WEBVTT
00:00:00.000 --> 00:00:03.520
Did you vote for Sam Altman, Mark Zuckerberg, Elon Musk?
00:00:03.759 --> 00:00:04.959
No, me neither.
00:00:05.120 --> 00:00:14.240
But somehow they and a handful of other tech billionaires are making decisions about AI that completely change your life, your kids' lives, and our democracy.
00:00:14.400 --> 00:00:20.000
So now it's time for us to ask: why do tech billionaires have so much power in deciding our lives?
00:00:20.160 --> 00:00:21.839
Is this actually democratic?
00:00:22.000 --> 00:00:23.600
And what can we do about it?
00:00:23.760 --> 00:00:26.079
So that's exactly what we will be discussing today.
00:00:26.239 --> 00:00:27.760
Hi, my name is Katarina Doria.
00:00:27.920 --> 00:00:35.280
I'm an AI ethicist, and this is Bots Don't Cry, the podcast where every Tuesday I'll teach you critical AI literacy with other critical AI experts.
00:00:35.439 --> 00:00:41.759
And today we are super duper lucky to actually get to unpack AI and democracy with political philosopher Dr.
00:00:41.920 --> 00:00:42.960
Annette Zimmerman.
00:00:43.039 --> 00:00:43.280
Dr.
00:00:43.359 --> 00:00:50.560
Annette Zimmerman is a tenured track professor of philosophy and co-founder of the Center for Uncertainty and AI at UW Madison.
00:00:50.799 --> 00:01:00.640
They wrote Democratizing AI, won the APA's Public Philosophy Prize, and have advised the US, UK, and German governments, as well as UNESCO and the OECD.
00:01:00.880 --> 00:01:02.719
So thank you so much for being here today with us.
00:01:02.799 --> 00:01:04.400
I'm so excited to talk about this topic with you.
00:01:04.640 --> 00:01:06.560
Yeah, thank you so much for having me, Katarina.
00:01:06.640 --> 00:01:08.079
It's uh exciting to be here.
00:01:08.560 --> 00:01:09.280
It's so exciting.
00:01:09.439 --> 00:01:12.000
Okay, so let's start with a really basic question.
00:01:12.239 --> 00:01:14.319
You are a political philosopher.
00:01:14.560 --> 00:01:20.159
Why do we need politics and philosophy in order to understand what is happening with AI today?
00:01:20.400 --> 00:01:22.400
Well, uh, there's political scientists, right?
00:01:22.480 --> 00:01:31.439
And like these people basically look at what's happening in politics and the structures around everyday politics, and they analyze how that works.
00:01:31.519 --> 00:01:34.319
So they kind of look at the world as it is.
00:01:34.480 --> 00:01:37.840
But political philosophers ask a slightly different question.
00:01:38.000 --> 00:01:42.400
So we're interested in this question of how should politics work?
00:01:42.719 --> 00:01:46.560
So it's a normative rather than empirical question.
00:01:46.719 --> 00:01:52.239
We're interested in the kinds of goals and values that should guide our collective decisions.
00:01:52.480 --> 00:02:06.400
And so this has become incredibly important in the AI space because all of a sudden we see these tech CEOs behaving as though they're elected officials, when in fact, as you pointed out before, they're really not.
00:02:06.719 --> 00:02:07.200
100%.
00:02:07.519 --> 00:02:11.039
No one voted for them, and then suddenly they have so much power.
00:02:11.199 --> 00:02:19.120
And what are the problems that that could lead us as a society to have just a handful of powerful people in Silicon Valley deciding for our lives?
00:02:19.520 --> 00:02:21.280
Well, I think there's multiple issues.
00:02:21.520 --> 00:02:26.719
One issue is like any concentration of power and wealth is always dangerous for a democracy.
00:02:27.039 --> 00:02:48.560
So we ideally don't want to get ourselves in a situation where just a handful of people andor companies concentrate like all of this money and all of this political influence in one part of society, but at the same time, they're actually able to have a really widespread distribution of all sorts of risks associated with AI.
00:02:48.719 --> 00:02:58.319
So we get this like really crazy asymmetry where, on the one hand, we concentrate the benefits, but we collectivize all of the risks and costs of AI.
00:02:58.479 --> 00:03:03.199
And so that stands in need of democratic justification, at least in my mind.
00:03:03.360 --> 00:03:13.199
So we need to ask, are we actually comfortable as a collective with that kind of asymmetrical distribution, or do we need to change some things?
00:03:13.360 --> 00:03:17.439
And so this is where democracy uh becomes relevant in this space.
00:03:18.319 --> 00:03:19.199
And I love what you're saying.
00:03:19.360 --> 00:03:20.479
Do we need to change or not?
00:03:20.560 --> 00:03:29.840
I could say that most of the people who follow my work and are into critical AI, they say, yes, we need to change this, but there's something that you discuss here, the costs and the risks.
00:03:30.000 --> 00:03:33.919
Could you tell us a little bit more about what are these costs and what are these risks?
00:03:34.319 --> 00:03:47.280
Yeah, so there's a cluster of more well-understood costs and risks that have already surfaced, that many experts in the field, many of whom you've already spoken to, and many of your uh audience members will know already.
00:03:47.439 --> 00:03:49.360
They've already investigated a lot of this.
00:03:49.520 --> 00:04:02.000
So racial bias, gender bias, machine hallucinations, like vulnerable users getting drawn into like very damaging behavior, mass uh you know, polarization and propaganda enabled by AI applications.
00:04:02.159 --> 00:04:09.680
So there's like a pretty broad cluster of very urgent kind of present-day risks that we actually already know about.
00:04:10.000 --> 00:04:17.519
Then on the other hand, there are some more mysterious future-oriented risks that are much more contentious.
00:04:17.759 --> 00:04:23.920
Audience members of yours might have encountered like recent debates about will AI kill all of us, right?
00:04:24.000 --> 00:04:26.480
Like, does AI pose an existential risk?
00:04:26.639 --> 00:04:40.079
And so this is where a lot of scholars actually say no, we need to be very cautious not to give people this seemingly confident forecast about there's like a 10% chance we're all gonna die in a few years.
00:04:40.319 --> 00:04:44.319
Because it's very hard to actually make that kind of forecast in a rigorous way.
00:04:44.480 --> 00:05:01.199
But at the same time, AI companies are very much emphasizing these ostensible risks because, of course, you know, it adds a lot of hype and mystique around their products if there's a discourse that says, look, like this is really powerful, but also really, really dangerous.
00:05:01.360 --> 00:05:11.360
So there's actually also been a critical backlash against that recent discourse that says, hey, listen, like, are your motives really just to expose this uh long-term risk?
00:05:11.680 --> 00:05:20.160
Or do maybe other political, economic, and societal factors play into the ways in which we want to forecast the future?
00:05:20.319 --> 00:05:25.920
And so the story might be a little bit more complex than just putting a percentage point on some sort of forecast.
00:05:26.240 --> 00:05:27.279
I think that's really interesting.
00:05:27.360 --> 00:05:35.199
And we're going to talk more about the ideologies that are kind of controlling the narratives that get said in AI today, exactly what you said.
00:05:35.360 --> 00:05:39.040
We have a handful of people saying that artificial intelligence will destroy the world.
00:05:39.199 --> 00:05:41.199
So we're going to leave that for the second part.
00:05:41.439 --> 00:05:48.000
But now I think that at the same time that you know these tech CEOs they're telling that artificial intelligence is really dangerous.
00:05:48.160 --> 00:06:06.240
They still want to double down on the idea that artificial intelligence is marvelous for us, that we should love artificial intelligence because it democratizes access to education or to the arts, and you know, because everyone has a phone today and you can have touch PT Claude in your pocket and you can do all of these things.
00:06:06.399 --> 00:06:12.959
So basically, trying to say that, hey, you should love this technology because everyone has access to this and that they're democratizing knowledge.
00:06:13.120 --> 00:06:16.000
Do you agree with this type of commentary?
00:06:16.240 --> 00:06:20.079
Yeah, nice job on clocking this important narrative, Katarina.
00:06:20.160 --> 00:06:40.000
I think this is really key to understanding what's going on in this debate, because uh this has become a really central argument pushed by AI industry that, like, look, there might be some sort of asymmetrical distribution, but at the end of the day, please don't complain, regular citizens, because like this is all good for you, like you want to have an easier life.
00:06:40.160 --> 00:06:44.160
And so now you don't need to be writing like boring emails all on your own.
00:06:44.319 --> 00:06:47.279
Like, you know, you get this like slight convenience increase.
00:06:47.439 --> 00:06:55.839
And so even if there's material inequality in society at like vast degrees, you should accept that because you're getting some sort of benefit.
00:06:56.079 --> 00:07:00.720
And so political philosophers have been very suspicious of this kind of argument.
00:07:00.879 --> 00:07:08.399
In our field, we have developed theories that show why these kinds of distributions are highly morally and politically suspect.
00:07:08.560 --> 00:07:24.240
Uh, political philosophers often talk about so-called win-win exploitation, where some exploiter gets like a huge benefit, and some other party, the exploited one, might still get like some benefit, but there's still an unfair transaction between these parties.
00:07:24.480 --> 00:07:31.759
And so not all moral problems disappear just because some sort of benefits are part of the story.
