WEBVTT
00:00:19.920 --> 00:00:21.600
Welcome to the future of you.
00:00:21.920 --> 00:00:24.719
In this episode, we're diving into the world of AI.
00:00:24.879 --> 00:00:31.920
We chat with a return guest to the pod, Lord Holmes, who's been busy pushing for new laws around artificial intelligence.
00:00:32.159 --> 00:00:38.719
He thinks it's time to get ahead of the game with AI and make sure that we have the rules that both encourage innovation and protect people.
00:00:38.880 --> 00:00:47.359
So he talks to me about the private members' bill he bought that includes the notion of an AI authority to oversee the changing landscape in the UK.
00:00:47.600 --> 00:01:04.319
Now the bill covers everything from consumer protection to intellectual property, but despite it moving through all of its stages in the Lords successfully, and as discussed before the general election, just as it was about to go to the Commons, it fell through the gaps in the wash-up of that general election.
00:01:04.560 --> 00:01:19.599
Several Lords had something to say about AI in the debate which followed the King's speech, but as we wait for greater clarity on the new government's position, it's probably best to look at the written update from Chris on his blog, and I've popped in a link in the show notes.
00:01:19.920 --> 00:01:24.079
It's also my great pleasure to speak with futurist Richard Yonk.
00:01:24.239 --> 00:01:35.599
He's also a former computing AI contributing editor for The Futurist magazine and a writer for Scientific American, World Future Review, Fast Company, GeekWire, Salon, and many more.
00:01:35.760 --> 00:01:47.599
He's been interviewed and quoted in the New York Times, Wall Street Journal, Wired, BBC News, etc., and guested on numerous podcasts and radio programmes, though the future review must surely be his favourite.
00:01:47.840 --> 00:01:58.560
He says it is his mission to provide organisations and individuals with insights that spark imagination, reduce risk exposure, and effectively connect the dots towards a preferred future.
00:01:58.719 --> 00:02:17.919
Now today he helps us connect some of the dots on AI, clearing up some of the myths and fears about it, reminding us it's still not anywhere close to human intelligence, and talking about the ups and downs of its development, the realistic timeline for things like self-driving cars, and the importance of balancing AI's benefits with ethics.
00:02:18.080 --> 00:02:21.599
So tune in as we break down some of these complex topics.
00:02:21.759 --> 00:02:23.280
Over to you, Richard.
00:02:42.960 --> 00:02:48.800
We have, in these last few years, reached a point where suddenly there's been some very big changes.
00:02:48.960 --> 00:02:54.400
People are of all sorts are very aware of it in a way that they never have been before.
00:02:54.639 --> 00:03:14.800
There's a lot of fear, there's a lot of concern built out of the narratives that we find from science fiction and movies and so forth, uh, that are perpetuated to a degree by uh some of the companies uh where we're worrying about AI taking over, AI replacing us entirely.
00:03:14.879 --> 00:03:34.879
Uh, but the fact is that when you talk to the people on the ground, the researchers, the people in the labs, the people who are actually doing the work, we are so far away from the point where AI is anything close to being uh human intelligence level.
00:03:35.039 --> 00:03:48.479
Uh we have a lot of things with generative AI, LLMs, uh Chat GPT, where we can point it at a particular task.
00:03:48.639 --> 00:04:10.319
It's it's taken in this enormous corpus of information from the internet, pretty much uh everything that we've been pouring onto the internet and the web for the last 30 years, and through statistical analysis and different types of algorithms, be able to take that information and answer amazing things.
00:04:10.479 --> 00:04:17.839
And it seems very creative, it seems very human-like, but in fact, what it is is it's us.
00:04:18.000 --> 00:04:21.279
We're we're seeing ourselves reflected back on us.
00:04:21.519 --> 00:04:47.360
And even though these programs can do amazing things, they can pass college exams, they can pass the bar, they can accomplish many different, very static tasks, they still don't have a lot of the very, very basic reasoning skills and understanding of causality and other aspects of our world that even a four-year-old has.
00:04:47.759 --> 00:05:05.839
So we're still at a very, very early phase with AI, but we're also at a phase where we we have to understand that it's a very different kind of intelligence, and we end up often misunderstanding what level it's at and what how we should be interacting with it.
00:05:06.160 --> 00:05:07.519
Yeah, so so that's very interesting.
00:05:07.600 --> 00:05:09.199
So what's driving the hype?
00:05:09.360 --> 00:05:10.319
Is it PR?
00:05:10.399 --> 00:05:11.199
Is it marketing?
00:05:11.360 --> 00:05:16.480
Is it all about research grants, or is it just, you know, ego of founders?
00:05:16.560 --> 00:05:20.319
What is the main driving force for the for the hype cycle?
00:05:20.800 --> 00:05:22.480
I think it's a lot of different things.
00:05:22.639 --> 00:05:25.040
We've been in a range of hype cycles.
00:05:25.120 --> 00:05:48.319
Uh I feel like they've been accelerating during the past decade or so uh as different uh business models have been kind of accelerating, and there's this feeling and need that we have to, that companies have to uh turn around and be showing profit very quickly, getting to market as quickly as possible, and essentially racing things out before they're ready.
00:05:48.560 --> 00:05:57.279
And so we get metaverses that are not ready for prime time, we get NFT markets that crash, and so forth.
00:05:57.519 --> 00:06:02.959
So, with AI, AI, as I say, has been around for 75, 80 years.
00:06:03.199 --> 00:06:29.680
It has gone through a range of ups and downs, AI winters, pullbacks in financing, and what these end up resulting in is actually a little bit of uh, you know, getting rid of the the chaff, having uh a little clearing, and there's an opportunity to kind of restart and and be funding uh viable uh you know directions and approaches again.
00:06:30.000 --> 00:06:36.480
The thing is that right now we're in a phase where there's a lot of money being thrown at it.
00:06:36.800 --> 00:06:44.959
You have the certainly the marketing departments, the founders wanting to talk things up as as far and as fast as they can.
00:06:45.279 --> 00:07:16.879
But a lot of the dialogue and concern, the ways that they're uh hyping this uh draw a lot on uh a range of dialogues, a range of conversations that have been on the certain corners of the web for and outside the web for several decades now, uh, where we're talking about and worrying about uh what happens when you know AI reaches the singularity, when it uh we reach uh an intelligence explosion, and who's going to protect us?
00:07:17.040 --> 00:07:21.360
Well, these companies have to, you know, they're the only ones who can take care of us.
00:07:21.519 --> 00:07:24.000
It at a certain level, I think they believe it.
00:07:24.160 --> 00:07:25.759
I think that there is ego there.
00:07:26.000 --> 00:07:38.399
Everyone who's working on something major in terms of technology throughout the ages thinks that what they're doing is the thing that's going to save humanity, that is the the great thing that nobody else can do without it.
00:07:38.480 --> 00:07:41.600
And they're the only ones who can bring it to light.
00:07:41.759 --> 00:07:51.600
The fact is that invention happens when it's ready, when the the time is right, when the the various infrastructure and uh conditions are set.
00:07:51.839 --> 00:07:54.000
So if not them, it will be somebody else.
00:07:54.319 --> 00:08:02.160
What do you think in terms of the um the regionalization of AI and the way we're treating it, particularly from a governance point of view, but not just that.
00:08:02.319 --> 00:08:06.639
I mean, even just thinking about data security, all those sorts of things.
00:08:06.720 --> 00:08:12.319
I mean, obviously what China are doing is a completely different approach to the EU, which again is a completely different approach to the states.
00:08:12.560 --> 00:08:15.279
You're in the states, I should say that for for for our listeners.
00:08:15.680 --> 00:08:18.959
What was your point of view on who's doing it best or better?
00:08:19.120 --> 00:08:21.199
Or can anybody do it that well right now?
00:08:21.600 --> 00:08:29.519
I think we're all finding our way, uh, most certainly, and everybody's got a different internal narrative that they're working from.
00:08:29.600 --> 00:08:38.240
They we've all got different political conditions and situations that we're starting from, so that alters the the path and the course.
00:08:38.480 --> 00:08:50.879
The EU, I feel, has a uh a really good start in terms of protecting the users' interests, the the citizens' interests.
00:08:51.120 --> 00:08:57.840
The GDPR is is an example pre-AI that certainly is involved there.
00:08:58.000 --> 00:09:03.919
I think that the idea of having that as a priority is important.
00:09:04.159 --> 00:09:23.279
I think that the idea where we have a free-for-all in the US where essentially we you know allow and have allowed, even with a small amount of noise and pushback, um, the scraping of all people's information and data that you know that you've got artists, you've got writers.
00:09:23.519 --> 00:09:28.480
I mean, I'm pretty sure your books are are you know part of Chat GPT.
00:09:28.559 --> 00:09:35.039
I know at least one of mine is, but it is uh a reality of our world right now.
00:09:35.279 --> 00:09:51.360
And the fact is that you've got these people who have said, okay, I'm going to basically create enormous profits for a very small number of people based on the product of the near sum of humanity.
00:09:51.519 --> 00:09:53.519
That's there's something wrong there.
00:09:53.840 --> 00:10:11.039
Because one, we're not having a say, but two, we need to have a range of different types of protections, not least of which is a level of balance and recognition toward, you know, if not compensation, at least control over our product and our data.
00:10:11.519 --> 00:10:13.279
What do you think about AI agents?
00:10:13.360 --> 00:10:26.480
Because um obviously my interest is in the future of identity and particularly identity, how it is influenced by technology, but very particularly by AI as we go forward, particularly with this podcast series.
00:10:26.799 --> 00:10:32.000
A lot of probably the work you do is about AI at an organizational level, and and some of mine is.
00:10:32.080 --> 00:10:41.039
But what do you think the impact is of agents or agentification on individuals, on persons, on personal identity?
00:10:41.840 --> 00:11:18.960
If you look at it from a very idealized standpoint, it has such enormous potential for helping us as individuals to whether it's possibly simplify our lives, possibly extend them in ways that simply has never been humanly possible before, uh, where our reality, our identity becomes something that extends well beyond us, be it is able to extend in ways where it diversifies in a way that you know we simply have never kind of inhabited throughout history.
00:11:19.200 --> 00:11:22.080
But that's a very idealized approach to it.
00:11:22.240 --> 00:11:33.759
At the more call it cynical viewpoint, these are typically going to be uh algorithms that are built by and controlled by corporations.
00:11:34.240 --> 00:11:42.639
So how do we silo our personal information, personal data, privacy in such a way that we are protected?
00:11:42.799 --> 00:11:49.120
And of course, that's even in an again idealized way, uh, that's much more easily said than done.
00:11:49.279 --> 00:11:58.879
Every everybody these days, as we've seen in in recent years, is is hacked on a regular basis and uh our our data and information exposed.
00:11:59.200 --> 00:12:22.720
So trying to look at the two poles of this and figure out, okay, where do we consider it as a reality in in somewhere hopefully in the middle where we're as protected as we can be, we get as much benefit as we can, but we're also not blind to the idea or the fact that we're probably being exposed and made vulnerable in ways that we never have been before.
00:12:23.440 --> 00:12:45.519
Because I've looked at a lot of models where one is using AI to create a twin, a digital twin, an avatar, more than one avatar's agents that go out and do things or say things on our behalf, but obviously reflect and represent or represent the self, you know, the tone of voice, the content, the the preferences, all those sorts of things.
00:12:45.600 --> 00:12:46.960
It's not there yet, obviously.
00:12:47.120 --> 00:12:48.480
There's some early examples.
00:12:48.799 --> 00:12:58.799
I think it's going to be very interesting when people are met with representations or representations of people and they don't really know whether it's the person or not.
00:12:58.879 --> 00:13:06.799
I think the assumption probably is going to be that you you know you're a representation, you're an avatar, rather than the assumption being you're a real person.
00:13:06.960 --> 00:13:15.279
And I think you've talked about this as well, that that sort of flip where we assume that we're dealing with a non-human rather than dealing with a human first.
00:13:15.759 --> 00:13:17.600
I think that's a fair assumption.
00:13:17.759 --> 00:13:19.840
It's going to depend on the modality.
00:13:20.159 --> 00:13:26.879
Right now, if we're back and forth with someone in text, where there's a whole range of things we don't get feedback on.
00:13:26.960 --> 00:13:39.039
And so it's pretty easy for us to say, okay, we're at a stage in technology where, yeah, there's a pretty good chance, you know, this is where I'm not dealing with the actual human being at this point.
00:13:39.279 --> 00:14:03.279
Uh, when we start talking about a range of you know, visual feedback, you know, people being represented as avatars, even entirely photorealistic, perfect representations of their voices and so forth, as we all become more familiar with the fact that yes, the technology is at this level right now, then we start to make a range of assumptions that we wouldn't have in the past.
00:14:03.519 --> 00:14:29.279
I think one of the things it brings to mind that I explored in my last book, Future Minds, was the idea that when you have all of these representations, all of these avatars, all of these identities going out into the world, if you go far enough into the future, and I'm not even sure how feasible this will ultimately be, you perhaps reintegrate that into your own memories, thoughts, and so forth.
00:14:29.440 --> 00:14:31.679
I'm not sure that if that's going to be feasible.
00:14:31.919 --> 00:14:36.879
But for the meantime, in the meantime, right now, what does that look like?
00:14:37.039 --> 00:14:53.519
If you have a hundred or a dozen entities that are you or represent you going out doing things, a lot of that is only useful to you if you're able to consolidate and take some of that information that they've gathered or interacted with back.
00:14:53.840 --> 00:14:58.080
Now we're already dealing with a world of massive data overload.
00:14:58.320 --> 00:15:02.320
Uh, how do we as human beings contend with that?
00:15:02.480 --> 00:15:11.840
I think that's a really big question we have to ask ourselves because that's going to be the the intermediary period before we ever get to that far future that I was talking about.
00:15:12.159 --> 00:15:13.759
I was going to ask you about consciousness.
00:15:14.080 --> 00:15:16.960
You mentioned it earlier on about whether AI can be conscious.
00:15:17.120 --> 00:15:23.600
What's your view on the sort of again another hype cycle or hype idea concept, the singularity?
00:15:24.159 --> 00:15:27.120
So um how long do you have?
00:15:27.840 --> 00:15:29.519
Till 2045.
00:15:30.879 --> 00:15:31.759
Well, yes.
00:15:31.919 --> 00:15:46.960
Um, so when I wrote Future Minds, uh, one of the main things that was driving that was this desire to really explore the nature of intelligence, not just human intelligence, but what is intelligence.
00:15:47.279 --> 00:15:52.960
And uh that in part because it's such a big thing.
00:15:53.039 --> 00:16:03.679
It it it really, if you look at every form of life on this planet, it has a form of intelligence that is appropriate to the niche that it occupies.
00:16:04.240 --> 00:16:13.360
Similarly, if we look out at uh artificial intelligence, machine intelligence, is that going to be a continuation of this path?
00:16:13.519 --> 00:16:21.039
Or is it fundamentally different for a range of reasons, not least of which is the fact that it doesn't begin from a biological substrate?
00:16:21.600 --> 00:16:32.399
So I think that uh the recognition that no entity, no species, uh shares the same kind of consciousness.
00:16:32.720 --> 00:16:40.639
All of it all species have a different level of intelligence, and what constitutes the consciousness they have is rather different.
00:16:40.960 --> 00:16:53.440
It goes back to the idea of you take a certain animal, uh a bat, for instance, we have no concept what it is like to be able to navigate with accolation.
00:16:53.600 --> 00:16:58.320
And so this is a a different way of being.
00:16:58.559 --> 00:17:01.360
Each entity has a very different form of it.
00:17:01.519 --> 00:17:02.720
What will AI have?
00:17:02.879 --> 00:17:04.799
What will machine intelligence have?
00:17:04.960 --> 00:17:15.119
I suspect there will be many, many different kinds of intelligence, many kinds of something of a form of consciousness, but it won't be like human consciousness.
00:17:15.279 --> 00:17:16.799
So let's back that up.
00:17:17.039 --> 00:17:21.759
Consciousness is what Marvin Minsky called an a suitcase word.
00:17:21.920 --> 00:17:29.279
It's essentially a word that means so many things to so many people, or depending on the context that you're using it in at that moment.
00:17:29.680 --> 00:17:38.160
So if we talk about it right now, most people are thinking of consciousness in terms of, well, is it self-aware like I am?
00:17:38.319 --> 00:17:40.480
Is it able to self-reflect?
00:17:40.559 --> 00:17:53.440
Uh, does it have uh a form of phenomenal uh consciousness, as Ned Bloch has talked about, where essentially it can, you know, register and appreciate a particular shade of red.
00:17:53.759 --> 00:17:54.720
But what is that?
00:17:54.880 --> 00:18:05.839
You know, it's it's a phenomenal experience and not something that can really be explained in a objective way, other than to talk about its wavelength.
00:18:06.160 --> 00:18:10.720
So, in that sense, will AI have that?
00:18:10.880 --> 00:18:14.160
And I suspect that it very possibly will not.
00:18:14.480 --> 00:18:25.759
I think we may have a level or a type of subjective consciousness, uh, because it's an emergent property of our minds, that is unique in the universe.
00:18:25.920 --> 00:18:40.720
Uh and until and unless we ever run across some other entity that has something we can at least partially agree is similar, then I suspect we may be very unique in that for a long, long time.
00:18:41.200 --> 00:18:44.960
Do you think we're going to be able to preserve our consciousness or bring it back?
00:18:45.039 --> 00:18:49.440
I'm thinking about cryonics and other technologies that are very emergent.
00:18:50.160 --> 00:18:55.839
So again, um so cryonics is a is kind of a subset of that.
00:18:56.000 --> 00:19:07.839
I think we have to you know back off a little bit and talk about okay, if we are such a very different substrate, if our minds exist in such a way, you have uh you know ideas.
00:19:08.000 --> 00:19:28.240
Ray Kurzweil has talked about the idea, oh, if we can emulate uh, and it's hardly he's hardly the only person, uh, if we can emulate each neuron perfectly, you know, through electronic means, right you know, transistors, and so forth, uh, then we can essentially replicate a human mind.
00:19:28.480 --> 00:19:52.319
Um it sounds good from a totally reductive standpoint, but the fact is that uh the nature of emergence, the nature of uh how some of these phenomena arise out of individual components that result through synergistic means to create a whole that is greater than the sum of its parts.
00:19:52.559 --> 00:19:55.279
Um I'm not sure that's going to be feasible.
00:19:55.440 --> 00:19:59.279
If you extend that then to okay, can we upload our minds?
00:19:59.440 --> 00:20:12.480
Can we record ourselves in a digital form in a computer and be able to continue to uh live and exist in that way, or you know, take that information and somehow put it back in a biological system?
00:20:12.799 --> 00:20:17.440
I think that that's going to be a really hard lift.
00:20:17.599 --> 00:20:24.400
I'm not sure that we have that capability in our future, but I'd love to be surprised.
00:20:24.799 --> 00:20:30.319
It's interesting, it is reductivist, and it's also a very rational way of thinking about human beings, I think.
00:20:30.400 --> 00:20:39.759
And I think we forget a lot of the time just how emotional we are, or so many of our responses are all, you know, they're all emotional or our decisions are generated from emotions.
00:20:39.839 --> 00:20:48.160
I mean, obviously, there's been lots and lots of work on this in the past couple of decades, but you've written a lot, I know, about AI and machines and emotions.
00:20:48.240 --> 00:20:50.799
And can they understand or detect our emotions?
00:20:50.960 --> 00:20:51.759
What could we do with that?
00:20:51.920 --> 00:20:53.599
What kind of effect will that have on us?
00:20:53.839 --> 00:20:56.319
Will we even be aware of the nudges?
00:20:56.480 --> 00:21:02.880
Can you talk us a bit through your sort of philosophy around that and your your beliefs about what will be possible and what won't?
00:21:03.359 --> 00:21:03.759
Sure.
00:21:04.000 --> 00:21:13.039
Well, thanks for the uh the the lead in uh yes, I I wrote a book, Heart of the Machine, Our Future in a World of Artificial Emotional Intelligence.
00:21:13.359 --> 00:21:20.319
And essentially it was looking at systems that are able to read, interpret, and interact with human emotions.
00:21:20.559 --> 00:21:57.200
This was work that started in really in the 90s, uh, really got going at MIT Media Lab in the mid-90s, started to be commercialized toward the end of the 2000s, and there was a lot of significant interest in it, and still is, but a range of concerns around the kinds of privacy uh that you know are it involves when you're interacting with and and being able to take in information about people visually, whether in a public environment or in a private one, you know, leads to a range of different types of problems.
00:21:57.440 --> 00:22:04.319
But then there's also the reality of Of what are these systems actually reading and recording?
00:22:04.640 --> 00:22:16.240
So human emotion is a core aspect of the human condition, something that essentially precedes our other forms of communication, such as speech and so forth.
00:22:16.559 --> 00:22:26.240
So, how fundamental that is, is I think something that is overlooked by a lot of people, especially in technology.
00:22:26.480 --> 00:22:32.799
But when you talk about taking a system that reads a person's face, for instance, a person smiles.
00:22:32.960 --> 00:22:34.400
Okay, it's they're smiling.
00:22:34.640 --> 00:22:37.519
But there is a huge number of nuances there.
00:22:37.599 --> 00:22:51.279
And beyond that, for any of us that as human beings have interacted with other people, we are continually in conversation, looking at their faces, taking the in expressions, looks in their eyes.
00:22:51.519 --> 00:22:57.200
I mean, so much information is being transmitted non-uhverbally.
00:22:57.680 --> 00:23:00.240
And there's been studies around that as well.
00:23:00.480 --> 00:23:07.440
And that information is continually feeding that conversation in a feedback loop between the two of you.
00:23:07.680 --> 00:23:10.799
It's referred to in psychology as theory of mind.
00:23:10.960 --> 00:23:20.480
Essentially, we are witnessing and experiencing this attempt to read each other, read each other's minds, you know, at a certain level.
00:23:20.640 --> 00:23:26.160
We're not reading each other's minds, but we are trying to understand what is that person thinking?
00:23:26.240 --> 00:23:28.160
How are they mad at me?
00:23:28.400 --> 00:23:30.640
Are they, you know, laughing at this?
00:23:30.799 --> 00:23:31.519
What have you?
00:23:31.759 --> 00:23:39.519
So when a machine takes that information, that visual information, and labels it, it's really not doing very much with it.
00:23:39.680 --> 00:23:43.119
It's just, you know, you could take an algorithm and say, well, they're smiling.
00:23:43.279 --> 00:23:44.480
We're going to say they're happy.
00:23:44.640 --> 00:23:46.000
Well, maybe they're not.
00:23:46.319 --> 00:23:51.440
You uh, I think mentioned in your book uh Lisa Feldman-Barrett, I mentioned her as well.
00:23:51.599 --> 00:23:57.599
And she definitely pushed back a few years ago on the whole emotion AI concept.
00:23:57.759 --> 00:23:58.960
And she's totally right.
00:23:59.200 --> 00:24:04.000
The fact is that these machines are not reading this information.
00:24:04.160 --> 00:24:15.200
And just as AI doesn't have the tools yet to reach full autonomous driving, for instance, it's still missing that last little piece.
00:24:15.440 --> 00:24:31.279
Right now, it doesn't have the reasoning capability that it may have in a few decades' time to be able to interact with us and have enough understanding of, you know, the nature of human beings, the nature of the world, the way we interact, and so forth.
00:24:31.440 --> 00:24:44.640
It may in the future be able to do a better job of essentially emulating or simulating that theory of mind kind of phenomena, but it's not going to be something that, again, it's not going to be as perfect as us for a long time.
00:24:44.880 --> 00:24:46.319
And we're not perfect.
00:24:47.119 --> 00:24:56.640
It's really interesting that you also talk there about driving because that was one of the big cells, wasn't it, a few years ago on emotive AI, emotional AI.
00:24:56.720 --> 00:25:06.319
You know, it's going to be in our cars, it's going to be reading our expressions, it's going to be know what mood we're in, and therefore it's going to create the atmosphere with the music, all these other things.
00:25:06.400 --> 00:25:20.720
I mean, we're so, so far off that we can't even get the car to drive properly without a human driver, let alone sort of program it all with all this, you know, emotional intelligence that is supposedly captured by a machine.
00:25:20.960 --> 00:25:26.000
So here's a question for a futurist, because it's one of the most difficult things, ones I think.
00:25:26.079 --> 00:25:26.240
Yeah.
00:25:26.480 --> 00:25:28.160
What are the timescales?
00:25:28.960 --> 00:25:31.599
What are the timescales for some of these technologies?
00:25:31.759 --> 00:25:35.680
What would be some of the milestones, particularly on AI, do you, do you think?
00:25:36.079 --> 00:25:36.880
Sure, sure.
00:25:37.200 --> 00:25:44.319
Try not to predict, but I do have a a range of you know general ideas that as far as where things are.
00:25:44.480 --> 00:25:58.000
Uh, I I typically tell people when they, you know, we we were told back in the 2000s, 2008, 2009, uh, by a range of executives, we were going to have the streets filled with self-driving cars by 2017.
00:25:58.240 --> 00:26:07.440
Every year in the last half dozen years, Elon Musk and others tell us we're one year away from full level five autonomous vehicles.
00:26:07.680 --> 00:26:09.200
We are not still not close.
00:26:09.440 --> 00:26:16.480
There's a range of problems, and these are these are that last mile problem that you have with a range of different technologies.
00:26:16.720 --> 00:26:29.599
So, in my mind, I've been saying to people, and we're closing in, and I'm getting I'm thinking about pushing it out a little bit, I've been telling people that we're not going to have it until at least 2030 in terms of technical capability.
00:26:29.759 --> 00:26:45.119
Then we have to get through a whole range of other factors because you have public sentiment, you have regulation, you have a whole transition period between the people giving up their cars, driving on those the roads at the same time.
00:26:45.279 --> 00:26:53.279
Uh, we're talking about well out into the 2030s before self-driving cars have fully taken over, and that may be optimistic.
00:26:53.519 --> 00:26:55.680
Other kinds of AI, it depends.
00:26:55.839 --> 00:27:13.119
I'm talking about you know, to people about the range of different types of programs and projects that are being done in research labs around uh different types of abstract reasoning emulation, you know, trying to understand, have key machine common sense and so forth.
00:27:13.279 --> 00:27:20.799
These are being developed, they're being worked on, but I think that it's a good 10, 20 years to see these fully be realized.
00:27:20.960 --> 00:27:33.519
In the meantime, the the thing that I really feel is going on now and will be driving a lot of AI for the next couple of decades is what I call, well, I'm not the only one, hybrid intelligence.
00:27:33.759 --> 00:28:01.359
The idea that humans and AIs working together in more and more frictionless ways, you know, between the intelligence of the machine, between the interfaces that are used, the different types of programs and algorithms to essentially help us to work together in a way where the systems do what they do so well, whether it's something highly repetitive, certain kinds of pattern recognition, and so forth.
00:28:01.519 --> 00:28:15.359
And we bring in a range of human-like skills that, you know, whether that's around reason, other aspects of emotion, values, and so forth, so that as a team, we're continually supporting each other.
00:28:15.519 --> 00:28:23.599
And you know, there's different small projects around this, but I think there's a whole range of ways this is going to expand over the coming decades.
00:28:24.000 --> 00:28:35.759
I'm really interested in this actually, because I think it's overlooked and underestimated just how difficult it is for human beings to work with AI, not for AI, but maybe in some cases it is.
00:28:35.920 --> 00:28:37.839
Um, nobody's taught us this.
00:28:38.079 --> 00:28:39.839
Who do we go to for it?
00:28:40.000 --> 00:28:43.839
And even for the people who are going to teach us, where's the pedagogy for it?
00:28:43.920 --> 00:28:47.759
I mean, there's if you track back, it's just so difficult to contemplate.
00:28:47.920 --> 00:28:56.160
I mean, I know how to go into a meeting with other people and discuss things in a conversation or a brainstorm or a you know, whatever the format is of the meeting.
00:28:56.240 --> 00:29:03.359
I know how to work on my own in front of my computer, not talking to it, just typing or whatever I'm doing, researching.
00:29:03.599 --> 00:29:09.599
But how do we work in these sort of whatever you want to call it, hybrid intelligence, hybrid working, co-piloting?
00:29:09.839 --> 00:29:11.759
I think it's a fascinating area.
00:29:12.160 --> 00:29:13.680
Yeah, no, it absolutely.
00:29:13.839 --> 00:29:15.759
And and it's rife with problems.
00:29:15.920 --> 00:29:17.920
I think there's going to be a lot of growing pains.
00:29:18.160 --> 00:29:32.240
And as human beings, again, kind of relearning or continually learning how to operate and how we use our values in the situation, I think is going to be difficult to, it's going to take us time.
00:29:32.480 --> 00:29:47.519
As an example, what's popping into my head is the other day I was on a uh platform with a number of other futurists, actually, uh being asked a range of questions for a conversation that was all being led by the platform's AI.
00:29:47.759 --> 00:29:55.839
So the platform was continually feeding us back information, prompting us for different things, and so forth.
00:29:56.000 --> 00:30:02.000
And it was definitely different, engaging, but it was also incredibly fast.
00:30:02.240 --> 00:30:04.880
And so, I mean, it has benefits.
00:30:05.039 --> 00:30:25.519
But for those of us who like to take time to do deep thinking, it really was not, I mean, it probably wasn't what it was geared for, but it really uh says something about uh we have to prioritize what kind of thinking and what kind of processes we're working on at a given time.
00:30:25.839 --> 00:30:29.680
Yeah, because immediately when you tell me that, I think, oh my God, that sounds awful.
00:30:29.839 --> 00:30:35.599
It feels like unless you're an ENTJ, Myers Briggs, then you aren't going to survive that meeting.
00:30:35.759 --> 00:30:38.240
And like we don't all want to be an ENTJ.
00:30:38.480 --> 00:30:44.880
No offense to ENTJs, but you know, um, you can have too much of a good thing.
00:30:45.599 --> 00:30:59.839
Um, yeah, so I mean, I feel like it's filtering potent potentially, obviously, filtering out some of the good stuff, some of the best stuff that humans do in order to be efficient, to operationalize, to optimize, and just to get an output and a result.
00:31:00.400 --> 00:31:00.960
Absolutely.
00:31:01.039 --> 00:31:07.359
It's something that I've I've been talking a little bit about about the nature of what we're doing with generative AI right now.
00:31:07.519 --> 00:31:20.160
Uh, the fact is that we have a lot of people who are touting and pushing uh this narrative that these different forms of generative AI can make us so much more efficient and we can do this and that.
00:31:20.400 --> 00:31:34.480
And when you look at some of the studies, there was a study six months ago, eight months ago, uh, where a range of consultants were using a form of generative AI, I want to say it was ChatGPT, but I'm not definite.