00:07:32.000 --> 00:07:40.319
It matters how we distribute them and who's actually able to call the shots about how those benefits go around.
00:07:40.560 --> 00:07:50.480
And so what we see in AI, we just have a lot of unilateral decision-making power in this very, very small handful of corporations.
00:07:50.639 --> 00:08:04.639
So we have an oligopoly in this market, and so this oligopoly is able to uh decide pretty much on their own in many cases when to deploy a tool, how soon to deploy it, under which conditions, and so on.
00:08:04.800 --> 00:08:06.639
And so that's really the core of the problem.
00:08:06.800 --> 00:08:10.319
We shouldn't get sidetracked with this benefits talk.
00:08:10.639 --> 00:08:14.399
Could you explain to us a little bit more about what do you mean by oligopoly?
00:08:14.720 --> 00:08:19.839
So this basically just means that there is a highly concentrated market.
00:08:20.079 --> 00:08:25.759
In a super free market, you have like many different companies and they're in competition with each other.
00:08:25.920 --> 00:08:32.000
And actually, interestingly, this idea of a free market kind of sounds like a libertarian or a conservative idea.
00:08:32.240 --> 00:08:46.159
But even like democratic socialists on the harder left are fans of you know having a more robust market in a democratic society in which we don't have a lot of powerful big fishes that can call the shots on everything.
00:08:46.480 --> 00:09:03.840
So, like uh under uh US President Biden, for instance, the FTC under uh Lena Khan started, you know, trying to break up big tech again after basically 12 years of not really doing all that much to curtail the power of corporations in this uh tech space in the US.
00:09:04.000 --> 00:09:14.240
And so briefly that actually changed because both progressives and conservatives recognized we actually have a problem if we only have a few companies calling the shots and amassing more and more power.
00:09:14.399 --> 00:09:24.159
So as soon as we move away from the free market, we either get an oligopoly, so like four or five companies, or even a monopoly, just like one company that dominates an entire sector.
00:09:24.320 --> 00:09:32.159
And so arguably both of these are not just bad for regular consumers, but they're actually bad for democracy.
00:09:32.399 --> 00:09:39.840
Because as soon as I have this kind of power concentration, it's not just anymore about, oh, this company can dictate prices.
00:09:40.000 --> 00:09:46.559
It's actually also about this company might be influencing all of our political decisions in some way.
00:09:46.720 --> 00:09:50.480
And so the problem actually goes much deeper than just economic problems.
00:09:50.799 --> 00:10:00.480
I think this is so fascinating because I do think that many times people think that just because, you know, open AI and then throw up, they're creating AI, that this is a tech problem.
00:10:00.639 --> 00:10:10.000
But we need to understand that when we have maybe five companies deciding artificial intelligence and technology that is impacting everyone, we're talking about politics here.
00:10:10.080 --> 00:10:15.200
We're talking about the agendas of what gets said, what doesn't get said, the things that are decided.
00:10:15.279 --> 00:10:22.720
And again, they're not decided by us, but only five companies or five people who are really, really wealthy and deciding these agendas.
00:10:22.879 --> 00:10:34.320
And I think one way that they were able to create this agenda setting is that somehow today, the artificial intelligence industry, they're considered different from anyone else.
00:10:34.480 --> 00:10:48.559
So for example, when we think about big pharma or if you want to create a new drug, what happens is that, you know, you come up with the idea, you do research, you do studies, and then you go to the standard body and you say, hey, I want to create this new drug, it's going to help human beings.
00:10:48.720 --> 00:10:52.799
The standard body says, okay, you did the test, or no, you didn't, push it to the market.
00:10:52.960 --> 00:11:09.039
Another example, if I want to sell chicken pie in my streets here in Sao Paulo, I need to take pictures of my kitchen, I need to, you know, say what is my credentials, put everything in a document, go to the food standard bodies and say, hey, I want to create this, I want to sell chicken pie, and they will say, fine.
00:11:09.200 --> 00:11:16.000
But then suddenly here we're talking about five people or five companies that are doing everything backwards.
00:11:16.240 --> 00:11:23.200
They say, I'm going to create this technology and I'm just going to push it in the market, and you, everyone, fend for yourself with the consequences.
00:11:23.360 --> 00:11:38.559
And that is partly because they have so much power, and also partly because they created this narrative that they are different and that when we think about AI innovation, that we shouldn't be playing the same rules of the game that basically all other industries.
00:11:38.720 --> 00:11:40.399
Could you tell us a little bit why?
00:11:40.559 --> 00:11:47.519
Not only they have so much power, but also why can they just play the game that everyone is playing, but with different rules?
00:11:47.759 --> 00:11:52.399
Yeah, that's a brilliant summary of a lot of my core uh concerns in this space.
00:11:52.559 --> 00:11:52.960
Thank you.
00:11:53.120 --> 00:12:02.399
I think, as you correctly pointed out, uh, the power to deploy a powerful new technology is itself a power to shift a democratic agenda.
00:12:02.559 --> 00:12:12.000
And so if I can totally change everyone's life by just saying, okay, hey, there's this new tool that I generously now give to you.
00:12:12.080 --> 00:12:13.759
And by the way, you can use it for free.
00:12:13.840 --> 00:12:22.639
So like please all start using it, and like that, of course, will help me optimize my tool so that I can come up with like some freemium model and like make a ton of money with that, right?
00:12:22.879 --> 00:12:40.799
So this is all part of a deployment strategy, of course, but it also immediately forces a political response because, of course, immediately people start wondering well, you know, now that this new thing happened, what should we perhaps regulate it, even though we might not have a good regulatory basis in place?
00:12:40.960 --> 00:12:46.159
And so this happened actually very early on in the space, even before the generative AI hype cycle.
00:12:46.320 --> 00:12:53.840
So if you recall, the predecessor model of Chat GPT, this was called GPT-2 at the time.
00:12:54.080 --> 00:13:08.799
And so back then, OpenAI was going around putting out all of these press releases saying, like, oh, we can't possibly deploy this out in the open, like no regular citizens can get their hands on it because it's way too dangerous, it'll kill us all, right?
00:13:08.879 --> 00:13:10.399
Like we've we've heard this before.
00:13:10.480 --> 00:13:13.600
So this is like an older narrative, many people don't realize.
00:13:13.840 --> 00:13:23.759
And so they, of course, completely reverted course as soon as they were ready to deploy a direct-to-consumer product in the form of Chat GPT.
00:13:23.840 --> 00:13:33.200
And so, of course, back then that triggered that generative AI deployment rush with all of these other companies that also wanted to go ahead and deploy their own tools.
00:13:33.440 --> 00:13:43.840
Now, on the side of politics, this triggered a very interesting type of discourse that in my book, Democratizing AI, I call AI exceptionalism.
00:13:44.000 --> 00:13:56.480
It's the idea that, as you just mentioned before, we should treat AI as a as a domain and as an industry as fundamentally mysterious and unique and unlike any other industry that we've ever seen before.
00:13:56.639 --> 00:14:06.559
And so when people buy into this narrative, it unfortunately makes it more likely that they'll actually also think about AI regulation in this AI exceptionalist way.
00:14:06.720 --> 00:14:17.279
And so we see this play out all the time when we see big tech CEOs going to like, you know, Senate or the Congress and saying things like, well, you know, it's just a black box.
00:14:17.440 --> 00:14:22.960
Like we need real deep technical expertise before we can even consider how to regulate that.
00:14:23.200 --> 00:14:31.360
And so, of course, we don't think about other highly complex domains of science and technology policy in the same way.
00:14:31.600 --> 00:14:35.440
I always think about climate change mitigation policies.
00:14:35.679 --> 00:14:40.559
So I don't have a meteorology PhD, nor do I have a physics PhD.
00:14:40.720 --> 00:14:47.440
But I have particular political preferences about how important it is, I think, to try to mitigate climate change.
00:14:47.519 --> 00:14:56.639
And so, as a regular citizen with no technical expertise, I can actually contribute to those policy debates by saying, like, I think this should be a key priority for us.
00:14:56.799 --> 00:15:00.320
I really want us to mitigate the threats associated with climate change.
00:15:00.480 --> 00:15:09.679
And so, very similarly, I think regular citizens can articulate at least some value-based preferences about what should happen with AI.
00:15:09.840 --> 00:15:23.200
For instance, if you ask people, hey, are you okay with mass surveillance powered by AI, ideally powered by a company that also has a really deep enmeshment with an increasingly authoritarian government?
00:15:23.440 --> 00:15:29.039
Chances are many people might say, actually, no, that doesn't align with my political preferences.
00:15:29.200 --> 00:15:34.559
And so, boom, all of a sudden, we're having a very non-exceptionalist discourse on AI policy and governments.
00:15:34.799 --> 00:15:35.759
It's very much possible.
00:15:35.840 --> 00:15:38.000
And I think that's actually where we need to get to.
00:15:38.159 --> 00:15:47.120
So I want to get more people who don't have a computer science PhD to really start asking themselves, like, what do I want our future to look like?
00:15:47.279 --> 00:15:52.320
And what role does AI play in that future or doesn't it play, perhaps?