00:31:34.720 --> 00:31:40.160
And they showed that they had a X percentage of increased efficiency.
00:31:40.400 --> 00:32:08.319
But when you actually took it a little more granular uh and saw that the the consultants with the least amount of skill had theirs raised the most, and the better uh performing consultants uh really did not have that level of benefit, it makes you wonder what is the long-term effect uh of that kind of AI influence on human thinking, on the ways that we perform.
00:32:08.799 --> 00:32:13.839
We operate right now in a form of general competition with other human beings.
00:32:14.000 --> 00:32:37.920
If you have this intermediary that is operating in such a way that it kind of diminishes the value of exceeding expectations, but actually tends to make us all gravitate toward the mean, which is what an LLM does anyway, that that really is um concerning for what that could mean for the future of human thinking, human knowledge, and so forth.
00:32:38.319 --> 00:32:52.640
Yeah, I mean, do you think that whether it's a meeting or whether it's a piece of creative output or whatever the content is, do you think we'll have several different outputs that then may be aggregated or one can choose between?
00:32:52.880 --> 00:33:06.480
Because as you just said, you know, we can't really rely on just the analysis of AI and what whatever mode or format that is, because obviously it is potentially discarding lots of stuff that we would as humans find valuable.
00:33:06.640 --> 00:33:08.960
So how do we create a system like that?
00:33:09.039 --> 00:33:20.319
I mean, as you were talking about that particular meeting with the air, I was thinking, I wonder if you could have 12 different AIs going and they're all programmed in slightly different ways to analyze the output slightly differently.
00:33:20.480 --> 00:33:26.160
But even then, it seems like a massive waste of um energy when you could just rely on the humans to do it.
00:33:26.480 --> 00:33:42.000
Well, and that's the thing is I think that when you're talking about wanting to keep human values in the loop, that really, if anything, I I want in that situation for there to be more human beings rather than saying, oh, we need more AIs.
00:33:42.240 --> 00:33:48.799
I want to see, you know, a range of different kinds of thinking, a range of different kinds of views.
00:33:48.960 --> 00:33:50.880
Sure, that's that's diversity in general.
00:33:51.039 --> 00:33:59.519
That the there's a lot of benefits, especially in futures thinking, to have a range of viewpoints so that we can make sure that we don't have blind spots.
00:33:59.839 --> 00:34:16.000
But I really would not be willing to I I think if you put it all on, say, 12 AIs, I've got a feeling you're going to have a huge number of blind spots, particularly in terms of what is valuable or important to human beings.
00:34:17.760 --> 00:34:35.039
When I've been looking at some of the particularly the tech entrepreneurs, but a lot of the smaller AI companies and all this, I'm getting concerned that there's um an assumption, even a desirability around the AI will program us with our values and rather than the other way around.
00:34:35.280 --> 00:34:40.880
And obviously there's a kind of there's a togling between the two, and it's a bit of a compromise and negotiation.
00:34:41.039 --> 00:34:45.840
But even then, the human being has to have the final say over those values.
00:34:46.320 --> 00:34:48.719
But can the human being always know?
00:34:48.960 --> 00:34:55.119
I mean, we've seen what happened with Google Gemini and lots of lots of things recently that have, yeah, exactly.
00:34:55.280 --> 00:34:57.119
You're rolling your eyes, I on too.
00:34:57.199 --> 00:35:01.360
I mean, I mean, sometimes I look at that and think, how how could they have gone out with that?
00:35:01.519 --> 00:35:06.239
Was that just a test to see how deplorable it would be and what people's reactions would be?
00:35:06.320 --> 00:35:08.320
Well, that was that actually a mistake.
00:35:08.639 --> 00:35:08.880
Sure.
00:35:09.039 --> 00:35:11.360
I'm just going to recap for some of your listeners.
00:35:11.599 --> 00:35:26.239
So back in February, Google released a version of uh Gemini that had it changed the algorithm slightly to try to implement more diversity into the uh results for different types of generation, particularly image generation.
00:35:26.480 --> 00:35:35.760
And they were creating these very, very and consistently, not just one-offs or just because you asked for it.
00:35:35.920 --> 00:35:44.400
They were creating uh representations of Vikings and World War, well, Nazis, I guess.
00:35:44.559 --> 00:35:57.199
And others you you had uh you had who were basically um people of various ethnicities that did not participate for very good reason in those events.
00:35:57.440 --> 00:36:00.320
There was a Native American in ancient Greece.
00:36:00.400 --> 00:36:12.000
Uh, you know, there was just all of these really ridiculous uh visuals that it's fine if you asked for it, but if it's just how it's going to generate it, it's like, hey, you guys messed up bad.
00:36:12.159 --> 00:36:21.599
Uh so yes, I unfortunately they've had a range of from a marketing PR standpoint, a range of bad demos here over the last number of years.
00:36:21.760 --> 00:36:28.079
I can actually think of a few different instances, and it's like, I'm not sure what's happening at that level or with that.
00:36:28.159 --> 00:36:30.960
But anyway, don't really want to get into that too much.
00:36:31.199 --> 00:36:41.440
I've said for a long time, and hardly the only person out there saying it, that technology has always changed us as it has developed.
00:36:41.599 --> 00:36:47.519
We create technology, it alters us in some way, regardless of what it is.
00:36:47.599 --> 00:36:53.039
Uh, and it's become more and more apparent with the modern technologies like social media and so forth.
00:36:53.280 --> 00:36:59.199
So that you know, now most of us, or a lot of us, uh, read short form only.
00:36:59.360 --> 00:37:02.639
Reading long form is much, much harder than it used to be.
00:37:02.719 --> 00:37:06.480
Our our the way that our brains process is actually quite different.
00:37:06.800 --> 00:37:24.639
We have uh some small, you know, little you know, bone growth in parts of our body, you know, because we have a different posture than we used to have, you know, years ago because of how we spend most of our time at desks and in front of machines these days or in front of uh mobile phones, so on and so forth.
00:37:24.800 --> 00:37:29.119
Technology continues either physically or mentally to alter us.
00:37:29.360 --> 00:37:33.280
The question then becomes is this something that we need to do?
00:37:33.519 --> 00:37:38.559
We behave and operate differently in terms of how we use a computer early on.
00:37:38.719 --> 00:37:40.639
We had to learn a command line interface.
00:37:40.800 --> 00:37:43.360
That was, you know, for a lot of people, that was tough.
00:37:43.519 --> 00:37:57.199
As it became an easier and easier interface, it became more natural, but we still do really weird things when you think about it in terms of how we ordinarily would operate in nature, in order to be able to make these machines work.
00:37:57.280 --> 00:38:01.039
We operate at a certain level on their terms rather than the other way around.
00:38:01.360 --> 00:38:03.599
We shape our tools and our tools shape us.
00:38:03.840 --> 00:38:24.320
With that in mind, um, my final question for you, I think, Richard, is given everything we've been talking about and the supposed benefits and advantages of AI, if it's used correctly, or we find ways to work with it in a hybrid manner, is the future going to be more complex or less complex with AI in it?
00:38:25.679 --> 00:38:29.519
So uh the future will be more complex.
00:38:29.760 --> 00:38:34.000
Decade after decade, century after century, we create abstraction upon abstraction.
00:38:34.079 --> 00:38:38.400
We we keep creating and adding complexity to our world and to our systems.
00:38:38.960 --> 00:38:49.280
We may feel at certain times or in certain ways that our lives are simpler, but overall, I think you we have to say that it will be more complex.
00:38:49.519 --> 00:38:50.400
Will it be better?
00:38:50.639 --> 00:38:51.840
Will it be worse?
00:38:52.000 --> 00:39:13.760
That's I think a far more important for me question in terms of is what we're creating going to be improving and beneficial to the human condition, or are we doing something that is ultimately going to be against our nature, be against what it means to be human beings?
00:39:14.000 --> 00:39:28.320
I don't want to live in a world where essentially all of us spend all of our time feeding the social media machines and all of the different systems in order that they can operate.
00:39:28.559 --> 00:39:40.719
I think we all have to still have the experiential life that we have in nature, in the world with other people, in order to be able to have full and rich lives.
00:39:43.519 --> 00:39:49.599
So welcome again to Lord Holmes, Chris, uh second appearance on the podcast.
00:39:49.760 --> 00:39:50.559
Thank you for returning.
00:39:51.039 --> 00:39:51.360
Pleasure.
00:39:51.760 --> 00:40:03.920
I wonder if you could tell us a bit about the artificial intelligence regulation bill that you have introduced and um bring us up to speed on where we are in its like is it its second reading, or is it it just received its second reading?
00:40:04.639 --> 00:40:17.599
Certainly, and the reason I wanted to introduce a piece of legislation on artificial intelligence was I think that we have an opportunity to lead and it's now time to legislate.
00:40:17.920 --> 00:40:21.440
The government's position is largely wait and see.
00:40:21.920 --> 00:40:34.880
But I think we need to know everything that we need to be in the right position for pro-innovation, pro-citizen rights, and pro-consumer protection, which is why I introduced my bill.
00:40:34.960 --> 00:40:40.559
And to give some context, every member of parliament has the opportunity to introduce a private member's bill.
00:40:40.639 --> 00:40:44.559
It can be on any subject they want, it needs to be relatively specific.
00:40:44.880 --> 00:40:58.320
And then there's an element of luck because it goes into a ballot at the beginning of every session, and you need to really come in the top 25 in the ballot if your bill is going to have a chance of getting before parliament.
00:40:58.559 --> 00:41:07.199
I came sixth, so I had first reading of my bill in November, and that means that's just when I present the bill to Parliament and read its long title.
00:41:07.360 --> 00:41:10.639
And we then had a second reading on the 22nd of March.
00:41:10.719 --> 00:41:18.960
So that's the opportunity for a full debate on all of the issues and all of the clauses within the bill.
00:41:19.280 --> 00:41:39.599
So I wanted to have a bill which addressed the innovation question, the investment question, crucially consumer protection, citizen rights, and to really drive a far greater level of public engagement with the opportunities and indeed the challenges of artificial intelligence.
00:41:40.320 --> 00:41:47.039
But why have you felt the need to offer up this bill and uh we'll we'll get on to some of the details.
00:41:47.199 --> 00:41:48.239
What do you think is missing?
00:41:48.320 --> 00:41:48.800
What's the gap?
00:41:48.960 --> 00:41:58.000
Obviously, the EU are doing their AI Act and other regions or territories are looking at their own sort of national or regional regulatory solutions.
00:41:58.159 --> 00:42:01.039
What is it that you think is missing in the UK?
00:42:01.840 --> 00:42:14.000
We had the Bletchley Safety Summit, the AI Safety Summit in November last year, and that was very successful and right to look at frontier risks and that question of safety.
00:42:14.320 --> 00:42:35.280
But having stood up a successful summit, it seems logical to then look at all of the other elements of AI which are already impacting people's lives, oftentimes without those people even knowing it in terms of shortlisting for recruitment, for example, in higher education, in healthcare decisions, people are already being impacted by AI.
00:42:35.519 --> 00:43:03.440
It's time that we legislate so we can cover all of the areas where AI is currently being deployed, all the areas where it may be, so we get the best outcomes for citizens, for companies, for innovators, for IP owners, for our nation, and we have a real opportunity with three L's I call them law, learning, and language.
00:43:03.599 --> 00:43:18.159
We have the great good fortune of English common law, we have a sensational higher and further education ecosystem, and we have the good fortune of English as our language.
00:43:18.480 --> 00:43:31.280
With that, we have a responsibility, I believe, not just an opportunity, a responsibility to lead internationally focused to lead on artificial intelligence.
00:43:31.760 --> 00:43:42.000
Okay, so why doesn't the existing landscape of many regulators in the UK, and I've sat on the board of one of them, so um I am familiar with it.
00:43:42.159 --> 00:43:54.400
For example, why doesn't the ICO or why doesn't Ofcom or even in some respects the advertising standards authority, why doesn't the current regulatory landscape fulfill the asks of your bill?
00:43:54.960 --> 00:43:59.599
It's interesting because the government's approach is very much to look to existing regulators.
00:44:00.320 --> 00:44:09.280
To take on this work around AI in their specific sector in a very vertical manner.
00:44:09.840 --> 00:44:28.800
Now, both in terms of consistency of approach, coherence, and ensuring that we identify all the gaps where there aren't competent regulators for a particular area and where existing regulators don't have competency in that specific field.
00:44:29.039 --> 00:44:32.800
It seems to me it would make sense to have an AI authority.
00:44:33.199 --> 00:44:42.400
And in no sense am I suggesting that that should be a do-it-all AI, huge, bureaucratic, cumbersome, expensive regulator, not a bit of it.
00:44:42.719 --> 00:44:59.519
Agile, flexible, horizontally focused, to do that piece of work, to look across existing regulators, to assess their competency, to address the issues, the challenges, the opportunities of AI, to look across all relevant existing legislation to assess its competency, to address the challenges and the opportunities of AI.
00:44:59.679 --> 00:45:02.239
And by doing that, identify the gaps.
00:45:02.559 --> 00:45:23.199
So then it wouldn't matter if you're an innovator, a consumer, a citizen, in whatever sphere of the economy or life you've found yourself in, if AI was in the mix, you could expect to have a consistent and a coherent approach wherever you happen to be experiencing that AI.
00:45:24.559 --> 00:45:31.440
So it's like it's smoothing out the potential wrinkles across all of the existing bodies, is it, in different sectors?
00:45:31.840 --> 00:45:46.639
But very much taking a horizontal approach to thread this across all of the existing regulators and to identify where there are gaps and uh changes which need to be made to enable that consistency and coherence across the whole of the economy and society when it comes to AI.
00:45:47.039 --> 00:45:49.679
Because it's it's interesting the timing of this as well.
00:45:49.840 --> 00:45:54.800
I mean, it's sort of surprising to me actually that this is coming at this time.
00:45:55.039 --> 00:45:59.199
Why didn't it happen five years ago when we knew some of this was on the horizon, for example?
00:45:59.280 --> 00:46:01.760
So I'm always going to ask this as a futurist, you know.
00:46:01.920 --> 00:46:03.440
We saw a lot of this coming.
00:46:03.599 --> 00:46:04.800
I know you did as well.
00:46:04.880 --> 00:46:12.960
Um, but you've got the ICO, for example, and not to pick on them, but the ICO, so only just going out to consultation on generative AI, for example.
00:46:13.119 --> 00:46:15.760
I mean, it's done, it's happening, it's happened.
00:46:15.920 --> 00:46:22.159
Um, so what's your point of view on getting ahead of some of these things and why haven't we on AI?
00:46:22.239 --> 00:46:23.360
Or perhaps you think we have?
00:46:23.760 --> 00:46:40.719
No, you're you're you're quite right that in terms of our startups, our scale-ups, our industries, many of our universities and further education establishments, they're doing great work when it comes to AI.
00:46:41.119 --> 00:46:46.880
But from a legislative and a regulatory perspective, we're certainly somewhat behind the curve.
00:46:46.960 --> 00:46:49.280
And we didn't need to get behind the curve.
00:46:49.599 --> 00:46:54.320
We can now, I suppose, take the opportunity of not being first mover.
00:46:54.400 --> 00:47:03.119
The EU have very much got out there and uh done that with the EU AI Act, so we don't have to have that fear of the focus on first mover.
00:47:03.920 --> 00:47:06.400
But we do need to get on with it.
00:47:06.719 --> 00:47:27.039
And as you rightly identify, if we don't, the danger is that so much, particularly in the generative AI space, so much will have happened, it'll be very difficult to retrospectively regulate and to try and have that positive, forward-facing approach, which we still could have at this stage, I believe.
00:47:27.199 --> 00:47:35.360
But as every week goes on, obviously you're giving yourselves less of an opportunity to influence to lead.
00:47:36.239 --> 00:47:44.480
Yeah, so so tell me a little bit about the AI authority, who would lead it, and also your idea of an AI officer.
00:47:44.639 --> 00:47:46.000
How would that work?
00:47:46.480 --> 00:47:47.039
Certainly.
00:47:47.119 --> 00:48:02.400
So again, it's all about trying to give us the best opportunity to enable the opportunities and be cognizant of the challenges and control and where appropriate, eradicate the risks.
00:48:02.559 --> 00:48:05.840
So the AI authority would be independent of government.
00:48:06.079 --> 00:48:10.960
It would have key functions, as I set out in terms of looking at existing legislation.
00:48:11.199 --> 00:48:31.760
Crucially, like the whole of the AI regulation bill, the regulator, the AI authority would be underpinned by principles, key principles that should run through all of our thinking on this: trust, transparency, inclusion, innovation, interoperability, an international approach, accountability, accessibility, and assurance.
00:48:32.000 --> 00:48:35.519
Great principles which can underpin our approach to this.
00:48:35.760 --> 00:48:51.119
And having then principles-based, outcomes focused, inputs understood, and where necessary remunerated gives us the best opportunity, I believe, to get the best out of AI.
00:48:51.360 --> 00:48:58.239
Added to that, the AI responsible officer, another key principle running through to build that of proportionality.
00:48:58.400 --> 00:49:04.559
So people shouldn't think this has to be an individual who all their time is on AI.
00:49:04.719 --> 00:49:12.320
It's proportionate to the size of the organization, it's proportionate to their use of artificial intelligence.
00:49:12.480 --> 00:49:39.119
But to have a function within every organization that develops, deploys, uses AI, where that function is responsible for the ethical, unbiased use of AI, with reporting obligations analogous to those for financial reporting within the companies act, so well understood by all the commercial entities, seems to be another good focal point.
00:49:39.280 --> 00:50:12.320
Not that then that AI officer is the only person responsible, everything is just abdicated to her or him, but to be that function within an organization, which then enables as a company grows, or if it already is of scale, for the entire board, the whole of Exco and layers of the organization to understand their responsibilities and to understand the leadership and the culture that's required for inclusive by design, unbiased, ethical AI development and deployment.
00:50:13.039 --> 00:50:14.559
It's interesting to say about boards.
00:50:14.639 --> 00:50:20.559
You and I were both at the same AI conference yesterday and the day before, which I think we both enjoyed.
00:50:20.719 --> 00:50:25.679
Um that the question of boards came up actually in the panel discussion I was doing.
00:50:25.840 --> 00:50:31.039
And one of the attendees asked, What do you think the board of the future is going to be like given AI?
00:50:31.199 --> 00:50:40.719
Do you have any thoughts or or visions or or hopes or even fears actually about what the future of an AI or AI-infused or assisted board might be?
00:50:41.679 --> 00:50:56.480
There should be confidence in all individuals and all humans that AI always needs to be human-led, always needs to be human in the loop, and then potentially in some ways human over the loop.
00:50:56.800 --> 00:51:28.480
But what that should mean for boards, if we get this right, is greatly enabled, greatly empowered boards to be able to have information in a different format, right in front of people, real-time information, more than real-time with predictive analytics, being able to have such a different view of your organization, of your market, of your sector, of your competition, of your environment.
00:51:28.719 --> 00:51:42.159
All this is it if we get it right and we must, it should lead to far more empowered boards being enabled to have even greater leadership capabilities and leadership potential.
00:51:42.880 --> 00:51:49.920
Now, businesses, I'm guessing, are going to say, here comes more regulation or more burdens on us.
00:51:50.079 --> 00:51:53.039
We've already got a quite a lot to cope with.
00:51:53.199 --> 00:51:58.639
Is the government going to help us, you know, either with funding or any other kinds of aid?
00:51:58.719 --> 00:52:01.039
What what do you have any thoughts on that?
00:52:01.679 --> 00:52:02.320
Very much.
00:52:02.400 --> 00:52:16.079
I and I was really clear when I started that draft in this bill that it can't be a burden, either bureaucratically, financially, operationally, on business.
00:52:16.480 --> 00:52:24.480
That key principle of proportionality is one guard against burden.
00:52:25.360 --> 00:52:32.559
And indeed, all of us can cite bits of legislation and regulation that is burdensome.
00:52:32.639 --> 00:52:39.119
But that doesn't mean regulation and legislation of itself is bad, it means that's the wrong regulation, that's the wrong legislation.
00:52:39.360 --> 00:52:58.880
So really pushing these principles around right-sized regulation, proportionate, a sense that if you are a startup, potentially there can be de minimis levels as to as to how the features will impact.
00:52:59.840 --> 00:53:23.760
So all of the provisions should really be pro-innovation, pro-consumer protection, pro-citizen rights, and it should be the job of any legislature and any regulator to have all those principles in their mind when they're drafting, because it makes it more complex, but it's absolutely essential.
00:53:24.239 --> 00:53:29.360
Was there anything that came through the second reading or anywhere during the process so far?
00:53:29.599 --> 00:53:35.519
Any comments from either any of the other lords or ladies, or anything that's surprised you?
00:53:35.679 --> 00:53:45.599
Any kind of either you know, challenges or what what what's what's been interesting in the debate that obviously your bill has kind of encouraged?
00:53:46.320 --> 00:54:06.800
I did a great deal of consultation ahead of drafting, never mind presenting the bill with civil society, with the creatives with IP, with copyright, with people interested in our democracy, with businesses, to try and find the right path through this.
00:54:07.199 --> 00:54:15.599
What I was really pleased with was the level of cross-party support for the bill.
00:54:15.840 --> 00:54:19.440
All major parties and none were in support.
00:54:19.760 --> 00:54:32.239
There were the obvious criticisms from some quarters saying that we shouldn't be regulating at all, we should just let this technology emerge and only at that point do something.
00:54:32.400 --> 00:54:38.079
And that's that's a a reasonable position for some colleagues to take.
00:54:42.000 --> 00:54:51.360
Because nothing which is in my bill would be a curb on innovation, on development, on growth.
00:54:51.519 --> 00:54:52.559
Not at all.
00:54:53.039 --> 00:55:10.079
The key is if we get right size regulation in place, as all the lessons from history demonstrate, that is a benefit to innovation that crowds in inward investment.
00:55:10.320 --> 00:55:22.639
And two obvious examples: the fintech regulatory sandbox, a clear regulatory intervention, measure of success, well, replicated in well over 50 jurisdictions around the world.
00:55:22.960 --> 00:55:34.639
Open banking, a great UK creation, a creature of a regulatory intervention with the CMA9 order, now replicated in well over 60 jurisdictions around the world.
00:55:34.800 --> 00:55:58.400
So we know how to do this, we know how to do it to market make, market shape, enable innovation, because right-size regulation is a deliverant of the clarity and the certainty that investors, the developers, the startups that really all of us would want to see in any particular area of our lives.
00:55:59.679 --> 00:56:05.199
It's fascinating listening to you talking about in such uh positive, innovative ways.
00:56:05.280 --> 00:56:13.360
Um, and I wonder what your point of view is on the EU AI Act, and because it's comes from such a different point of view.
00:56:13.519 --> 00:56:18.079
Obviously, they're you know looking at the very specific threats and they've got these risk levels.
00:56:18.239 --> 00:56:25.360
Everything feels terrifying uh rather than you know, in any way sort of optimistic and and and innovative.
00:56:25.440 --> 00:56:36.320
And I mean, obviously, you've got interoperability as one of your principles, and of course that stands across you know businesses, uh regulators in in the UK.
00:56:36.400 --> 00:56:40.400
But what about the interoperability with the EU and their act?
00:56:40.480 --> 00:56:43.519
I'd love to get your point of view on that.
00:56:44.400 --> 00:56:50.800
Regulatory interoperability is a critical part of my bill as well.
00:56:51.280 --> 00:56:59.119
The EU AI Act is a valiant effort in all of its 892 pages.
00:57:00.000 --> 00:57:17.280
It's largely a creature of the legal framework in which it necessarily finds itself, so a code-based legal system, which necessarily tends towards a more prescriptive approach.
00:57:17.519 --> 00:57:29.760
I think you'll see the development of the act with all of the elements around uh generative AI coming in at some somewhat of a late stage of the drafting process.
00:57:30.719 --> 00:57:34.480
And I think it will be quite difficult in some ways.
00:57:34.559 --> 00:57:44.960
It's quite difficult to see in some circumstances where the right of a redress would come from, which authority you would be appealing to.
00:57:45.119 --> 00:57:47.440
There's some clashes there, certainly.
00:57:47.679 --> 00:58:09.840
And almost in many ways, don't listen to my words, listen to the words of someone who knows more about this than me who said there's very little point being a lead regulator in a sector in which you do not operate, and that not me, that was Emmanuel Macron.
00:58:10.079 --> 00:58:36.159
So there's clear tension between not least the member states and the EU because potentially it will have more of a lean towards risk and the control thereof, rather than the balance of understanding how to have a pro-innovation, pro-citizen right, pro-consumer protection approach.
00:58:36.320 --> 00:58:49.599
So there are there are opportunities for the UK to understand how we can both be interoperable, but also in our common law jurisdiction, take a legislative leadership opportunity.
00:58:50.239 --> 00:59:00.079
I was wondering about the area of biometrics and the interoperability on that, because I mean they're quite um they've been quite strident on where they will and and won't accept biometrics.
00:59:00.239 --> 00:59:16.800
And then back in the UK, obviously I've been following biometrics for a while, and uh UK biometrics commissioner resigned and he had a few things to say about what he thought the government were going to do and how they were potentially going to relax some of the the um the regulation around that.
00:59:16.880 --> 00:59:20.000
And I think a new biometrics commissioner has been appointed.
00:59:20.079 --> 00:59:22.320
Um, I'm sure you can correct me if I'm wrong on that.
00:59:22.480 --> 00:59:29.280
Um, but it seems to me like the biometrics, I mean, you talk a lot about public engagement, but AI and biometrics working together.
00:59:29.360 --> 00:59:41.360
And you and I talked about digital identity before, and the three things are very much connected, and it feels to me like they're they're very important and most obvious and visible to the to the general public.
00:59:41.599 --> 00:59:48.159
Um, where do you think we are with biometrics and what would you hope that your bill could achieve on that?
00:59:49.360 --> 01:00:04.880
It's a really good way to take us right back to basic principles, and that is that none of these AI systems are anything without data.
01:00:05.599 --> 01:00:09.840
And when we say di data, largely we mean our data.
01:00:10.239 --> 01:00:20.559
Some of that is very personal, some of it is IP protected, some of it is copyrighted, all of it is incredibly important.
01:00:22.079 --> 01:00:43.280
If we go to the most sensitive part of data, to my mind it really seems that biometrics are that such critical pieces of data, they are part of us, who we actually are.
01:00:44.559 --> 01:00:57.760
We're not where we need to be when it comes to biometrics, and yet there are very simple ways in which we need to get there, and what's within my bill is clear on that.
01:00:57.920 --> 01:01:05.039
It it goes to the sense of the need for everybody in this context.
01:01:05.280 --> 01:01:45.920
If biometrics are being captured, used, deployed, it's clear that has to be firstly done in an ethical way, done with consent, reported upon, all of that data declared that that has been used in the mix, effectively getting to that point of transparency, so everybody knows that that is being used, and crucially, the labelling clause that I have in the bill would also be assistive in this because it would require all goods or services where AI is in the mix to be labelled as such clearly.
01:01:46.159 --> 01:01:52.960
So citizens, customers, consumers would be well aware of that way before getting involved.
01:01:53.119 --> 01:02:01.679
So they could decide, we could decide, we could choose whether we wanted to be part of that, safe in the knowledge that AI is in the mix.
01:02:01.920 --> 01:02:08.559
So I've got the bill through second reading, it will then go to committee stage and report stage, that's where amendments can be made to it.
01:02:08.719 --> 01:02:18.000
Then third reading, it will then go down to the commons and do those stages down there, and if it gets through that, then it will go to the palace for the king to sign it into law.
01:02:18.320 --> 01:02:22.800
My hope is that I can get the entire bill enacted.
01:02:23.039 --> 01:02:35.360
If that's not possible, then I certainly hope that some of the key provisions can be adopted, not least around public engagement, not least around the protection of IP and copyright.
01:02:35.440 --> 01:02:52.559
Because at the moment, for anybody, you know, some of our fabulous creatives, our musicians, artists, writers, they're having their work just completely taken with no consent, no remuneration.
01:02:52.880 --> 01:02:54.159
That can't be right.
01:02:54.400 --> 01:03:03.119
And again, we know what we need to know to stop that being a problematic force in our society, in our economy right now.
01:03:03.519 --> 01:03:05.199
So positive.
01:03:05.360 --> 01:03:11.199
There's still a lot to do in terms of legislative process, in terms of further engagement with government.
01:03:11.360 --> 01:03:19.519
But I hope that certainly some of the key principles, some of the key elements will be taken on and will become part of legislation.
01:03:19.599 --> 01:03:41.679
And like all good legislation, once passed, it should seem so logical and just melt into part of how we go about living our lives alongside, enabled, empowered by AI, and being very much able to control, to eradicate, to mitigate, to manage the risks.
01:03:42.239 --> 01:03:42.880
Fantastic.
01:03:42.960 --> 01:03:50.239
Well, I'm sure lots of our listeners to this podcast will be really um interested in the bill and in helping any public engagement around it.
01:03:50.320 --> 01:03:55.519
So we'll obviously put a link to it in the show notes and we'll follow its progress very closely.
01:03:55.599 --> 01:03:57.679
Um, good luck with it, and thank you so much.
01:03:57.840 --> 01:04:00.000
Always a pleasure and privilege to speak to you, Chris.
01:04:00.079 --> 01:04:00.880
Thank you so much.
01:04:01.199 --> 01:04:08.639
No, likewise, thank you very much indeed, and really would love to hear from anybody their thoughts on any element of this.
01:04:09.119 --> 01:04:12.800
Do please hit me up at LinkedIn at Lord Chris Holmes.
01:04:12.960 --> 01:04:18.159
Lots more about the bill and the wider technology piece there, but we'd love to hear from all of you.
01:04:18.400 --> 01:04:18.960
Thank you very much.
01:04:19.119 --> 01:04:21.039
Great to speak to you as always.
01:04:26.800 --> 01:04:30.239
Thank you for listening to the Future of View, hosted by me, Tracy Follows.
01:04:30.320 --> 01:04:34.239
Be sure to check out the show notes for more info about the topics we covered today.
01:04:34.400 --> 01:04:38.960
If you enjoyed this episode, please like and subscribe wherever you get your podcasts.
01:04:39.119 --> 01:04:42.800
And if you know someone who would love this episode, please share it with them.
01:04:42.960 --> 01:04:48.800
For more on the future of identity in a digital world, visit the future made.group slash the future of you.
01:04:49.199 --> 01:04:53.840
To explore the future of everything else, head over to future made.group.
01:04:54.000 --> 01:04:57.599
The Future of View podcast is produced by Big Tent Media.