00:15:52.559 --> 00:16:03.039
I think that's also really interesting because in getting people to understand how they could impact the narrative, they also need to understand how artificial intelligence is impacting them day to day, right?
00:16:03.200 --> 00:16:19.200
I have many followers that say, hey, I don't want to learn about AI, I hate AI, I don't want to be part of this conversation, or that somehow just because they dislike it, artificial intelligence is not impacting their lives, not only in a technical space, also in an ideological place.
00:16:19.440 --> 00:16:44.879
So I would like to talk about how having this oligopoly and having these tech CEOs creating these products and pushing them in the market, they also are able to do something that everyone needs to understand, which is setting up political agendas or the agendas of what gets spoken about, what doesn't get spoken about, the things that we're deciding in terms of education or things about climate change as well.
00:16:45.120 --> 00:16:50.320
Just by pushing products in the market, they're also ideologically shaping that agenda.
00:16:50.559 --> 00:17:00.080
Could you tell us a little bit more about that and also how that ends up impacting people's lives, not only in a technical sense, but in a living day-by-day sense?
00:17:00.399 --> 00:17:07.359
The most concerning thing that I think happens in this regard is a narrowing of our imagination of what's possible.
00:17:07.519 --> 00:17:19.519
And so normally when somebody doesn't mess with our collective agenda setting process in a democracy, that means that we're having a lot of debates in society, not just in the context of elections, by the way, right?
00:17:19.599 --> 00:17:25.039
So like democracy is something bigger than just like, oh, you go and like cast your vote every four years.
00:17:25.200 --> 00:17:29.039
That's like a really thin notion of democracy that I want to overcome.
00:17:29.359 --> 00:17:46.000
I think democracy happens daily when you talk to people, ideally also people that you disagree with, and you form opinions, you deliberate about different options, and ideally you come to a fuller understanding of what you and your neighbors and your co-citizens think is desirable and possible.
00:17:46.240 --> 00:17:58.880
But of course, if you have this massive concentration of power, this can lead to some policy preferences and some arguments gaining like a lot of traction also in the media and in the public imagination.
00:17:59.119 --> 00:18:08.559
This actually gets exacerbated by the fact that some tech CEOs are also doing a lot of very aggressive media company acquisitions.
00:18:08.720 --> 00:18:20.400
So Larry Allison comes to mind, a figure who now has really, really far-reaching control over a vast media ecosystem right before the midterm elections.
00:18:20.640 --> 00:18:24.240
And also, of course, this has global reach far beyond the United States.
00:18:24.480 --> 00:18:35.039
And so these are tech CEOs that don't just see themselves as executives, they see themselves as political activists who actively shape these narratives and these options.
00:18:35.279 --> 00:18:42.799
And so two examples of potentially misleading or artificially narrow options that I've seen recently are the following.
00:18:43.039 --> 00:18:52.400
One is this idea that comes from OpenAI, but also from Palantir that says, look, we're in an arms race with China and maybe other uh foreign adversaries.
00:18:52.559 --> 00:19:06.880
And so, therefore, if you want democratic AI, you gotta back American companies because, of course, these American companies are operating in uh ostensibly democratic structures, while uh, you know, competitor companies abroad are not doing that.
00:19:07.039 --> 00:19:13.440
And so this argument often gets fused with a bid for lesser regulation in the tech space.
00:19:13.599 --> 00:19:25.440
I noticed that, you know, Alex Carp, the Palantir CEO, wrote a book justifying this idea, also justifying the idea that actually we shouldn't separate corporations from government.
00:19:25.599 --> 00:19:27.599
Actually, we should have a deep enmeshment.
00:19:27.759 --> 00:19:30.559
We should welcome the idea of a military-industrial complex.
00:19:30.720 --> 00:19:40.559
And so, you know, for obvious reasons, I think this can quickly be a very narrow type of view that uh, you know, suggests to people that they only have two choices, right?
00:19:40.640 --> 00:19:43.519
When actually more choices might be available to them.
00:19:43.599 --> 00:19:51.680
Like maybe it's not just, you know, either we all die under authoritarian dictatorship or we we regulate some tech companies.
00:19:51.759 --> 00:19:54.160
You know, like maybe these are not the only two options.
00:19:54.400 --> 00:19:58.720
I also was very interested to see another type of narrative emerge.
00:19:58.880 --> 00:20:00.160
This is over the past few years.
00:20:00.319 --> 00:20:06.400
I noticed that Sam Altman went to a Senate hearing that I covered very extensively in my research.
00:20:06.480 --> 00:20:10.799
I kind of did a deep dive on this particular hearing in 2023.
00:20:11.039 --> 00:20:18.160
And I noticed that when Altman was asked by lawmakers, hey, how should we actually regulate this technology?
00:20:18.319 --> 00:20:21.599
They were kind of saying, like, oh, we're clueless, like, we don't even know where to start.
00:20:21.839 --> 00:20:23.279
He actually came prepared.
00:20:23.440 --> 00:20:27.359
So he said, Look, I have an idea, I think you should do it.
00:20:27.599 --> 00:20:35.039
And so while he was outlining this idea, he was also asked directly by these lawmakers, so who should implement this idea?
00:20:35.200 --> 00:20:38.000
Who should oversee the governance of AI in America?
00:20:38.160 --> 00:20:44.799
And so he just laughs and he just says, Well, we can send your office a list of the people we think are qualified.
00:20:44.960 --> 00:20:51.440
And so he gets this response in the hearing of people on both sides of the aisle saying, Well, thank you.
00:20:51.599 --> 00:20:53.599
We really want your recommendations.
00:20:53.759 --> 00:21:04.079
And so I think this is very indicative of an artificially narrow option space where all of these other possible voices aren't really part of the conversation anymore.
00:21:04.240 --> 00:21:11.839
And all of these other ideas that they might have about how to regulate AI aren't really heard in this discourse either.
00:21:12.000 --> 00:21:19.839
And so it was no surprise then that at that hearing, Altman floated this idea of, you know, you don't need to regulate all that much.
00:21:20.000 --> 00:21:22.480
You just need to implement a licensing regime.
00:21:22.640 --> 00:21:32.799
The idea that in order to keep deploying AI and to reject democratic agendas, you should just be a big company that is able to apply for a license.
00:21:32.960 --> 00:21:34.880
And so the government oversees that.
00:21:34.960 --> 00:21:38.640
And so, you know, this handful of companies can then get that license.
00:21:38.880 --> 00:21:54.319
That sounds kind of responsible and nice, but what he didn't say was that actually only a small number of corporate actors would be able to apply for that kind of license successfully under the conditions that he had articulated.
00:21:54.559 --> 00:21:59.680
And so it would have actually solidified the immense market power that layers.
00:22:00.079 --> 00:22:09.599
Like OpenAI and Anthropic have, and artificially keep out any new competition, like any new startup, any small research team, any foreign competitor, anything.
00:22:09.839 --> 00:22:20.480
And so this is a very sneaky narrative because it sounds really ethical, but it actually really misleads us about what else might be possible.
00:22:20.799 --> 00:22:24.319
So this is what I have in mind when I think about AI exceptionalism.
00:22:24.720 --> 00:22:43.839
Whenever, and and for people who are listening this episode, I actually, we already talked before, and when you told me about the concept of AI exceptionalism, I started seeing this everywhere, everywhere in the narratives, everywhere around me, how artificial intelligence is seen completely different, how AI CEOs are seen in a completely different way.
00:22:44.000 --> 00:22:46.480
And it's something that I started sharing with my friends.
00:22:46.640 --> 00:22:54.160
It's, you know, we're not asking the sugar CEOs if they recommend people eating sugar or not, because we know that they are biased.
00:22:54.240 --> 00:22:58.480
We know that whatever they say will be fully biased in order for us to eat more sugar.
00:22:58.559 --> 00:23:03.680
But then suddenly, when it comes to AI CEOs, they are invited into Congress hearings.
00:23:03.759 --> 00:23:20.559
They're invited in these interviews to tell us what we should be thinking about in terms of regulating them, which seems to me that their agenda in big terms is also bringing this sort of neutrality in how they benefit from this discourse.
00:23:20.799 --> 00:23:34.559
So that narrow way of thinking also comes a lot with their ideologies and what they see is right, what they see is wrong, and what they see as to what our society should be striving for.
00:23:34.720 --> 00:23:38.799
So I think that this sort of conversation, we need to start talking about ideologies.
00:23:38.960 --> 00:23:42.799
We can't just talk about this without understanding what they actually believe in.
00:23:42.960 --> 00:23:47.839
So could it be possible for us to start talking about what are these ideologies that they're following?
00:23:48.079 --> 00:23:49.279
What is going around?
00:23:49.359 --> 00:23:54.640
And I know that effective altruism comes to play here, also effective accelerationism.
00:23:54.799 --> 00:24:00.079
Could you tell us a little bit more about that and how that gets implemented into the narrow narratives that we see today?
00:24:00.400 --> 00:24:12.960
Yeah, I started getting really interested in this question because uh, you know, I'm a philosophy professor and I started noticing that a bunch of these tech CEOs are suddenly saying, oh, you know, like I wrote a philosophical manifesto.
00:24:13.119 --> 00:24:16.400
That's usually an indicator that something has gone terribly wrong.
00:24:16.559 --> 00:24:19.200
Uh it's not easy to like just do philosophy.