00:00:19.920 --> 00:00:21.600
Welcome to the future of you.
00:00:21.920 --> 00:00:24.719
In this episode, we're diving into the world of AI.
00:00:24.879 --> 00:00:31.920
We chat with a return guest to the pod, Lord Holmes, who's been busy pushing for new laws around artificial intelligence.
00:00:32.159 --> 00:00:38.719
He thinks it's time to get ahead of the game with AI and make sure that we have the rules that both encourage innovation and protect people.
00:00:38.880 --> 00:00:47.359
So he talks to me about the private members' bill he bought that includes the notion of an AI authority to oversee the changing landscape in the UK.
00:00:47.600 --> 00:01:04.319
Now the bill covers everything from consumer protection to intellectual property, but despite it moving through all of its stages in the Lords successfully, and as discussed before the general election, just as it was about to go to the Commons, it fell through the gaps in the wash-up of that general election.
00:01:04.560 --> 00:01:19.599
Several Lords had something to say about AI in the debate which followed the King's speech, but as we wait for greater clarity on the new government's position, it's probably best to look at the written update from Chris on his blog, and I've popped in a link in the show notes.
00:01:19.920 --> 00:01:24.079
It's also my great pleasure to speak with futurist Richard Yonk.
00:01:24.239 --> 00:01:35.599
He's also a former computing AI contributing editor for The Futurist magazine and a writer for Scientific American, World Future Review, Fast Company, GeekWire, Salon, and many more.
00:01:35.760 --> 00:01:47.599
He's been interviewed and quoted in the New York Times, Wall Street Journal, Wired, BBC News, etc., and guested on numerous podcasts and radio programmes, though the future review must surely be his favourite.
00:01:47.840 --> 00:01:58.560
He says it is his mission to provide organisations and individuals with insights that spark imagination, reduce risk exposure, and effectively connect the dots towards a preferred future.
00:01:58.719 --> 00:02:17.919
Now today he helps us connect some of the dots on AI, clearing up some of the myths and fears about it, reminding us it's still not anywhere close to human intelligence, and talking about the ups and downs of its development, the realistic timeline for things like self-driving cars, and the importance of balancing AI's benefits with ethics.
00:02:18.080 --> 00:02:21.599
So tune in as we break down some of these complex topics.
00:02:21.759 --> 00:02:23.280
Over to you, Richard.
00:02:42.960 --> 00:02:48.800
We have, in these last few years, reached a point where suddenly there's been some very big changes.
00:02:48.960 --> 00:02:54.400
People are of all sorts are very aware of it in a way that they never have been before.
00:02:54.639 --> 00:03:14.800
There's a lot of fear, there's a lot of concern built out of the narratives that we find from science fiction and movies and so forth, uh, that are perpetuated to a degree by uh some of the companies uh where we're worrying about AI taking over, AI replacing us entirely.
00:03:14.879 --> 00:03:34.879
Uh, but the fact is that when you talk to the people on the ground, the researchers, the people in the labs, the people who are actually doing the work, we are so far away from the point where AI is anything close to being uh human intelligence level.
00:03:35.039 --> 00:03:48.479
Uh we have a lot of things with generative AI, LLMs, uh Chat GPT, where we can point it at a particular task.
00:03:48.639 --> 00:04:10.319
It's it's taken in this enormous corpus of information from the internet, pretty much uh everything that we've been pouring onto the internet and the web for the last 30 years, and through statistical analysis and different types of algorithms, be able to take that information and answer amazing things.
00:04:10.479 --> 00:04:17.839
And it seems very creative, it seems very human-like, but in fact, what it is is it's us.
00:04:18.000 --> 00:04:21.279
We're we're seeing ourselves reflected back on us.
00:04:21.519 --> 00:04:47.360
And even though these programs can do amazing things, they can pass college exams, they can pass the bar, they can accomplish many different, very static tasks, they still don't have a lot of the very, very basic reasoning skills and understanding of causality and other aspects of our world that even a four-year-old has.
00:04:47.759 --> 00:05:05.839
So we're still at a very, very early phase with AI, but we're also at a phase where we we have to understand that it's a very different kind of intelligence, and we end up often misunderstanding what level it's at and what how we should be interacting with it.
00:05:06.160 --> 00:05:07.519
Yeah, so so that's very interesting.
00:05:07.600 --> 00:05:09.199
So what's driving the hype?
00:05:09.360 --> 00:05:10.319
Is it PR?
00:05:10.399 --> 00:05:11.199
Is it marketing?
00:05:11.360 --> 00:05:16.480
Is it all about research grants, or is it just, you know, ego of founders?
00:05:16.560 --> 00:05:20.319
What is the main driving force for the for the hype cycle?
00:05:20.800 --> 00:05:22.480
I think it's a lot of different things.
00:05:22.639 --> 00:05:25.040
We've been in a range of hype cycles.
00:05:25.120 --> 00:05:48.319
Uh I feel like they've been accelerating during the past decade or so uh as different uh business models have been kind of accelerating, and there's this feeling and need that we have to, that companies have to uh turn around and be showing profit very quickly, getting to market as quickly as possible, and essentially racing things out before they're ready.
00:05:48.560 --> 00:05:57.279
And so we get metaverses that are not ready for prime time, we get NFT markets that crash, and so forth.
00:05:57.519 --> 00:06:02.959
So, with AI, AI, as I say, has been around for 75, 80 years.
00:06:03.199 --> 00:06:29.680
It has gone through a range of ups and downs, AI winters, pullbacks in financing, and what these end up resulting in is actually a little bit of uh, you know, getting rid of the the chaff, having uh a little clearing, and there's an opportunity to kind of restart and and be funding uh viable uh you know directions and approaches again.
00:06:30.000 --> 00:06:36.480
The thing is that right now we're in a phase where there's a lot of money being thrown at it.
00:06:36.800 --> 00:06:44.959
You have the certainly the marketing departments, the founders wanting to talk things up as as far and as fast as they can.
00:06:45.279 --> 00:07:16.879
But a lot of the dialogue and concern, the ways that they're uh hyping this uh draw a lot on uh a range of dialogues, a range of conversations that have been on the certain corners of the web for and outside the web for several decades now, uh, where we're talking about and worrying about uh what happens when you know AI reaches the singularity, when it uh we reach uh an intelligence explosion, and who's going to protect us?
00:07:17.040 --> 00:07:21.360
Well, these companies have to, you know, they're the only ones who can take care of us.
00:07:21.519 --> 00:07:24.000
It at a certain level, I think they believe it.
00:07:24.160 --> 00:07:25.759
I think that there is ego there.
00:07:26.000 --> 00:07:38.399
Everyone who's working on something major in terms of technology throughout the ages thinks that what they're doing is the thing that's going to save humanity, that is the the great thing that nobody else can do without it.
00:07:38.480 --> 00:07:41.600
And they're the only ones who can bring it to light.
00:07:41.759 --> 00:07:51.600
The fact is that invention happens when it's ready, when the the time is right, when the the various infrastructure and uh conditions are set.
00:07:51.839 --> 00:07:54.000
So if not them, it will be somebody else.
00:07:54.319 --> 00:08:02.160
What do you think in terms of the um the regionalization of AI and the way we're treating it, particularly from a governance point of view, but not just that.
00:08:02.319 --> 00:08:06.639
I mean, even just thinking about data security, all those sorts of things.
00:08:06.720 --> 00:08:12.319
I mean, obviously what China are doing is a completely different approach to the EU, which again is a completely different approach to the states.
00:08:12.560 --> 00:08:15.279
You're in the states, I should say that for for for our listeners.
00:08:15.680 --> 00:08:18.959
What was your point of view on who's doing it best or better?
00:08:19.120 --> 00:08:21.199
Or can anybody do it that well right now?
00:08:21.600 --> 00:08:29.519
I think we're all finding our way, uh, most certainly, and everybody's got a different internal narrative that they're working from.
00:08:29.600 --> 00:08:38.240
They we've all got different political conditions and situations that we're starting from, so that alters the the path and the course.
00:08:38.480 --> 00:08:50.879
The EU, I feel, has a uh a really good start in terms of protecting the users' interests, the the citizens' interests.
00:08:51.120 --> 00:08:57.840
The GDPR is is an example pre-AI that certainly is involved there.
00:08:58.000 --> 00:09:03.919
I think that the idea of having that as a priority is important.
00:09:04.159 --> 00:09:23.279
I think that the idea where we have a free-for-all in the US where essentially we you know allow and have allowed, even with a small amount of noise and pushback, um, the scraping of all people's information and data that you know that you've got artists, you've got writers.
00:09:23.519 --> 00:09:28.480
I mean, I'm pretty sure your books are are you know part of Chat GPT.
00:09:28.559 --> 00:09:35.039
I know at least one of mine is, but it is uh a reality of our world right now.
00:09:35.279 --> 00:09:51.360
And the fact is that you've got these people who have said, okay, I'm going to basically create enormous profits for a very small number of people based on the product of the near sum of humanity.
00:09:51.519 --> 00:09:53.519
That's there's something wrong there.
00:09:53.840 --> 00:10:11.039
Because one, we're not having a say, but two, we need to have a range of different types of protections, not least of which is a level of balance and recognition toward, you know, if not compensation, at least control over our product and our data.
00:10:11.519 --> 00:10:13.279
What do you think about AI agents?
00:10:13.360 --> 00:10:26.480
Because um obviously my interest is in the future of identity and particularly identity, how it is influenced by technology, but very particularly by AI as we go forward, particularly with this podcast series.
00:10:26.799 --> 00:10:32.000
A lot of probably the work you do is about AI at an organizational level, and and some of mine is.
00:10:32.080 --> 00:10:41.039
But what do you think the impact is of agents or agentification on individuals, on persons, on personal identity?
00:10:41.840 --> 00:11:18.960
If you look at it from a very idealized standpoint, it has such enormous potential for helping us as individuals to whether it's possibly simplify our lives, possibly extend them in ways that simply has never been humanly possible before, uh, where our reality, our identity becomes something that extends well beyond us, be it is able to extend in ways where it diversifies in a way that you know we simply have never kind of inhabited throughout history.
00:11:19.200 --> 00:11:22.080
But that's a very idealized approach to it.
00:11:22.240 --> 00:11:33.759
At the more call it cynical viewpoint, these are typically going to be uh algorithms that are built by and controlled by corporations.
00:11:34.240 --> 00:11:42.639
So how do we silo our personal information, personal data, privacy in such a way that we are protected?
00:11:42.799 --> 00:11:49.120
And of course, that's even in an again idealized way, uh, that's much more easily said than done.
00:11:49.279 --> 00:11:58.879
Every everybody these days, as we've seen in in recent years, is is hacked on a regular basis and uh our our data and information exposed.
00:11:59.200 --> 00:12:22.720
So trying to look at the two poles of this and figure out, okay, where do we consider it as a reality in in somewhere hopefully in the middle where we're as protected as we can be, we get as much benefit as we can, but we're also not blind to the idea or the fact that we're probably being exposed and made vulnerable in ways that we never have been before.
00:12:23.440 --> 00:12:45.519
Because I've looked at a lot of models where one is using AI to create a twin, a digital twin, an avatar, more than one avatar's agents that go out and do things or say things on our behalf, but obviously reflect and represent or represent the self, you know, the tone of voice, the content, the the preferences, all those sorts of things.
00:12:45.600 --> 00:12:46.960
It's not there yet, obviously.
00:12:47.120 --> 00:12:48.480
There's some early examples.
00:12:48.799 --> 00:12:58.799
I think it's going to be very interesting when people are met with representations or representations of people and they don't really know whether it's the person or not.
00:12:58.879 --> 00:13:06.799
I think the assumption probably is going to be that you you know you're a representation, you're an avatar, rather than the assumption being you're a real person.
00:13:06.960 --> 00:13:15.279
And I think you've talked about this as well, that that sort of flip where we assume that we're dealing with a non-human rather than dealing with a human first.
00:13:15.759 --> 00:13:17.600
I think that's a fair assumption.
00:13:17.759 --> 00:13:19.840
It's going to depend on the modality.
00:13:20.159 --> 00:13:26.879
Right now, if we're back and forth with someone in text, where there's a whole range of things we don't get feedback on.
00:13:26.960 --> 00:13:39.039
And so it's pretty easy for us to say, okay, we're at a stage in technology where, yeah, there's a pretty good chance, you know, this is where I'm not dealing with the actual human being at this point.
00:13:39.279 --> 00:14:03.279
Uh, when we start talking about a range of you know, visual feedback, you know, people being represented as avatars, even entirely photorealistic, perfect representations of their voices and so forth, as we all become more familiar with the fact that yes, the technology is at this level right now, then we start to make a range of assumptions that we wouldn't have in the past.
00:14:03.519 --> 00:14:29.279
I think one of the things it brings to mind that I explored in my last book, Future Minds, was the idea that when you have all of these representations, all of these avatars, all of these identities going out into the world, if you go far enough into the future, and I'm not even sure how feasible this will ultimately be, you perhaps reintegrate that into your own memories, thoughts, and so forth.
00:14:29.440 --> 00:14:31.679
I'm not sure that if that's going to be feasible.
00:14:31.919 --> 00:14:36.879
But for the meantime, in the meantime, right now, what does that look like?
00:14:37.039 --> 00:14:53.519
If you have a hundred or a dozen entities that are you or represent you going out doing things, a lot of that is only useful to you if you're able to consolidate and take some of that information that they've gathered or interacted with back.
00:14:53.840 --> 00:14:58.080
Now we're already dealing with a world of massive data overload.
00:14:58.320 --> 00:15:02.320
Uh, how do we as human beings contend with that?
00:15:02.480 --> 00:15:11.840
I think that's a really big question we have to ask ourselves because that's going to be the the intermediary period before we ever get to that far future that I was talking about.
00:15:12.159 --> 00:15:13.759
I was going to ask you about consciousness.
00:15:14.080 --> 00:15:16.960
You mentioned it earlier on about whether AI can be conscious.
00:15:17.120 --> 00:15:23.600
What's your view on the sort of again another hype cycle or hype idea concept, the singularity?
00:15:24.159 --> 00:15:27.120
So um how long do you have?
00:15:27.840 --> 00:15:29.519
Till 2045.
00:15:30.879 --> 00:15:31.759
Well, yes.
00:15:31.919 --> 00:15:46.960
Um, so when I wrote Future Minds, uh, one of the main things that was driving that was this desire to really explore the nature of intelligence, not just human intelligence, but what is intelligence.
00:15:47.279 --> 00:15:52.960
And uh that in part because it's such a big thing.
00:15:53.039 --> 00:16:03.679
It it it really, if you look at every form of life on this planet, it has a form of intelligence that is appropriate to the niche that it occupies.
00:16:04.240 --> 00:16:13.360
Similarly, if we look out at uh artificial intelligence, machine intelligence, is that going to be a continuation of this path?
00:16:13.519 --> 00:16:21.039
Or is it fundamentally different for a range of reasons, not least of which is the fact that it doesn't begin from a biological substrate?
00:16:21.600 --> 00:16:32.399
So I think that uh the recognition that no entity, no species, uh shares the same kind of consciousness.
00:16:32.720 --> 00:16:40.639
All of it all species have a different level of intelligence, and what constitutes the consciousness they have is rather different.
00:16:40.960 --> 00:16:53.440
It goes back to the idea of you take a certain animal, uh a bat, for instance, we have no concept what it is like to be able to navigate with accolation.
00:16:53.600 --> 00:16:58.320
And so this is a a different way of being.