00:24:19.359 --> 00:24:27.440
And so surprise, surprise, a lot of this stuff isn't, well, let's put it this way, it wouldn't pass uh in a philosophy class.
00:24:27.519 --> 00:24:34.960
Uh, because there's often actually many logical flaws in uh in these manifestos and treatises that we get from uh CEOs.
00:24:35.119 --> 00:24:41.599
I think it's also important to not put all of these pseudo-philosophical outputs in just one bucket.
00:24:41.759 --> 00:24:45.759
So it's not the case that there is one Silicon Valley ideology.
00:24:46.000 --> 00:24:50.319
They have actually splintered out into all of these different and weird camps.
00:24:50.480 --> 00:24:58.319
We started out earlier, like the older Silicon Valley that is still, you know, very much present in people's imagination and in the media.
00:24:58.559 --> 00:25:10.400
That is a kind of simpler story where nerds in Silicon Valley used to be kind of economically libertarian, like they like to tinker, they like to like, you know, deploy a cool product in the market and like get massively rich with that.
00:25:10.559 --> 00:25:16.240
But they also used to have this brand of being kind of socially liberal, at least in a kind of superficial sense, right?
00:25:16.400 --> 00:25:19.279
But so things have really shifted in recent times.
00:25:19.440 --> 00:25:26.799
Uh, you already mentioned effective accelerationism and effective altruism as two really important movements in this space.
00:25:26.960 --> 00:25:37.039
Uh, to that, let's add a third, and that is right-wing libertarianism, which is increasingly turning into dark enlightenment kind of alt-right authoritarianism.
00:25:37.200 --> 00:25:41.599
So that's a kind of different ideological trajectory that that works in parallel there.
00:25:41.759 --> 00:25:46.640
Let's maybe start with that last bit because I think it's it's increasingly gaining a lot of traction.
00:25:46.799 --> 00:25:58.400
So I'm thinking in particular of people like Peter Thiel here, who even like back in the day, like decades ago, would write these essays about how you know freedom and democracy are no longer compatible.
00:25:58.480 --> 00:26:04.880
And so if you really care about capitalism, then uh, you know, you gotta toss democracy to the side.
00:26:05.119 --> 00:26:14.240
Because ever since states decided to enfranchise women and poor people, that really messed things up for the capitalists, and so democracy stopped working, according to him.
00:26:14.480 --> 00:26:19.519
He later apologized for making that claim in the essay, but it was a pretty lukewarm apology.
00:26:19.680 --> 00:26:25.039
And actually, he seems to have doubled down on this democracy skepticism since.
00:26:25.119 --> 00:26:45.519
And instead, writing further philosophical essays in which he talks about the decline of the Western world and threats from mysterious foreign places against that Western world, that coincides with a more general political ideological drift to the hard right that has recently happened, especially over the past decade.
00:26:45.680 --> 00:26:52.319
And so one of Peter Thiel's mentees, Curtis Jarvin, is the figurehead of part of this movement.
00:26:52.559 --> 00:27:02.240
Curtis Jarvin is a blogger who is known for publicizing explicitly racist views in his writings.
00:27:02.400 --> 00:27:09.759
He's also known for articulating the view that actually we should abolish democracy, we should be run by a CEO monarchy.
00:27:09.920 --> 00:27:15.119
So the democracy skepticism actually is really key to understanding this whole ideology.
00:27:15.359 --> 00:27:27.440
There's a whole bunch of other kind of socially conservative influences in this ideology too, but it would be a mistake to think of them just as, oh, these are Republicans or social conservatives.
00:27:27.680 --> 00:27:41.119
These are people who go much further to the right and who basically think that the Enlightenment, the idea of giving people equal rights and like having a transparent discourse in society, that was a big mistake.
00:27:41.200 --> 00:27:43.839
And so that's why it's called Doric Enlightenment ideology.
00:27:43.920 --> 00:27:48.640
So they want to backtrack on all of these ideas of human rights, civil rights, and so on.
00:27:48.799 --> 00:27:53.920
Let's pivot a little bit to the other two twin theories that you mentioned before.
00:27:54.160 --> 00:28:03.359
So we have effective altruism, which started out as a kind of niche philosophical view that was originally much more concerned about like non-AI issues.
00:28:03.440 --> 00:28:15.839
So like back in the day, 10, 15 years ago, this was a group of people who were committed to the moral theory of consequentialism, the idea that the morally right thing to do is to maximize good outcomes.
00:28:16.000 --> 00:28:24.079
And so with this logic of maximizing good outcomes, they started thinking about well, how can we do good things in a better way?
00:28:24.240 --> 00:28:42.319
But so as a community, at some point recently-ish, they decided actually, if we want to maximize good outcomes, then we should be very concerned about a potential scenario in which AI could kill us all, because blocking that doomsday scenario is then absolutely important.
00:28:42.480 --> 00:28:45.440
It kind of trumps all of these other nice charity considerations.
00:28:45.599 --> 00:28:53.119
And so this movement really pivoted to this idea of being singularly focused on this one particular threat model.
00:28:53.359 --> 00:29:03.359
At the same time, the movement has been criticized by, first of all, being led by people who all tend to fit the same kind of narrow sociodemographic criteria.
00:29:03.440 --> 00:29:08.400
So it doesn't seem to be a very diverse movement, especially when we look at the leading figures.
00:29:08.640 --> 00:29:18.160
And this movement tends to de-prioritize a lot of issues that people in fact care about right here and right now, not even just thinking about doomsday scenarios.
00:29:18.319 --> 00:29:22.079
So, like right now, society has racism and sexism.
00:29:22.319 --> 00:29:25.599
Right now we have massive political division and unrest.
00:29:25.680 --> 00:29:31.119
And like there's so many real-world problems, we don't have to wait until this doomsday scenario to fix them.
00:29:31.279 --> 00:29:38.240
It seems like effective altruists don't really want to pivot their attention to these shorter term considerations.
00:29:38.559 --> 00:29:47.680
We can define the last one of those three super quickly, effective accelerationism, is just the backlash movement against that, uh, against effective altruism.
00:29:47.839 --> 00:30:04.240
And so this happened because a lot of people in the valley started getting really annoyed with all of this AI doomer discourse, and they started saying, no, that's instead of stopping AI, we actually need to pump the gas and just like really accelerate and keep deploying way more AI.
00:30:04.400 --> 00:30:06.319
And that's the thing that's gonna save humanity.
00:30:06.559 --> 00:30:16.319
And so these people, for instance, represented by venture capitalist Mark Andreessen, who wrote the so-called techno-optimist manifesto, they just want to push forward, right?
00:30:16.480 --> 00:30:24.240
So they their kind of a salvation narrative is something like well, the best possible AI is the thing that makes all of our lives better.
00:30:24.480 --> 00:30:43.359
What's interesting about this movement is that they too don't seem to be particularly diverse, and they also seem to center this idea of a few very, very smart people being able to save everyone if we just stop regulating them and annoying them so much.
00:30:43.519 --> 00:30:50.799
And so this idea of the hero genius who just needs to be let loose looms really large in this movement too.
00:30:51.119 --> 00:31:01.200
And why do you think that everyday people need to understand these ideologies to understand artificial intelligence today and how our democracy is being impacted by them?
00:31:01.519 --> 00:31:09.200
I think it's really important to be able to clock a narrative that might not be the only analysis of what's actually happening to us right now.
00:31:09.440 --> 00:31:24.799
So when a lot of people hear AI might kill us all, uh they understandably start getting upset and scared because here we have so-called experts telling us, oh, you know, like we might all die unless we follow this movement or like this preferred policy solution.
00:31:24.960 --> 00:31:34.319
And so often things are presented as facts when they're actually just more like opinions or at least kind of contested predictions about the future.
00:31:34.559 --> 00:31:51.519
Here it's important to notice that it's very hard to make very rigorous, scientifically defensible predictions of the future, especially when we look at the kinds of doomsday forecasts that we get from some people who are behind these ideological movements.
00:31:51.759 --> 00:32:00.720
So very often, an argument that sounds a lot like it's inevitable that X will happen is actually just one possibility.
00:32:00.880 --> 00:32:22.240
And so it's important for us as normal citizens and also as members of the academic community, as activists, as members of nonprofits, and so on, to think about okay, what are some alternative future paths that we want to pay attention to in order to contest this narrative that some specific course of events is actually inevitable.
00:32:22.480 --> 00:32:30.799
And so that's why it's really important not to just let policymakers try to ask themselves these questions, because this actually affects all of us, right?
00:32:30.880 --> 00:32:42.559
Like it's gonna affect my decision about whether I support AI-powered surveillance, if I think that I have to accept AI-powered surveillance because otherwise the free world will die, right?
00:32:42.640 --> 00:32:55.359
Like if I just accept that meta-narrative, then a whole bunch of my other opinions that somehow intersect with AI will be totally distorted by whatever these tech CEOs are telling us.
00:32:55.519 --> 00:33:06.640
And so being able to identify that and to say, oh, actually, that sounds a lot more like an opinion or a preference rather than an unshakable fact is really important in order to be an empowered citizen.