00:16:58.559 --> 00:17:01.360
Each entity has a very different form of it.
00:17:01.519 --> 00:17:02.720
What will AI have?
00:17:02.879 --> 00:17:04.799
What will machine intelligence have?
00:17:04.960 --> 00:17:15.119
I suspect there will be many, many different kinds of intelligence, many kinds of something of a form of consciousness, but it won't be like human consciousness.
00:17:15.279 --> 00:17:16.799
So let's back that up.
00:17:17.039 --> 00:17:21.759
Consciousness is what Marvin Minsky called an a suitcase word.
00:17:21.920 --> 00:17:29.279
It's essentially a word that means so many things to so many people, or depending on the context that you're using it in at that moment.
00:17:29.680 --> 00:17:38.160
So if we talk about it right now, most people are thinking of consciousness in terms of, well, is it self-aware like I am?
00:17:38.319 --> 00:17:40.480
Is it able to self-reflect?
00:17:40.559 --> 00:17:53.440
Uh, does it have uh a form of phenomenal uh consciousness, as Ned Bloch has talked about, where essentially it can, you know, register and appreciate a particular shade of red.
00:17:53.759 --> 00:17:54.720
But what is that?
00:17:54.880 --> 00:18:05.839
You know, it's it's a phenomenal experience and not something that can really be explained in a objective way, other than to talk about its wavelength.
00:18:06.160 --> 00:18:10.720
So, in that sense, will AI have that?
00:18:10.880 --> 00:18:14.160
And I suspect that it very possibly will not.
00:18:14.480 --> 00:18:25.759
I think we may have a level or a type of subjective consciousness, uh, because it's an emergent property of our minds, that is unique in the universe.
00:18:25.920 --> 00:18:40.720
Uh and until and unless we ever run across some other entity that has something we can at least partially agree is similar, then I suspect we may be very unique in that for a long, long time.
00:18:41.200 --> 00:18:44.960
Do you think we're going to be able to preserve our consciousness or bring it back?
00:18:45.039 --> 00:18:49.440
I'm thinking about cryonics and other technologies that are very emergent.
00:18:50.160 --> 00:18:55.839
So again, um so cryonics is a is kind of a subset of that.
00:18:56.000 --> 00:19:07.839
I think we have to you know back off a little bit and talk about okay, if we are such a very different substrate, if our minds exist in such a way, you have uh you know ideas.
00:19:08.000 --> 00:19:28.240
Ray Kurzweil has talked about the idea, oh, if we can emulate uh, and it's hardly he's hardly the only person, uh, if we can emulate each neuron perfectly, you know, through electronic means, right you know, transistors, and so forth, uh, then we can essentially replicate a human mind.
00:19:28.480 --> 00:19:52.319
Um it sounds good from a totally reductive standpoint, but the fact is that uh the nature of emergence, the nature of uh how some of these phenomena arise out of individual components that result through synergistic means to create a whole that is greater than the sum of its parts.
00:19:52.559 --> 00:19:55.279
Um I'm not sure that's going to be feasible.
00:19:55.440 --> 00:19:59.279
If you extend that then to okay, can we upload our minds?
00:19:59.440 --> 00:20:12.480
Can we record ourselves in a digital form in a computer and be able to continue to uh live and exist in that way, or you know, take that information and somehow put it back in a biological system?
00:20:12.799 --> 00:20:17.440
I think that that's going to be a really hard lift.
00:20:17.599 --> 00:20:24.400
I'm not sure that we have that capability in our future, but I'd love to be surprised.
00:20:24.799 --> 00:20:30.319
It's interesting, it is reductivist, and it's also a very rational way of thinking about human beings, I think.
00:20:30.400 --> 00:20:39.759
And I think we forget a lot of the time just how emotional we are, or so many of our responses are all, you know, they're all emotional or our decisions are generated from emotions.
00:20:39.839 --> 00:20:48.160
I mean, obviously, there's been lots and lots of work on this in the past couple of decades, but you've written a lot, I know, about AI and machines and emotions.
00:20:48.240 --> 00:20:50.799
And can they understand or detect our emotions?
00:20:50.960 --> 00:20:51.759
What could we do with that?
00:20:51.920 --> 00:20:53.599
What kind of effect will that have on us?
00:20:53.839 --> 00:20:56.319
Will we even be aware of the nudges?
00:20:56.480 --> 00:21:02.880
Can you talk us a bit through your sort of philosophy around that and your your beliefs about what will be possible and what won't?
00:21:03.359 --> 00:21:03.759
Sure.
00:21:04.000 --> 00:21:13.039
Well, thanks for the uh the the lead in uh yes, I I wrote a book, Heart of the Machine, Our Future in a World of Artificial Emotional Intelligence.
00:21:13.359 --> 00:21:20.319
And essentially it was looking at systems that are able to read, interpret, and interact with human emotions.
00:21:20.559 --> 00:21:57.200
This was work that started in really in the 90s, uh, really got going at MIT Media Lab in the mid-90s, started to be commercialized toward the end of the 2000s, and there was a lot of significant interest in it, and still is, but a range of concerns around the kinds of privacy uh that you know are it involves when you're interacting with and and being able to take in information about people visually, whether in a public environment or in a private one, you know, leads to a range of different types of problems.
00:21:57.440 --> 00:22:04.319
But then there's also the reality of Of what are these systems actually reading and recording?
00:22:04.640 --> 00:22:16.240
So human emotion is a core aspect of the human condition, something that essentially precedes our other forms of communication, such as speech and so forth.
00:22:16.559 --> 00:22:26.240
So, how fundamental that is, is I think something that is overlooked by a lot of people, especially in technology.
00:22:26.480 --> 00:22:32.799
But when you talk about taking a system that reads a person's face, for instance, a person smiles.
00:22:32.960 --> 00:22:34.400
Okay, it's they're smiling.
00:22:34.640 --> 00:22:37.519
But there is a huge number of nuances there.
00:22:37.599 --> 00:22:51.279
And beyond that, for any of us that as human beings have interacted with other people, we are continually in conversation, looking at their faces, taking the in expressions, looks in their eyes.
00:22:51.519 --> 00:22:57.200
I mean, so much information is being transmitted non-uhverbally.
00:22:57.680 --> 00:23:00.240
And there's been studies around that as well.
00:23:00.480 --> 00:23:07.440
And that information is continually feeding that conversation in a feedback loop between the two of you.
00:23:07.680 --> 00:23:10.799
It's referred to in psychology as theory of mind.
00:23:10.960 --> 00:23:20.480
Essentially, we are witnessing and experiencing this attempt to read each other, read each other's minds, you know, at a certain level.
00:23:20.640 --> 00:23:26.160
We're not reading each other's minds, but we are trying to understand what is that person thinking?
00:23:26.240 --> 00:23:28.160
How are they mad at me?
00:23:28.400 --> 00:23:30.640
Are they, you know, laughing at this?
00:23:30.799 --> 00:23:31.519
What have you?
00:23:31.759 --> 00:23:39.519
So when a machine takes that information, that visual information, and labels it, it's really not doing very much with it.
00:23:39.680 --> 00:23:43.119
It's just, you know, you could take an algorithm and say, well, they're smiling.
00:23:43.279 --> 00:23:44.480
We're going to say they're happy.
00:23:44.640 --> 00:23:46.000
Well, maybe they're not.
00:23:46.319 --> 00:23:51.440
You uh, I think mentioned in your book uh Lisa Feldman-Barrett, I mentioned her as well.
00:23:51.599 --> 00:23:57.599
And she definitely pushed back a few years ago on the whole emotion AI concept.
00:23:57.759 --> 00:23:58.960
And she's totally right.
00:23:59.200 --> 00:24:04.000
The fact is that these machines are not reading this information.
00:24:04.160 --> 00:24:15.200
And just as AI doesn't have the tools yet to reach full autonomous driving, for instance, it's still missing that last little piece.
00:24:15.440 --> 00:24:31.279
Right now, it doesn't have the reasoning capability that it may have in a few decades' time to be able to interact with us and have enough understanding of, you know, the nature of human beings, the nature of the world, the way we interact, and so forth.
00:24:31.440 --> 00:24:44.640
It may in the future be able to do a better job of essentially emulating or simulating that theory of mind kind of phenomena, but it's not going to be something that, again, it's not going to be as perfect as us for a long time.
00:24:44.880 --> 00:24:46.319
And we're not perfect.
00:24:47.119 --> 00:24:56.640
It's really interesting that you also talk there about driving because that was one of the big cells, wasn't it, a few years ago on emotive AI, emotional AI.
00:24:56.720 --> 00:25:06.319
You know, it's going to be in our cars, it's going to be reading our expressions, it's going to be know what mood we're in, and therefore it's going to create the atmosphere with the music, all these other things.
00:25:06.400 --> 00:25:20.720
I mean, we're so, so far off that we can't even get the car to drive properly without a human driver, let alone sort of program it all with all this, you know, emotional intelligence that is supposedly captured by a machine.
00:25:20.960 --> 00:25:26.000
So here's a question for a futurist, because it's one of the most difficult things, ones I think.
00:25:26.079 --> 00:25:26.240
Yeah.
00:25:26.480 --> 00:25:28.160
What are the timescales?
00:25:28.960 --> 00:25:31.599
What are the timescales for some of these technologies?
00:25:31.759 --> 00:25:35.680
What would be some of the milestones, particularly on AI, do you, do you think?
00:25:36.079 --> 00:25:36.880
Sure, sure.
00:25:37.200 --> 00:25:44.319
Try not to predict, but I do have a a range of you know general ideas that as far as where things are.
00:25:44.480 --> 00:25:58.000
Uh, I I typically tell people when they, you know, we we were told back in the 2000s, 2008, 2009, uh, by a range of executives, we were going to have the streets filled with self-driving cars by 2017.
00:25:58.240 --> 00:26:07.440
Every year in the last half dozen years, Elon Musk and others tell us we're one year away from full level five autonomous vehicles.
00:26:07.680 --> 00:26:09.200
We are not still not close.
00:26:09.440 --> 00:26:16.480
There's a range of problems, and these are these are that last mile problem that you have with a range of different technologies.
00:26:16.720 --> 00:26:29.599
So, in my mind, I've been saying to people, and we're closing in, and I'm getting I'm thinking about pushing it out a little bit, I've been telling people that we're not going to have it until at least 2030 in terms of technical capability.
00:26:29.759 --> 00:26:45.119
Then we have to get through a whole range of other factors because you have public sentiment, you have regulation, you have a whole transition period between the people giving up their cars, driving on those the roads at the same time.
00:26:45.279 --> 00:26:53.279
Uh, we're talking about well out into the 2030s before self-driving cars have fully taken over, and that may be optimistic.
00:26:53.519 --> 00:26:55.680
Other kinds of AI, it depends.
00:26:55.839 --> 00:27:13.119
I'm talking about you know, to people about the range of different types of programs and projects that are being done in research labs around uh different types of abstract reasoning emulation, you know, trying to understand, have key machine common sense and so forth.
00:27:13.279 --> 00:27:20.799
These are being developed, they're being worked on, but I think that it's a good 10, 20 years to see these fully be realized.
00:27:20.960 --> 00:27:33.519
In the meantime, the the thing that I really feel is going on now and will be driving a lot of AI for the next couple of decades is what I call, well, I'm not the only one, hybrid intelligence.
00:27:33.759 --> 00:28:01.359
The idea that humans and AIs working together in more and more frictionless ways, you know, between the intelligence of the machine, between the interfaces that are used, the different types of programs and algorithms to essentially help us to work together in a way where the systems do what they do so well, whether it's something highly repetitive, certain kinds of pattern recognition, and so forth.
00:28:01.519 --> 00:28:15.359
And we bring in a range of human-like skills that, you know, whether that's around reason, other aspects of emotion, values, and so forth, so that as a team, we're continually supporting each other.
00:28:15.519 --> 00:28:23.599
And you know, there's different small projects around this, but I think there's a whole range of ways this is going to expand over the coming decades.
00:28:24.000 --> 00:28:35.759
I'm really interested in this actually, because I think it's overlooked and underestimated just how difficult it is for human beings to work with AI, not for AI, but maybe in some cases it is.
00:28:35.920 --> 00:28:37.839
Um, nobody's taught us this.
00:28:38.079 --> 00:28:39.839
Who do we go to for it?
00:28:40.000 --> 00:28:43.839
And even for the people who are going to teach us, where's the pedagogy for it?
00:28:43.920 --> 00:28:47.759
I mean, there's if you track back, it's just so difficult to contemplate.
00:28:47.920 --> 00:28:56.160
I mean, I know how to go into a meeting with other people and discuss things in a conversation or a brainstorm or a you know, whatever the format is of the meeting.
00:28:56.240 --> 00:29:03.359
I know how to work on my own in front of my computer, not talking to it, just typing or whatever I'm doing, researching.
00:29:03.599 --> 00:29:09.599
But how do we work in these sort of whatever you want to call it, hybrid intelligence, hybrid working, co-piloting?
00:29:09.839 --> 00:29:11.759
I think it's a fascinating area.
00:29:12.160 --> 00:29:13.680
Yeah, no, it absolutely.
00:29:13.839 --> 00:29:15.759
And and it's rife with problems.
00:29:15.920 --> 00:29:17.920
I think there's going to be a lot of growing pains.
00:29:18.160 --> 00:29:32.240
And as human beings, again, kind of relearning or continually learning how to operate and how we use our values in the situation, I think is going to be difficult to, it's going to take us time.
00:29:32.480 --> 00:29:47.519
As an example, what's popping into my head is the other day I was on a uh platform with a number of other futurists, actually, uh being asked a range of questions for a conversation that was all being led by the platform's AI.
00:29:47.759 --> 00:29:55.839
So the platform was continually feeding us back information, prompting us for different things, and so forth.
00:29:56.000 --> 00:30:02.000
And it was definitely different, engaging, but it was also incredibly fast.
00:30:02.240 --> 00:30:04.880
And so, I mean, it has benefits.
00:30:05.039 --> 00:30:25.519
But for those of us who like to take time to do deep thinking, it really was not, I mean, it probably wasn't what it was geared for, but it really uh says something about uh we have to prioritize what kind of thinking and what kind of processes we're working on at a given time.
00:30:25.839 --> 00:30:29.680
Yeah, because immediately when you tell me that, I think, oh my God, that sounds awful.
00:30:29.839 --> 00:30:35.599
It feels like unless you're an ENTJ, Myers Briggs, then you aren't going to survive that meeting.
00:30:35.759 --> 00:30:38.240
And like we don't all want to be an ENTJ.
00:30:38.480 --> 00:30:44.880
No offense to ENTJs, but you know, um, you can have too much of a good thing.
00:30:45.599 --> 00:30:59.839
Um, yeah, so I mean, I feel like it's filtering potent potentially, obviously, filtering out some of the good stuff, some of the best stuff that humans do in order to be efficient, to operationalize, to optimize, and just to get an output and a result.
00:31:00.400 --> 00:31:00.960
Absolutely.
00:31:01.039 --> 00:31:07.359
It's something that I've I've been talking a little bit about about the nature of what we're doing with generative AI right now.
00:31:07.519 --> 00:31:20.160
Uh, the fact is that we have a lot of people who are touting and pushing uh this narrative that these different forms of generative AI can make us so much more efficient and we can do this and that.
00:31:20.400 --> 00:31:34.480
And when you look at some of the studies, there was a study six months ago, eight months ago, uh, where a range of consultants were using a form of generative AI, I want to say it was ChatGPT, but I'm not definite.