00:33:07.039 --> 00:33:22.000
It seems like understanding and clocking out and not trusting AI CEOs at face value is one of the ways that we're able to democratize AI in a way that I understand how this impacts my life, and I can clock that out because that's an ideological assertion.
00:33:22.240 --> 00:33:26.000
I might not need that, not want that, and I think that there's a different way for that.
00:33:26.079 --> 00:33:27.279
So maybe that is one.
00:33:27.440 --> 00:33:36.079
But are there more ways that we as a society, we everyday people, can push for this democratization of AI that clearly is not happening today?
00:33:36.400 --> 00:33:47.680
I think one very interesting opportunity that can be important in some cases, at least, is to actually say, hey, we don't actually need AI for this particular task.
00:33:47.920 --> 00:33:58.160
And so this is an argument that often gets vilified right now by the media and by tech industry, and it gets called something like AI populism by some.
00:33:58.240 --> 00:34:16.000
So there have been a number of recent think pieces that, for example, commented on the fact that like young people at graduation ceremonies started booing Eric Schmidt and other speakers who said, Oh, you know, I know you guys are worried about never getting a job, but you just need to adapt to AI and like this is the future.
00:34:16.159 --> 00:34:17.280
And so people weren't having it.
00:34:17.360 --> 00:34:19.199
Like people were really upset about that.
00:34:19.360 --> 00:34:38.960
And I think it's important not to dismiss that as just populism, but just because it happens to be a negative opinion about AI in domains like education or like in algorithmic hiring, which plenty of studies have shown to be very, very biased against women and people of color.
00:34:39.199 --> 00:34:40.079
And so people know this.
00:34:40.159 --> 00:34:42.239
People have good reasons to be very concerned about this.
00:34:42.320 --> 00:34:48.880
Like upon graduation, they'll be interviewed by some AI interviewer, they'll be unfairly assessed in many cases.
00:34:49.119 --> 00:34:55.519
Nobody keeps these systems in check, and so they'll just not get a fair shake after spending years and like a lot of money on their education.
00:34:55.760 --> 00:35:02.159
And so it's it's really key to allow that as like part of the democratic process.
00:35:02.320 --> 00:35:10.159
It can't just be that we deliberate about, oh, we should regulate AI this way or that way, but we have to accept it nonetheless.
00:35:10.320 --> 00:35:11.840
So this is like one important thing.
00:35:12.079 --> 00:35:17.599
Then I think the question is well, how do we actually translate this into collective decision making?
00:35:17.760 --> 00:35:21.599
And so here it really depends what kind of constituency we're talking about.
00:35:21.760 --> 00:35:27.360
I've seen some very positive examples of very local decision making about AI.
00:35:27.440 --> 00:35:33.360
So like often we kind of think, oh, democracy is just about like state-level elections and so on.
00:35:33.519 --> 00:35:35.679
But actually, democracy can happen in the city.
00:35:35.840 --> 00:35:46.239
Democracy can happen in like a federal state, democracy can even happen transnationally when different local groups decide to push for some agenda and they inspire each other.
00:35:46.480 --> 00:35:52.239
One concrete example I can think of is actually even before this generative AI wave.
00:35:52.400 --> 00:35:58.639
And that's the public debate in some cities on facial recognition technologies used in policing.
00:35:58.880 --> 00:36:23.840
I remember that in 2019, in San Francisco, no less, right in the middle of the valley, people locally decided, and this was very much led by local activists working on criminal justice issues, not just technical experts, they decided, hey, actually, we're very skeptical of using facial recognition tools for police surveillance because we're concerned about racial bias in this context.
00:36:24.000 --> 00:36:31.519
At that time, plenty of academic experts had already mapped many important risks for the population and their civil rights.
00:36:31.679 --> 00:36:34.079
And so they were able to successfully push this through.
00:36:34.159 --> 00:36:54.400
In 2019, San Francisco passed the stop secret surveillance ordinance, and that actually triggered this fascinating snowball effect across the globe, where like multiple American cities, but also multiple cities in other countries decided to also impose a local deployment moratorium until these risks can be better understood.
00:36:54.559 --> 00:36:58.719
And many of them have actually upheld that block, some of them have reverted it.
00:36:58.880 --> 00:37:18.800
But nonetheless, we see there is like a lot of space for bottom-up grassroots activism, even if it doesn't end up being this like national movement, it can be extremely transformative for local populations, and it can have this really inspirational effect across like different constituencies.
00:37:18.960 --> 00:37:33.440
And so I think we should look at that example and think about well, where else can normal citizens who see a problem in their city, in their local community, where can they push for a decision that makes a small difference?
00:37:33.679 --> 00:37:40.239
And so this can be the beginning point of a much larger push for a positive change.
00:37:40.480 --> 00:37:48.320
And so, of course, at the same time, it's really important to have like legitimate, democratically legitimate, top-down regulation, right?
00:37:48.400 --> 00:37:55.039
Where we get our elected officials to make some better choices that don't just accept an industry narrative.
00:37:55.199 --> 00:38:04.000
And so we do need like legal change, we need like good regulatory frameworks, but that can't be all that there is to democratizing AI.
00:38:04.159 --> 00:38:13.119
We also need these spaces for more flexible, more bottom-up kinds of activism inspired by these positive use cases that we've already seen.
00:38:13.360 --> 00:38:44.079
So I guess one message that I'd really want people to remember from this conversation is that even though it seems that democracy is really messy and like really hard and can often go wrong, and like it might also seem like it's impossible to do anything against these powerful corporations, actually, the data we have shows that democracy is possible even in a highly complex technical space like this, and we've seen it working multiple times.
00:38:44.320 --> 00:38:53.920
And so that should actually give us much more motivation to keep learning about these topics and to think about well, how does this apply to me and my local community?
00:38:54.239 --> 00:39:06.800
Is there something or different things that people should be paying attention to that maybe AI CEOs or maybe even policymakers are saying that this is what good democratization of AI should look like?
00:39:06.960 --> 00:39:09.440
But they're actually not good whatsoever.
00:39:09.519 --> 00:39:15.119
And they're just a way to hide the real good ways that we should be fighting against big tech.
00:39:15.199 --> 00:39:21.519
Is there something that you could tell us to protect us from trying to push democratization of AI in a bad way?
00:39:22.000 --> 00:39:30.639
Yeah, I think that's that's exactly the right question to ask, because uh I think there's actually a lot of democracy washing going on right now.
00:39:30.800 --> 00:39:35.599
And so this is a very important danger that I look at in my research.
00:39:35.840 --> 00:39:45.679
And so, what I mean by democracy washing are these kinds of approaches where some company or some government decides to just do like a little survey with a bunch of people.
00:39:45.840 --> 00:39:49.920
OpenAI and Anthropic did this as did a few governments.
00:39:50.079 --> 00:40:03.519
And so often that just means you told people about their current views, you do this like one time, so there isn't like a kind of continuous sense in which you engage with citizens, like you kind of just study citizens.
00:40:03.599 --> 00:40:14.480
So you look at citizens and you try to map their views, but you're not really engaging on the same level in terms of like equal engagement with these people.
00:40:14.639 --> 00:40:21.119
So I think a lot has already gone wrong with these approaches just in terms of like mindset and like how they view citizens.
00:40:21.280 --> 00:40:28.239
And so then, of course, an additional problem is that often these exercises are merely consultative, they're not binding.
00:40:28.400 --> 00:40:34.800
So, what that means is that companies and governments might get a better picture of what normal citizens want.
00:40:34.880 --> 00:40:38.400
And so they might realize, oh, I guess it's very different than what we want.
00:40:38.639 --> 00:40:40.079
And so, guess what?
00:40:40.239 --> 00:40:42.559
Nobody forces them to actually implement that.
00:40:42.800 --> 00:40:56.800
But what makes democracy a unique form of government is that it endows us with binding decision-making power, and that's actually the thing that keeps justice and transparency as intact values, right?
00:40:56.960 --> 00:41:04.639
Otherwise, if we don't have any tools to enforce that, then we have no chance against powerful actors in the system.
00:41:04.800 --> 00:41:17.280
And so the problem I see with these consultative approaches, they might treat citizens like, oh, you know, uh, let's understand your wish list, and then maybe we'll give you this one wish, but nobody's gonna like force us to follow through on that.
00:41:17.440 --> 00:41:23.199
And like, best case, you're gonna feel validated, but you're never gonna have any real control over us.
00:41:23.440 --> 00:41:38.800
And so I think this is just a way of pacifying this dissent and resistance that I mentioned before, while at the same time diminishing it as inherently populist or inherently fear-driven or misguided.
00:41:39.039 --> 00:41:47.760
And so the better way to deal with these kinds of problems actually would be to think about well, how do we get binding decision-making powers in the hands of citizens?
00:41:47.920 --> 00:41:58.239
And so this can happen either via a referendum, which is very bottom-up, uh, or it could mean that citizens successfully pressure their elected officials who then enforce their preferences.
00:41:58.400 --> 00:42:06.239
And so this also means probably really rethinking the kind of guardrails that we have against things like regulatory capture.
00:42:06.400 --> 00:42:15.599
So if we have too much lobbying and too much money in politics, well, we can't really enforce our elected officials doing what we're asking them to do.
00:42:15.760 --> 00:42:21.519
So we might need fundamental legal change on a lot of these anti-lobbying protections.
00:42:21.679 --> 00:42:25.519
So some countries are doing a better job with that currently than others.
00:42:25.679 --> 00:42:37.440
There's a lot of potential to pressure elected officials to make sure that these general democratic guardrails that actually intersect with AI but also go beyond it, that they're intact.
00:42:37.679 --> 00:42:40.639
Otherwise, we're always gonna play catch up, right?
00:42:40.719 --> 00:42:50.480
So if democracy is already hurting at this kind of structural level, it's always gonna be really hard to ask for accountability when it comes to AI policy and governance.
00:42:50.719 --> 00:42:55.119
If we could maybe rapid fire, what are the AI?
00:42:55.440 --> 00:43:05.119
It's gonna be hard to say that AI democracy washing that they should be paying attention to to make sure that they're not sidetracked now that they understand what should be happening.
00:43:05.360 --> 00:43:13.760
What are the agendas or the titles or the just the things that they are seeing in the media that can sidetrack them in terms of AI democracy washing?
00:43:14.000 --> 00:43:15.840
Yeah, maybe let's focus on three.
00:43:16.000 --> 00:43:17.679
First, the doomsday trap.
00:43:17.840 --> 00:43:24.159
So this idea that like we're all gonna die unless we do what corporations want us to do and like please don't regulate too much.
00:43:24.320 --> 00:43:26.960
The second, let's call it the wish list trap.
00:43:27.119 --> 00:43:34.639
So the idea that as long as companies just keep delivering what we want them to deliver subjectively, then actually there isn't a democratic problem.
00:43:34.800 --> 00:43:38.880
Clearly, this is misguided because power still remains outside of civic hands.
00:43:39.039 --> 00:43:42.719
And then finally, let's maybe say the arms race trap.
00:43:42.880 --> 00:43:52.800
So this is a little bit different than the doom state trap because we don't actually need to conjure up this fear about existential risk posed by AI itself.
00:43:53.039 --> 00:43:59.199
What we just need is this idea that another state could develop more powerful AI.
00:43:59.280 --> 00:43:59.760
And so that.
00:44:00.079 --> 00:44:09.440
That's why here, back home, wherever we are, we shouldn't stifle corporate power too much because otherwise we're just gonna lose and there's gonna be war and other conflict.
00:44:09.679 --> 00:44:12.800
And so these three traps I think are highly misleading.
00:44:12.960 --> 00:44:28.639
In all cases, I can imagine mechanisms, or we've already seen examples of mechanisms in which companies claim to include civic preferences, but they have actually not succeeded in fundamentally altering the decisional structures that are at play.
00:44:28.960 --> 00:44:29.920
I think that's so important.
00:44:30.000 --> 00:44:45.280
And I hope that everyone who is watching this or listening to this just pay attention to this agenda that they're trying to set, pay attention to the ways that they're going to try to make you think that yes, they respect our democracy, yes, they respect our preferences when they actually don't.
00:44:45.440 --> 00:44:56.960
And then lastly, I would like to ask you with everything that you know and everything that you have studied, do you really think that it is possible for us to create democratic AI?
00:44:57.360 --> 00:45:02.719
And if we really want to pursue this, what will we need to make that happen?
00:45:03.039 --> 00:45:06.639
I think that's fundamentally an open question, actually, and that's kind of the point, right?
00:45:06.880 --> 00:45:12.159
So we have to decide together what we want to do and what the democratic AI really means.
00:45:12.320 --> 00:45:18.079
If it means something like, okay, we develop a tool that, you know, we control locally in some way.
00:45:18.159 --> 00:45:29.840
So we're not dependent on, say, five US companies, but instead, like we have our own resources and we're able to, you know, design this tool in a way that best fits our preferences.
00:45:30.079 --> 00:45:36.480
This is something that some people think it's a kind of popular approach in the EU and in some other local contexts.
00:45:36.719 --> 00:45:45.760
A lot of people in that discourse talk about sovereign AI, so they're really concerned about dependency of the US and other geopolitical players.
00:45:46.000 --> 00:45:53.840
I think this could be an interesting approach in the sense that it could be more responsive to what local citizens actually want their tools to do.
00:45:54.079 --> 00:46:00.079
But I think it would be dangerous to assume that this is the only possible solution or even the best possible solution.
00:46:00.400 --> 00:46:08.320
Because I can imagine cases in which we build our own sovereign tool, uh, but we still actually have a concentration of power on our ground.
00:46:08.480 --> 00:46:22.079
And so I think the kind of value problem that we need to pay attention to is exactly what we started with oligopolistic concentration of power, authoritarian decision making, and then distorted agenda setting, right?
00:46:22.159 --> 00:46:30.400
So as soon as civic control over this trajectory of our future is diminished by any of these forces, that's the problem.
00:46:30.559 --> 00:46:36.480
It's not really just about, you know, have we solved the dependence on this company or that company?
00:46:36.639 --> 00:46:38.480
So we should resist easy answers.
00:46:38.559 --> 00:46:44.239
And I think we should also be open to the idea that different local solutions can and should look different, right?
00:46:44.320 --> 00:46:53.199
Because like different constituencies are different and they might have different local needs, they might have different ecological constraints, they might have different priorities.
00:46:53.360 --> 00:46:56.239
And so those priorities are actually currently getting ignored.
00:46:56.400 --> 00:47:03.440
One way of democratizing this space is to bring those priorities back into the conversation and to allow people to articulate them.
00:47:03.679 --> 00:47:30.320
It is so important to talk about this localization of artificial intelligence because it's exactly what we have been talking throughout the past hour, which is right now we are using the ideologies of Silicon Valley, we are using the technologies from Silicon Valley, and they have such a specific way of seeing the world, such a specific culture that at the end of the day, in Brazil, in Germany, in uh Japan or wherever country it is, people are using these technologies that are not made for us.
00:47:30.480 --> 00:47:36.400
They're not taking into consideration our culture, our language, our ideologies, the ways that we think, the ways that we don't think.
00:47:36.639 --> 00:47:40.639
We're just using something that was created there and pretending that it works for everyone.
00:47:40.800 --> 00:47:53.039
So big part of democracy is understanding that we are all different, that we have different ways of thinking, and that we should be considering these different ways of thinking in order to create a technology or different technologies that work for everyone.
00:47:53.199 --> 00:47:57.119
So I think that that is such a great way to end this conversation.
00:47:57.280 --> 00:48:06.239
And Leslie, is there anything that you would like people, everyday people, to understand about artificial intelligence that everyone should take away from this conversation?
00:48:06.480 --> 00:48:12.320
I guess uh the main thing I'd like to emphasize is resist this idea of learned helplessness.
00:48:12.559 --> 00:48:19.440
So whenever somebody says it's impossible to learn about this super complex space, you should question it.
00:48:19.679 --> 00:48:27.920
Because in very many cases, the real heart of these decisions is something that we can all understand and engage with.
00:48:28.079 --> 00:48:33.679
We might not always agree, but it's something that is achievable for all of us citizens.
00:48:33.920 --> 00:48:35.760
So, everyone, take this takeaway.
00:48:36.079 --> 00:48:37.039
Learn about AI.
00:48:37.119 --> 00:48:50.880
It doesn't matter if you like it, it doesn't matter if you hate it, you need to learn how these technologies work, what are the ideologies behind them, because that is the only way that we will be able to make artificial intelligence democratic for everyone and create technologies that work for us.
00:48:51.119 --> 00:48:53.280
So thank you so much, Annette, for joining us today.
00:48:53.440 --> 00:48:56.079
Don't forget to pick up their book on democratizing AI.
00:48:56.159 --> 00:48:59.599
And I'll see you next Tuesday for a new episode of Bots Don't Cry.
00:48:59.679 --> 00:49:09.519
And if you have heard the podcast until the end, please write down what was the biggest takeaway that you're taking away from this conversation today and also post the emoji of the planet.
00:49:09.679 --> 00:49:09.920
Why?
00:49:10.079 --> 00:49:15.519
Because it is really important for us to understand that everyone needs to be part of this conversation in order to democratize AI.
00:49:15.599 --> 00:49:19.199
So if you heard it until the end, let me know that you're part of the Stay Until the End gang.
00:49:19.360 --> 00:49:19.840
Bye everyone.
00:49:20.000 --> 00:49:20.639
See you next Tuesday.
00:49:20.719 --> 00:49:21.280
Chao, ciao.
00:00:00.000 --> 00:00:03.520
Did you vote for Sam Altman, Mark Zuckerberg, Elon Musk?
00:00:03.759 --> 00:00:04.959
No, me neither.
00:00:05.120 --> 00:00:14.240
But somehow they and a handful of other tech billionaires are making decisions about AI that completely change your life, your kids' lives, and our democracy.
00:00:14.400 --> 00:00:20.000
So now it's time for us to ask: why do tech billionaires have so much power in deciding our lives?
00:00:20.160 --> 00:00:21.839
Is this actually democratic?
00:00:22.000 --> 00:00:23.600
And what can we do about it?
00:00:23.760 --> 00:00:26.079
So that's exactly what we will be discussing today.
00:00:26.239 --> 00:00:27.760
Hi, my name is Katarina Doria.