00:31:34.720 --> 00:31:40.160
And they showed that they had a X percentage of increased efficiency.
00:31:40.400 --> 00:32:08.319
But when you actually took it a little more granular uh and saw that the the consultants with the least amount of skill had theirs raised the most, and the better uh performing consultants uh really did not have that level of benefit, it makes you wonder what is the long-term effect uh of that kind of AI influence on human thinking, on the ways that we perform.
00:32:08.799 --> 00:32:13.839
We operate right now in a form of general competition with other human beings.
00:32:14.000 --> 00:32:37.920
If you have this intermediary that is operating in such a way that it kind of diminishes the value of exceeding expectations, but actually tends to make us all gravitate toward the mean, which is what an LLM does anyway, that that really is um concerning for what that could mean for the future of human thinking, human knowledge, and so forth.
00:32:38.319 --> 00:32:52.640
Yeah, I mean, do you think that whether it's a meeting or whether it's a piece of creative output or whatever the content is, do you think we'll have several different outputs that then may be aggregated or one can choose between?
00:32:52.880 --> 00:33:06.480
Because as you just said, you know, we can't really rely on just the analysis of AI and what whatever mode or format that is, because obviously it is potentially discarding lots of stuff that we would as humans find valuable.
00:33:06.640 --> 00:33:08.960
So how do we create a system like that?
00:33:09.039 --> 00:33:20.319
I mean, as you were talking about that particular meeting with the air, I was thinking, I wonder if you could have 12 different AIs going and they're all programmed in slightly different ways to analyze the output slightly differently.
00:33:20.480 --> 00:33:26.160
But even then, it seems like a massive waste of um energy when you could just rely on the humans to do it.
00:33:26.480 --> 00:33:42.000
Well, and that's the thing is I think that when you're talking about wanting to keep human values in the loop, that really, if anything, I I want in that situation for there to be more human beings rather than saying, oh, we need more AIs.
00:33:42.240 --> 00:33:48.799
I want to see, you know, a range of different kinds of thinking, a range of different kinds of views.
00:33:48.960 --> 00:33:50.880
Sure, that's that's diversity in general.
00:33:51.039 --> 00:33:59.519
That the there's a lot of benefits, especially in futures thinking, to have a range of viewpoints so that we can make sure that we don't have blind spots.
00:33:59.839 --> 00:34:16.000
But I really would not be willing to I I think if you put it all on, say, 12 AIs, I've got a feeling you're going to have a huge number of blind spots, particularly in terms of what is valuable or important to human beings.
00:34:17.760 --> 00:34:35.039
When I've been looking at some of the particularly the tech entrepreneurs, but a lot of the smaller AI companies and all this, I'm getting concerned that there's um an assumption, even a desirability around the AI will program us with our values and rather than the other way around.
00:34:35.280 --> 00:34:40.880
And obviously there's a kind of there's a togling between the two, and it's a bit of a compromise and negotiation.
00:34:41.039 --> 00:34:45.840
But even then, the human being has to have the final say over those values.
00:34:46.320 --> 00:34:48.719
But can the human being always know?
00:34:48.960 --> 00:34:55.119
I mean, we've seen what happened with Google Gemini and lots of lots of things recently that have, yeah, exactly.
00:34:55.280 --> 00:34:57.119
You're rolling your eyes, I on too.
00:34:57.199 --> 00:35:01.360
I mean, I mean, sometimes I look at that and think, how how could they have gone out with that?
00:35:01.519 --> 00:35:06.239
Was that just a test to see how deplorable it would be and what people's reactions would be?
00:35:06.320 --> 00:35:08.320
Well, that was that actually a mistake.
00:35:08.639 --> 00:35:08.880
Sure.
00:35:09.039 --> 00:35:11.360
I'm just going to recap for some of your listeners.
00:35:11.599 --> 00:35:26.239
So back in February, Google released a version of uh Gemini that had it changed the algorithm slightly to try to implement more diversity into the uh results for different types of generation, particularly image generation.
00:35:26.480 --> 00:35:35.760
And they were creating these very, very and consistently, not just one-offs or just because you asked for it.
00:35:35.920 --> 00:35:44.400
They were creating uh representations of Vikings and World War, well, Nazis, I guess.
00:35:44.559 --> 00:35:57.199
And others you you had uh you had who were basically um people of various ethnicities that did not participate for very good reason in those events.
00:35:57.440 --> 00:36:00.320
There was a Native American in ancient Greece.
00:36:00.400 --> 00:36:12.000
Uh, you know, there was just all of these really ridiculous uh visuals that it's fine if you asked for it, but if it's just how it's going to generate it, it's like, hey, you guys messed up bad.
00:36:12.159 --> 00:36:21.599
Uh so yes, I unfortunately they've had a range of from a marketing PR standpoint, a range of bad demos here over the last number of years.
00:36:21.760 --> 00:36:28.079
I can actually think of a few different instances, and it's like, I'm not sure what's happening at that level or with that.
00:36:28.159 --> 00:36:30.960
But anyway, don't really want to get into that too much.
00:36:31.199 --> 00:36:41.440
I've said for a long time, and hardly the only person out there saying it, that technology has always changed us as it has developed.
00:36:41.599 --> 00:36:47.519
We create technology, it alters us in some way, regardless of what it is.
00:36:47.599 --> 00:36:53.039
Uh, and it's become more and more apparent with the modern technologies like social media and so forth.
00:36:53.280 --> 00:36:59.199
So that you know, now most of us, or a lot of us, uh, read short form only.
00:36:59.360 --> 00:37:02.639
Reading long form is much, much harder than it used to be.
00:37:02.719 --> 00:37:06.480
Our our the way that our brains process is actually quite different.
00:37:06.800 --> 00:37:24.639
We have uh some small, you know, little you know, bone growth in parts of our body, you know, because we have a different posture than we used to have, you know, years ago because of how we spend most of our time at desks and in front of machines these days or in front of uh mobile phones, so on and so forth.
00:37:24.800 --> 00:37:29.119
Technology continues either physically or mentally to alter us.
00:37:29.360 --> 00:37:33.280
The question then becomes is this something that we need to do?
00:37:33.519 --> 00:37:38.559
We behave and operate differently in terms of how we use a computer early on.
00:37:38.719 --> 00:37:40.639
We had to learn a command line interface.
00:37:40.800 --> 00:37:43.360
That was, you know, for a lot of people, that was tough.
00:37:43.519 --> 00:37:57.199
As it became an easier and easier interface, it became more natural, but we still do really weird things when you think about it in terms of how we ordinarily would operate in nature, in order to be able to make these machines work.
00:37:57.280 --> 00:38:01.039
We operate at a certain level on their terms rather than the other way around.
00:38:01.360 --> 00:38:03.599
We shape our tools and our tools shape us.
00:38:03.840 --> 00:38:24.320
With that in mind, um, my final question for you, I think, Richard, is given everything we've been talking about and the supposed benefits and advantages of AI, if it's used correctly, or we find ways to work with it in a hybrid manner, is the future going to be more complex or less complex with AI in it?
00:38:25.679 --> 00:38:29.519
So uh the future will be more complex.
00:38:29.760 --> 00:38:34.000
Decade after decade, century after century, we create abstraction upon abstraction.
00:38:34.079 --> 00:38:38.400
We we keep creating and adding complexity to our world and to our systems.
00:38:38.960 --> 00:38:49.280
We may feel at certain times or in certain ways that our lives are simpler, but overall, I think you we have to say that it will be more complex.
00:38:49.519 --> 00:38:50.400
Will it be better?
00:38:50.639 --> 00:38:51.840
Will it be worse?
00:38:52.000 --> 00:39:13.760
That's I think a far more important for me question in terms of is what we're creating going to be improving and beneficial to the human condition, or are we doing something that is ultimately going to be against our nature, be against what it means to be human beings?
00:39:14.000 --> 00:39:28.320
I don't want to live in a world where essentially all of us spend all of our time feeding the social media machines and all of the different systems in order that they can operate.
00:39:28.559 --> 00:39:40.719
I think we all have to still have the experiential life that we have in nature, in the world with other people, in order to be able to have full and rich lives.
00:39:43.519 --> 00:39:49.599
So welcome again to Lord Holmes, Chris, uh second appearance on the podcast.
00:39:49.760 --> 00:39:50.559
Thank you for returning.
00:39:51.039 --> 00:39:51.360
Pleasure.
00:39:51.760 --> 00:40:03.920
I wonder if you could tell us a bit about the artificial intelligence regulation bill that you have introduced and um bring us up to speed on where we are in its like is it its second reading, or is it it just received its second reading?
00:40:04.639 --> 00:40:17.599
Certainly, and the reason I wanted to introduce a piece of legislation on artificial intelligence was I think that we have an opportunity to lead and it's now time to legislate.
00:40:17.920 --> 00:40:21.440
The government's position is largely wait and see.
00:40:21.920 --> 00:40:34.880
But I think we need to know everything that we need to be in the right position for pro-innovation, pro-citizen rights, and pro-consumer protection, which is why I introduced my bill.
00:40:34.960 --> 00:40:40.559
And to give some context, every member of parliament has the opportunity to introduce a private member's bill.
00:40:40.639 --> 00:40:44.559
It can be on any subject they want, it needs to be relatively specific.
00:40:44.880 --> 00:40:58.320
And then there's an element of luck because it goes into a ballot at the beginning of every session, and you need to really come in the top 25 in the ballot if your bill is going to have a chance of getting before parliament.
00:40:58.559 --> 00:41:07.199
I came sixth, so I had first reading of my bill in November, and that means that's just when I present the bill to Parliament and read its long title.
00:41:07.360 --> 00:41:10.639
And we then had a second reading on the 22nd of March.
00:41:10.719 --> 00:41:18.960
So that's the opportunity for a full debate on all of the issues and all of the clauses within the bill.
00:41:19.280 --> 00:41:39.599
So I wanted to have a bill which addressed the innovation question, the investment question, crucially consumer protection, citizen rights, and to really drive a far greater level of public engagement with the opportunities and indeed the challenges of artificial intelligence.
00:41:40.320 --> 00:41:47.039
But why have you felt the need to offer up this bill and uh we'll we'll get on to some of the details.
00:41:47.199 --> 00:41:48.239
What do you think is missing?
00:41:48.320 --> 00:41:48.800
What's the gap?
00:41:48.960 --> 00:41:58.000
Obviously, the EU are doing their AI Act and other regions or territories are looking at their own sort of national or regional regulatory solutions.
00:41:58.159 --> 00:42:01.039
What is it that you think is missing in the UK?
00:42:01.840 --> 00:42:14.000
We had the Bletchley Safety Summit, the AI Safety Summit in November last year, and that was very successful and right to look at frontier risks and that question of safety.
00:42:14.320 --> 00:42:35.280
But having stood up a successful summit, it seems logical to then look at all of the other elements of AI which are already impacting people's lives, oftentimes without those people even knowing it in terms of shortlisting for recruitment, for example, in higher education, in healthcare decisions, people are already being impacted by AI.
00:42:35.519 --> 00:43:03.440
It's time that we legislate so we can cover all of the areas where AI is currently being deployed, all the areas where it may be, so we get the best outcomes for citizens, for companies, for innovators, for IP owners, for our nation, and we have a real opportunity with three L's I call them law, learning, and language.
00:43:03.599 --> 00:43:18.159
We have the great good fortune of English common law, we have a sensational higher and further education ecosystem, and we have the good fortune of English as our language.
00:43:18.480 --> 00:43:31.280
With that, we have a responsibility, I believe, not just an opportunity, a responsibility to lead internationally focused to lead on artificial intelligence.
00:43:31.760 --> 00:43:42.000
Okay, so why doesn't the existing landscape of many regulators in the UK, and I've sat on the board of one of them, so um I am familiar with it.
00:43:42.159 --> 00:43:54.400
For example, why doesn't the ICO or why doesn't Ofcom or even in some respects the advertising standards authority, why doesn't the current regulatory landscape fulfill the asks of your bill?
00:43:54.960 --> 00:43:59.599
It's interesting because the government's approach is very much to look to existing regulators.
00:44:00.320 --> 00:44:09.280
To take on this work around AI in their specific sector in a very vertical manner.
00:44:09.840 --> 00:44:28.800
Now, both in terms of consistency of approach, coherence, and ensuring that we identify all the gaps where there aren't competent regulators for a particular area and where existing regulators don't have competency in that specific field.
00:44:29.039 --> 00:44:32.800
It seems to me it would make sense to have an AI authority.
00:44:33.199 --> 00:44:42.400
And in no sense am I suggesting that that should be a do-it-all AI, huge, bureaucratic, cumbersome, expensive regulator, not a bit of it.
00:44:42.719 --> 00:44:59.519
Agile, flexible, horizontally focused, to do that piece of work, to look across existing regulators, to assess their competency, to address the issues, the challenges, the opportunities of AI, to look across all relevant existing legislation to assess its competency, to address the challenges and the opportunities of AI.
00:44:59.679 --> 00:45:02.239
And by doing that, identify the gaps.
00:45:02.559 --> 00:45:23.199
So then it wouldn't matter if you're an innovator, a consumer, a citizen, in whatever sphere of the economy or life you've found yourself in, if AI was in the mix, you could expect to have a consistent and a coherent approach wherever you happen to be experiencing that AI.
00:45:24.559 --> 00:45:31.440
So it's like it's smoothing out the potential wrinkles across all of the existing bodies, is it, in different sectors?
00:45:31.840 --> 00:45:46.639
But very much taking a horizontal approach to thread this across all of the existing regulators and to identify where there are gaps and uh changes which need to be made to enable that consistency and coherence across the whole of the economy and society when it comes to AI.
00:45:47.039 --> 00:45:49.679
Because it's it's interesting the timing of this as well.
00:45:49.840 --> 00:45:54.800
I mean, it's sort of surprising to me actually that this is coming at this time.
00:45:55.039 --> 00:45:59.199
Why didn't it happen five years ago when we knew some of this was on the horizon, for example?
00:45:59.280 --> 00:46:01.760
So I'm always going to ask this as a futurist, you know.
00:46:01.920 --> 00:46:03.440
We saw a lot of this coming.
00:46:03.599 --> 00:46:04.800
I know you did as well.
00:46:04.880 --> 00:46:12.960
Um, but you've got the ICO, for example, and not to pick on them, but the ICO, so only just going out to consultation on generative AI, for example.
00:46:13.119 --> 00:46:15.760
I mean, it's done, it's happening, it's happened.
00:46:15.920 --> 00:46:22.159
Um, so what's your point of view on getting ahead of some of these things and why haven't we on AI?
00:46:22.239 --> 00:46:23.360
Or perhaps you think we have?
00:46:23.760 --> 00:46:40.719
No, you're you're you're quite right that in terms of our startups, our scale-ups, our industries, many of our universities and further education establishments, they're doing great work when it comes to AI.
00:46:41.119 --> 00:46:46.880
But from a legislative and a regulatory perspective, we're certainly somewhat behind the curve.
00:46:46.960 --> 00:46:49.280
And we didn't need to get behind the curve.
00:46:49.599 --> 00:46:54.320
We can now, I suppose, take the opportunity of not being first mover.
00:46:54.400 --> 00:47:03.119
The EU have very much got out there and uh done that with the EU AI Act, so we don't have to have that fear of the focus on first mover.
00:47:03.920 --> 00:47:06.400
But we do need to get on with it.