00:00:27.920 --> 00:00:35.280
I'm an AI ethicist, and this is Bots Don't Cry, the podcast where every Tuesday I'll teach you critical AI literacy with other critical AI experts.
00:00:35.439 --> 00:00:41.759
And today we are super duper lucky to actually get to unpack AI and democracy with political philosopher Dr.
00:00:41.920 --> 00:00:42.960
Annette Zimmerman.
00:00:43.039 --> 00:00:43.280
Dr.
00:00:43.359 --> 00:00:50.560
Annette Zimmerman is a tenured track professor of philosophy and co-founder of the Center for Uncertainty and AI at UW Madison.
00:00:50.799 --> 00:01:00.640
They wrote Democratizing AI, won the APA's Public Philosophy Prize, and have advised the US, UK, and German governments, as well as UNESCO and the OECD.
00:01:00.880 --> 00:01:02.719
So thank you so much for being here today with us.
00:01:02.799 --> 00:01:04.400
I'm so excited to talk about this topic with you.
00:01:04.640 --> 00:01:06.560
Yeah, thank you so much for having me, Katarina.
00:01:06.640 --> 00:01:08.079
It's uh exciting to be here.
00:01:08.560 --> 00:01:09.280
It's so exciting.
00:01:09.439 --> 00:01:12.000
Okay, so let's start with a really basic question.
00:01:12.239 --> 00:01:14.319
You are a political philosopher.
00:01:14.560 --> 00:01:20.159
Why do we need politics and philosophy in order to understand what is happening with AI today?
00:01:20.400 --> 00:01:22.400
Well, uh, there's political scientists, right?
00:01:22.480 --> 00:01:31.439
And like these people basically look at what's happening in politics and the structures around everyday politics, and they analyze how that works.
00:01:31.519 --> 00:01:34.319
So they kind of look at the world as it is.
00:01:34.480 --> 00:01:37.840
But political philosophers ask a slightly different question.
00:01:38.000 --> 00:01:42.400
So we're interested in this question of how should politics work?
00:01:42.719 --> 00:01:46.560
So it's a normative rather than empirical question.
00:01:46.719 --> 00:01:52.239
We're interested in the kinds of goals and values that should guide our collective decisions.
00:01:52.480 --> 00:02:06.400
And so this has become incredibly important in the AI space because all of a sudden we see these tech CEOs behaving as though they're elected officials, when in fact, as you pointed out before, they're really not.
00:02:06.719 --> 00:02:07.200
100%.
00:02:07.519 --> 00:02:11.039
No one voted for them, and then suddenly they have so much power.
00:02:11.199 --> 00:02:19.120
And what are the problems that that could lead us as a society to have just a handful of powerful people in Silicon Valley deciding for our lives?
00:02:19.520 --> 00:02:21.280
Well, I think there's multiple issues.
00:02:21.520 --> 00:02:26.719
One issue is like any concentration of power and wealth is always dangerous for a democracy.
00:02:27.039 --> 00:02:48.560
So we ideally don't want to get ourselves in a situation where just a handful of people andor companies concentrate like all of this money and all of this political influence in one part of society, but at the same time, they're actually able to have a really widespread distribution of all sorts of risks associated with AI.
00:02:48.719 --> 00:02:58.319
So we get this like really crazy asymmetry where, on the one hand, we concentrate the benefits, but we collectivize all of the risks and costs of AI.
00:02:58.479 --> 00:03:03.199
And so that stands in need of democratic justification, at least in my mind.
00:03:03.360 --> 00:03:13.199
So we need to ask, are we actually comfortable as a collective with that kind of asymmetrical distribution, or do we need to change some things?
00:03:13.360 --> 00:03:17.439
And so this is where democracy uh becomes relevant in this space.
00:03:18.319 --> 00:03:19.199
And I love what you're saying.
00:03:19.360 --> 00:03:20.479
Do we need to change or not?
00:03:20.560 --> 00:03:29.840
I could say that most of the people who follow my work and are into critical AI, they say, yes, we need to change this, but there's something that you discuss here, the costs and the risks.
00:03:30.000 --> 00:03:33.919
Could you tell us a little bit more about what are these costs and what are these risks?
00:03:34.319 --> 00:03:47.280
Yeah, so there's a cluster of more well-understood costs and risks that have already surfaced, that many experts in the field, many of whom you've already spoken to, and many of your uh audience members will know already.
00:03:47.439 --> 00:03:49.360
They've already investigated a lot of this.
00:03:49.520 --> 00:04:02.000
So racial bias, gender bias, machine hallucinations, like vulnerable users getting drawn into like very damaging behavior, mass uh you know, polarization and propaganda enabled by AI applications.
00:04:02.159 --> 00:04:09.680
So there's like a pretty broad cluster of very urgent kind of present-day risks that we actually already know about.
00:04:10.000 --> 00:04:17.519
Then on the other hand, there are some more mysterious future-oriented risks that are much more contentious.
00:04:17.759 --> 00:04:23.920
Audience members of yours might have encountered like recent debates about will AI kill all of us, right?
00:04:24.000 --> 00:04:26.480
Like, does AI pose an existential risk?
00:04:26.639 --> 00:04:40.079
And so this is where a lot of scholars actually say no, we need to be very cautious not to give people this seemingly confident forecast about there's like a 10% chance we're all gonna die in a few years.
00:04:40.319 --> 00:04:44.319
Because it's very hard to actually make that kind of forecast in a rigorous way.
00:04:44.480 --> 00:05:01.199
But at the same time, AI companies are very much emphasizing these ostensible risks because, of course, you know, it adds a lot of hype and mystique around their products if there's a discourse that says, look, like this is really powerful, but also really, really dangerous.
00:05:01.360 --> 00:05:11.360
So there's actually also been a critical backlash against that recent discourse that says, hey, listen, like, are your motives really just to expose this uh long-term risk?
00:05:11.680 --> 00:05:20.160
Or do maybe other political, economic, and societal factors play into the ways in which we want to forecast the future?
00:05:20.319 --> 00:05:25.920
And so the story might be a little bit more complex than just putting a percentage point on some sort of forecast.
00:05:26.240 --> 00:05:27.279
I think that's really interesting.
00:05:27.360 --> 00:05:35.199
And we're going to talk more about the ideologies that are kind of controlling the narratives that get said in AI today, exactly what you said.
00:05:35.360 --> 00:05:39.040
We have a handful of people saying that artificial intelligence will destroy the world.
00:05:39.199 --> 00:05:41.199
So we're going to leave that for the second part.
00:05:41.439 --> 00:05:48.000
But now I think that at the same time that you know these tech CEOs they're telling that artificial intelligence is really dangerous.
00:05:48.160 --> 00:06:06.240
They still want to double down on the idea that artificial intelligence is marvelous for us, that we should love artificial intelligence because it democratizes access to education or to the arts, and you know, because everyone has a phone today and you can have touch PT Claude in your pocket and you can do all of these things.
00:06:06.399 --> 00:06:12.959
So basically, trying to say that, hey, you should love this technology because everyone has access to this and that they're democratizing knowledge.
00:06:13.120 --> 00:06:16.000
Do you agree with this type of commentary?
00:06:16.240 --> 00:06:20.079
Yeah, nice job on clocking this important narrative, Katarina.
00:06:20.160 --> 00:06:40.000
I think this is really key to understanding what's going on in this debate, because uh this has become a really central argument pushed by AI industry that, like, look, there might be some sort of asymmetrical distribution, but at the end of the day, please don't complain, regular citizens, because like this is all good for you, like you want to have an easier life.
00:06:40.160 --> 00:06:44.160
And so now you don't need to be writing like boring emails all on your own.
00:06:44.319 --> 00:06:47.279
Like, you know, you get this like slight convenience increase.
00:06:47.439 --> 00:06:55.839
And so even if there's material inequality in society at like vast degrees, you should accept that because you're getting some sort of benefit.
00:06:56.079 --> 00:07:00.720
And so political philosophers have been very suspicious of this kind of argument.
00:07:00.879 --> 00:07:08.399
In our field, we have developed theories that show why these kinds of distributions are highly morally and politically suspect.
00:07:08.560 --> 00:07:24.240
Uh, political philosophers often talk about so-called win-win exploitation, where some exploiter gets like a huge benefit, and some other party, the exploited one, might still get like some benefit, but there's still an unfair transaction between these parties.
00:07:24.480 --> 00:07:31.759
And so not all moral problems disappear just because some sort of benefits are part of the story.
00:07:32.000 --> 00:07:40.319
It matters how we distribute them and who's actually able to call the shots about how those benefits go around.
00:07:40.560 --> 00:07:50.480
And so what we see in AI, we just have a lot of unilateral decision-making power in this very, very small handful of corporations.
00:07:50.639 --> 00:08:04.639
So we have an oligopoly in this market, and so this oligopoly is able to uh decide pretty much on their own in many cases when to deploy a tool, how soon to deploy it, under which conditions, and so on.
00:08:04.800 --> 00:08:06.639
And so that's really the core of the problem.
00:08:06.800 --> 00:08:10.319
We shouldn't get sidetracked with this benefits talk.
00:08:10.639 --> 00:08:14.399
Could you explain to us a little bit more about what do you mean by oligopoly?