00:47:06.719 --> 00:47:27.039
And as you rightly identify, if we don't, the danger is that so much, particularly in the generative AI space, so much will have happened, it'll be very difficult to retrospectively regulate and to try and have that positive, forward-facing approach, which we still could have at this stage, I believe.
00:47:27.199 --> 00:47:35.360
But as every week goes on, obviously you're giving yourselves less of an opportunity to influence to lead.
00:47:36.239 --> 00:47:44.480
Yeah, so so tell me a little bit about the AI authority, who would lead it, and also your idea of an AI officer.
00:47:44.639 --> 00:47:46.000
How would that work?
00:47:46.480 --> 00:47:47.039
Certainly.
00:47:47.119 --> 00:48:02.400
So again, it's all about trying to give us the best opportunity to enable the opportunities and be cognizant of the challenges and control and where appropriate, eradicate the risks.
00:48:02.559 --> 00:48:05.840
So the AI authority would be independent of government.
00:48:06.079 --> 00:48:10.960
It would have key functions, as I set out in terms of looking at existing legislation.
00:48:11.199 --> 00:48:31.760
Crucially, like the whole of the AI regulation bill, the regulator, the AI authority would be underpinned by principles, key principles that should run through all of our thinking on this: trust, transparency, inclusion, innovation, interoperability, an international approach, accountability, accessibility, and assurance.
00:48:32.000 --> 00:48:35.519
Great principles which can underpin our approach to this.
00:48:35.760 --> 00:48:51.119
And having then principles-based, outcomes focused, inputs understood, and where necessary remunerated gives us the best opportunity, I believe, to get the best out of AI.
00:48:51.360 --> 00:48:58.239
Added to that, the AI responsible officer, another key principle running through to build that of proportionality.
00:48:58.400 --> 00:49:04.559
So people shouldn't think this has to be an individual who all their time is on AI.
00:49:04.719 --> 00:49:12.320
It's proportionate to the size of the organization, it's proportionate to their use of artificial intelligence.
00:49:12.480 --> 00:49:39.119
But to have a function within every organization that develops, deploys, uses AI, where that function is responsible for the ethical, unbiased use of AI, with reporting obligations analogous to those for financial reporting within the companies act, so well understood by all the commercial entities, seems to be another good focal point.
00:49:39.280 --> 00:50:12.320
Not that then that AI officer is the only person responsible, everything is just abdicated to her or him, but to be that function within an organization, which then enables as a company grows, or if it already is of scale, for the entire board, the whole of Exco and layers of the organization to understand their responsibilities and to understand the leadership and the culture that's required for inclusive by design, unbiased, ethical AI development and deployment.
00:50:13.039 --> 00:50:14.559
It's interesting to say about boards.
00:50:14.639 --> 00:50:20.559
You and I were both at the same AI conference yesterday and the day before, which I think we both enjoyed.
00:50:20.719 --> 00:50:25.679
Um that the question of boards came up actually in the panel discussion I was doing.
00:50:25.840 --> 00:50:31.039
And one of the attendees asked, What do you think the board of the future is going to be like given AI?
00:50:31.199 --> 00:50:40.719
Do you have any thoughts or or visions or or hopes or even fears actually about what the future of an AI or AI-infused or assisted board might be?
00:50:41.679 --> 00:50:56.480
There should be confidence in all individuals and all humans that AI always needs to be human-led, always needs to be human in the loop, and then potentially in some ways human over the loop.
00:50:56.800 --> 00:51:28.480
But what that should mean for boards, if we get this right, is greatly enabled, greatly empowered boards to be able to have information in a different format, right in front of people, real-time information, more than real-time with predictive analytics, being able to have such a different view of your organization, of your market, of your sector, of your competition, of your environment.
00:51:28.719 --> 00:51:42.159
All this is it if we get it right and we must, it should lead to far more empowered boards being enabled to have even greater leadership capabilities and leadership potential.
00:51:42.880 --> 00:51:49.920
Now, businesses, I'm guessing, are going to say, here comes more regulation or more burdens on us.
00:51:50.079 --> 00:51:53.039
We've already got a quite a lot to cope with.
00:51:53.199 --> 00:51:58.639
Is the government going to help us, you know, either with funding or any other kinds of aid?
00:51:58.719 --> 00:52:01.039
What what do you have any thoughts on that?
00:52:01.679 --> 00:52:02.320
Very much.
00:52:02.400 --> 00:52:16.079
I and I was really clear when I started that draft in this bill that it can't be a burden, either bureaucratically, financially, operationally, on business.
00:52:16.480 --> 00:52:24.480
That key principle of proportionality is one guard against burden.
00:52:25.360 --> 00:52:32.559
And indeed, all of us can cite bits of legislation and regulation that is burdensome.
00:52:32.639 --> 00:52:39.119
But that doesn't mean regulation and legislation of itself is bad, it means that's the wrong regulation, that's the wrong legislation.
00:52:39.360 --> 00:52:58.880
So really pushing these principles around right-sized regulation, proportionate, a sense that if you are a startup, potentially there can be de minimis levels as to as to how the features will impact.
00:52:59.840 --> 00:53:23.760
So all of the provisions should really be pro-innovation, pro-consumer protection, pro-citizen rights, and it should be the job of any legislature and any regulator to have all those principles in their mind when they're drafting, because it makes it more complex, but it's absolutely essential.
00:53:24.239 --> 00:53:29.360
Was there anything that came through the second reading or anywhere during the process so far?
00:53:29.599 --> 00:53:35.519
Any comments from either any of the other lords or ladies, or anything that's surprised you?
00:53:35.679 --> 00:53:45.599
Any kind of either you know, challenges or what what what's what's been interesting in the debate that obviously your bill has kind of encouraged?
00:53:46.320 --> 00:54:06.800
I did a great deal of consultation ahead of drafting, never mind presenting the bill with civil society, with the creatives with IP, with copyright, with people interested in our democracy, with businesses, to try and find the right path through this.
00:54:07.199 --> 00:54:15.599
What I was really pleased with was the level of cross-party support for the bill.
00:54:15.840 --> 00:54:19.440
All major parties and none were in support.
00:54:19.760 --> 00:54:32.239
There were the obvious criticisms from some quarters saying that we shouldn't be regulating at all, we should just let this technology emerge and only at that point do something.
00:54:32.400 --> 00:54:38.079
And that's that's a a reasonable position for some colleagues to take.
00:54:42.000 --> 00:54:51.360
Because nothing which is in my bill would be a curb on innovation, on development, on growth.
00:54:51.519 --> 00:54:52.559
Not at all.
00:54:53.039 --> 00:55:10.079
The key is if we get right size regulation in place, as all the lessons from history demonstrate, that is a benefit to innovation that crowds in inward investment.
00:55:10.320 --> 00:55:22.639
And two obvious examples: the fintech regulatory sandbox, a clear regulatory intervention, measure of success, well, replicated in well over 50 jurisdictions around the world.
00:55:22.960 --> 00:55:34.639
Open banking, a great UK creation, a creature of a regulatory intervention with the CMA9 order, now replicated in well over 60 jurisdictions around the world.
00:55:34.800 --> 00:55:58.400
So we know how to do this, we know how to do it to market make, market shape, enable innovation, because right-size regulation is a deliverant of the clarity and the certainty that investors, the developers, the startups that really all of us would want to see in any particular area of our lives.
00:55:59.679 --> 00:56:05.199
It's fascinating listening to you talking about in such uh positive, innovative ways.
00:56:05.280 --> 00:56:13.360
Um, and I wonder what your point of view is on the EU AI Act, and because it's comes from such a different point of view.
00:56:13.519 --> 00:56:18.079
Obviously, they're you know looking at the very specific threats and they've got these risk levels.
00:56:18.239 --> 00:56:25.360
Everything feels terrifying uh rather than you know, in any way sort of optimistic and and and innovative.
00:56:25.440 --> 00:56:36.320
And I mean, obviously, you've got interoperability as one of your principles, and of course that stands across you know businesses, uh regulators in in the UK.
00:56:36.400 --> 00:56:40.400
But what about the interoperability with the EU and their act?
00:56:40.480 --> 00:56:43.519
I'd love to get your point of view on that.
00:56:44.400 --> 00:56:50.800
Regulatory interoperability is a critical part of my bill as well.
00:56:51.280 --> 00:56:59.119
The EU AI Act is a valiant effort in all of its 892 pages.
00:57:00.000 --> 00:57:17.280
It's largely a creature of the legal framework in which it necessarily finds itself, so a code-based legal system, which necessarily tends towards a more prescriptive approach.
00:57:17.519 --> 00:57:29.760
I think you'll see the development of the act with all of the elements around uh generative AI coming in at some somewhat of a late stage of the drafting process.
00:57:30.719 --> 00:57:34.480
And I think it will be quite difficult in some ways.
00:57:34.559 --> 00:57:44.960
It's quite difficult to see in some circumstances where the right of a redress would come from, which authority you would be appealing to.
00:57:45.119 --> 00:57:47.440
There's some clashes there, certainly.
00:57:47.679 --> 00:58:09.840
And almost in many ways, don't listen to my words, listen to the words of someone who knows more about this than me who said there's very little point being a lead regulator in a sector in which you do not operate, and that not me, that was Emmanuel Macron.
00:58:10.079 --> 00:58:36.159
So there's clear tension between not least the member states and the EU because potentially it will have more of a lean towards risk and the control thereof, rather than the balance of understanding how to have a pro-innovation, pro-citizen right, pro-consumer protection approach.
00:58:36.320 --> 00:58:49.599
So there are there are opportunities for the UK to understand how we can both be interoperable, but also in our common law jurisdiction, take a legislative leadership opportunity.
00:58:50.239 --> 00:59:00.079
I was wondering about the area of biometrics and the interoperability on that, because I mean they're quite um they've been quite strident on where they will and and won't accept biometrics.
00:59:00.239 --> 00:59:16.800
And then back in the UK, obviously I've been following biometrics for a while, and uh UK biometrics commissioner resigned and he had a few things to say about what he thought the government were going to do and how they were potentially going to relax some of the the um the regulation around that.
00:59:16.880 --> 00:59:20.000
And I think a new biometrics commissioner has been appointed.
00:59:20.079 --> 00:59:22.320
Um, I'm sure you can correct me if I'm wrong on that.
00:59:22.480 --> 00:59:29.280
Um, but it seems to me like the biometrics, I mean, you talk a lot about public engagement, but AI and biometrics working together.
00:59:29.360 --> 00:59:41.360
And you and I talked about digital identity before, and the three things are very much connected, and it feels to me like they're they're very important and most obvious and visible to the to the general public.
00:59:41.599 --> 00:59:48.159
Um, where do you think we are with biometrics and what would you hope that your bill could achieve on that?
00:59:49.360 --> 01:00:04.880
It's a really good way to take us right back to basic principles, and that is that none of these AI systems are anything without data.
01:00:05.599 --> 01:00:09.840
And when we say di data, largely we mean our data.
01:00:10.239 --> 01:00:20.559
Some of that is very personal, some of it is IP protected, some of it is copyrighted, all of it is incredibly important.
01:00:22.079 --> 01:00:43.280
If we go to the most sensitive part of data, to my mind it really seems that biometrics are that such critical pieces of data, they are part of us, who we actually are.
01:00:44.559 --> 01:00:57.760
We're not where we need to be when it comes to biometrics, and yet there are very simple ways in which we need to get there, and what's within my bill is clear on that.
01:00:57.920 --> 01:01:05.039
It it goes to the sense of the need for everybody in this context.
01:01:05.280 --> 01:01:45.920
If biometrics are being captured, used, deployed, it's clear that has to be firstly done in an ethical way, done with consent, reported upon, all of that data declared that that has been used in the mix, effectively getting to that point of transparency, so everybody knows that that is being used, and crucially, the labelling clause that I have in the bill would also be assistive in this because it would require all goods or services where AI is in the mix to be labelled as such clearly.
01:01:46.159 --> 01:01:52.960
So citizens, customers, consumers would be well aware of that way before getting involved.
01:01:53.119 --> 01:02:01.679
So they could decide, we could decide, we could choose whether we wanted to be part of that, safe in the knowledge that AI is in the mix.
01:02:01.920 --> 01:02:08.559
So I've got the bill through second reading, it will then go to committee stage and report stage, that's where amendments can be made to it.
01:02:08.719 --> 01:02:18.000
Then third reading, it will then go down to the commons and do those stages down there, and if it gets through that, then it will go to the palace for the king to sign it into law.
01:02:18.320 --> 01:02:22.800
My hope is that I can get the entire bill enacted.
01:02:23.039 --> 01:02:35.360
If that's not possible, then I certainly hope that some of the key provisions can be adopted, not least around public engagement, not least around the protection of IP and copyright.
01:02:35.440 --> 01:02:52.559
Because at the moment, for anybody, you know, some of our fabulous creatives, our musicians, artists, writers, they're having their work just completely taken with no consent, no remuneration.
01:02:52.880 --> 01:02:54.159
That can't be right.
01:02:54.400 --> 01:03:03.119
And again, we know what we need to know to stop that being a problematic force in our society, in our economy right now.
01:03:03.519 --> 01:03:05.199
So positive.
01:03:05.360 --> 01:03:11.199
There's still a lot to do in terms of legislative process, in terms of further engagement with government.
01:03:11.360 --> 01:03:19.519
But I hope that certainly some of the key principles, some of the key elements will be taken on and will become part of legislation.
01:03:19.599 --> 01:03:41.679
And like all good legislation, once passed, it should seem so logical and just melt into part of how we go about living our lives alongside, enabled, empowered by AI, and being very much able to control, to eradicate, to mitigate, to manage the risks.
01:03:42.239 --> 01:03:42.880
Fantastic.
01:03:42.960 --> 01:03:50.239
Well, I'm sure lots of our listeners to this podcast will be really um interested in the bill and in helping any public engagement around it.
01:03:50.320 --> 01:03:55.519
So we'll obviously put a link to it in the show notes and we'll follow its progress very closely.
01:03:55.599 --> 01:03:57.679
Um, good luck with it, and thank you so much.
01:03:57.840 --> 01:04:00.000
Always a pleasure and privilege to speak to you, Chris.
01:04:00.079 --> 01:04:00.880
Thank you so much.
01:04:01.199 --> 01:04:08.639
No, likewise, thank you very much indeed, and really would love to hear from anybody their thoughts on any element of this.
01:04:09.119 --> 01:04:12.800
Do please hit me up at LinkedIn at Lord Chris Holmes.
01:04:12.960 --> 01:04:18.159
Lots more about the bill and the wider technology piece there, but we'd love to hear from all of you.
01:04:18.400 --> 01:04:18.960
Thank you very much.
01:04:19.119 --> 01:04:21.039
Great to speak to you as always.
01:04:26.800 --> 01:04:30.239
Thank you for listening to the Future of View, hosted by me, Tracy Follows.
01:04:30.320 --> 01:04:34.239
Be sure to check out the show notes for more info about the topics we covered today.
01:04:34.400 --> 01:04:38.960
If you enjoyed this episode, please like and subscribe wherever you get your podcasts.
01:04:39.119 --> 01:04:42.800
And if you know someone who would love this episode, please share it with them.
01:04:42.960 --> 01:04:48.800
For more on the future of identity in a digital world, visit the future made.group slash the future of you.
01:04:49.199 --> 01:04:53.840
To explore the future of everything else, head over to future made.group.
01:04:54.000 --> 01:04:57.599
The Future of View podcast is produced by Big Tent Media.