00:08:14.720 --> 00:08:19.839
So this basically just means that there is a highly concentrated market.
00:08:20.079 --> 00:08:25.759
In a super free market, you have like many different companies and they're in competition with each other.
00:08:25.920 --> 00:08:32.000
And actually, interestingly, this idea of a free market kind of sounds like a libertarian or a conservative idea.
00:08:32.240 --> 00:08:46.159
But even like democratic socialists on the harder left are fans of you know having a more robust market in a democratic society in which we don't have a lot of powerful big fishes that can call the shots on everything.
00:08:46.480 --> 00:09:03.840
So, like uh under uh US President Biden, for instance, the FTC under uh Lena Khan started, you know, trying to break up big tech again after basically 12 years of not really doing all that much to curtail the power of corporations in this uh tech space in the US.
00:09:04.000 --> 00:09:14.240
And so briefly that actually changed because both progressives and conservatives recognized we actually have a problem if we only have a few companies calling the shots and amassing more and more power.
00:09:14.399 --> 00:09:24.159
So as soon as we move away from the free market, we either get an oligopoly, so like four or five companies, or even a monopoly, just like one company that dominates an entire sector.
00:09:24.320 --> 00:09:32.159
And so arguably both of these are not just bad for regular consumers, but they're actually bad for democracy.
00:09:32.399 --> 00:09:39.840
Because as soon as I have this kind of power concentration, it's not just anymore about, oh, this company can dictate prices.
00:09:40.000 --> 00:09:46.559
It's actually also about this company might be influencing all of our political decisions in some way.
00:09:46.720 --> 00:09:50.480
And so the problem actually goes much deeper than just economic problems.
00:09:50.799 --> 00:10:00.480
I think this is so fascinating because I do think that many times people think that just because, you know, open AI and then throw up, they're creating AI, that this is a tech problem.
00:10:00.639 --> 00:10:10.000
But we need to understand that when we have maybe five companies deciding artificial intelligence and technology that is impacting everyone, we're talking about politics here.
00:10:10.080 --> 00:10:15.200
We're talking about the agendas of what gets said, what doesn't get said, the things that are decided.
00:10:15.279 --> 00:10:22.720
And again, they're not decided by us, but only five companies or five people who are really, really wealthy and deciding these agendas.
00:10:22.879 --> 00:10:34.320
And I think one way that they were able to create this agenda setting is that somehow today, the artificial intelligence industry, they're considered different from anyone else.
00:10:34.480 --> 00:10:48.559
So for example, when we think about big pharma or if you want to create a new drug, what happens is that, you know, you come up with the idea, you do research, you do studies, and then you go to the standard body and you say, hey, I want to create this new drug, it's going to help human beings.
00:10:48.720 --> 00:10:52.799
The standard body says, okay, you did the test, or no, you didn't, push it to the market.
00:10:52.960 --> 00:11:09.039
Another example, if I want to sell chicken pie in my streets here in Sao Paulo, I need to take pictures of my kitchen, I need to, you know, say what is my credentials, put everything in a document, go to the food standard bodies and say, hey, I want to create this, I want to sell chicken pie, and they will say, fine.
00:11:09.200 --> 00:11:16.000
But then suddenly here we're talking about five people or five companies that are doing everything backwards.
00:11:16.240 --> 00:11:23.200
They say, I'm going to create this technology and I'm just going to push it in the market, and you, everyone, fend for yourself with the consequences.
00:11:23.360 --> 00:11:38.559
And that is partly because they have so much power, and also partly because they created this narrative that they are different and that when we think about AI innovation, that we shouldn't be playing the same rules of the game that basically all other industries.
00:11:38.720 --> 00:11:40.399
Could you tell us a little bit why?
00:11:40.559 --> 00:11:47.519
Not only they have so much power, but also why can they just play the game that everyone is playing, but with different rules?
00:11:47.759 --> 00:11:52.399
Yeah, that's a brilliant summary of a lot of my core uh concerns in this space.
00:11:52.559 --> 00:11:52.960
Thank you.
00:11:53.120 --> 00:12:02.399
I think, as you correctly pointed out, uh, the power to deploy a powerful new technology is itself a power to shift a democratic agenda.
00:12:02.559 --> 00:12:12.000
And so if I can totally change everyone's life by just saying, okay, hey, there's this new tool that I generously now give to you.
00:12:12.080 --> 00:12:13.759
And by the way, you can use it for free.
00:12:13.840 --> 00:12:22.639
So like please all start using it, and like that, of course, will help me optimize my tool so that I can come up with like some freemium model and like make a ton of money with that, right?
00:12:22.879 --> 00:12:40.799
So this is all part of a deployment strategy, of course, but it also immediately forces a political response because, of course, immediately people start wondering well, you know, now that this new thing happened, what should we perhaps regulate it, even though we might not have a good regulatory basis in place?
00:12:40.960 --> 00:12:46.159
And so this happened actually very early on in the space, even before the generative AI hype cycle.
00:12:46.320 --> 00:12:53.840
So if you recall, the predecessor model of Chat GPT, this was called GPT-2 at the time.
00:12:54.080 --> 00:13:08.799
And so back then, OpenAI was going around putting out all of these press releases saying, like, oh, we can't possibly deploy this out in the open, like no regular citizens can get their hands on it because it's way too dangerous, it'll kill us all, right?
00:13:08.879 --> 00:13:10.399
Like we've we've heard this before.
00:13:10.480 --> 00:13:13.600
So this is like an older narrative, many people don't realize.
00:13:13.840 --> 00:13:23.759
And so they, of course, completely reverted course as soon as they were ready to deploy a direct-to-consumer product in the form of Chat GPT.
00:13:23.840 --> 00:13:33.200
And so, of course, back then that triggered that generative AI deployment rush with all of these other companies that also wanted to go ahead and deploy their own tools.
00:13:33.440 --> 00:13:43.840
Now, on the side of politics, this triggered a very interesting type of discourse that in my book, Democratizing AI, I call AI exceptionalism.
00:13:44.000 --> 00:13:56.480
It's the idea that, as you just mentioned before, we should treat AI as a as a domain and as an industry as fundamentally mysterious and unique and unlike any other industry that we've ever seen before.
00:13:56.639 --> 00:14:06.559
And so when people buy into this narrative, it unfortunately makes it more likely that they'll actually also think about AI regulation in this AI exceptionalist way.
00:14:06.720 --> 00:14:17.279
And so we see this play out all the time when we see big tech CEOs going to like, you know, Senate or the Congress and saying things like, well, you know, it's just a black box.
00:14:17.440 --> 00:14:22.960
Like we need real deep technical expertise before we can even consider how to regulate that.
00:14:23.200 --> 00:14:31.360
And so, of course, we don't think about other highly complex domains of science and technology policy in the same way.
00:14:31.600 --> 00:14:35.440
I always think about climate change mitigation policies.
00:14:35.679 --> 00:14:40.559
So I don't have a meteorology PhD, nor do I have a physics PhD.
00:14:40.720 --> 00:14:47.440
But I have particular political preferences about how important it is, I think, to try to mitigate climate change.
00:14:47.519 --> 00:14:56.639
And so, as a regular citizen with no technical expertise, I can actually contribute to those policy debates by saying, like, I think this should be a key priority for us.
00:14:56.799 --> 00:15:00.320
I really want us to mitigate the threats associated with climate change.
00:15:00.480 --> 00:15:09.679
And so, very similarly, I think regular citizens can articulate at least some value-based preferences about what should happen with AI.
00:15:09.840 --> 00:15:23.200
For instance, if you ask people, hey, are you okay with mass surveillance powered by AI, ideally powered by a company that also has a really deep enmeshment with an increasingly authoritarian government?
00:15:23.440 --> 00:15:29.039
Chances are many people might say, actually, no, that doesn't align with my political preferences.
00:15:29.200 --> 00:15:34.559
And so, boom, all of a sudden, we're having a very non-exceptionalist discourse on AI policy and governments.
00:15:34.799 --> 00:15:35.759
It's very much possible.
00:15:35.840 --> 00:15:38.000
And I think that's actually where we need to get to.
00:15:38.159 --> 00:15:47.120
So I want to get more people who don't have a computer science PhD to really start asking themselves, like, what do I want our future to look like?
00:15:47.279 --> 00:15:52.320
And what role does AI play in that future or doesn't it play, perhaps?
00:15:52.559 --> 00:16:03.039
I think that's also really interesting because in getting people to understand how they could impact the narrative, they also need to understand how artificial intelligence is impacting them day to day, right?
00:16:03.200 --> 00:16:19.200
I have many followers that say, hey, I don't want to learn about AI, I hate AI, I don't want to be part of this conversation, or that somehow just because they dislike it, artificial intelligence is not impacting their lives, not only in a technical space, also in an ideological place.
00:16:19.440 --> 00:16:44.879
So I would like to talk about how having this oligopoly and having these tech CEOs creating these products and pushing them in the market, they also are able to do something that everyone needs to understand, which is setting up political agendas or the agendas of what gets spoken about, what doesn't get spoken about, the things that we're deciding in terms of education or things about climate change as well.
00:16:45.120 --> 00:16:50.320
Just by pushing products in the market, they're also ideologically shaping that agenda.
00:16:50.559 --> 00:17:00.080
Could you tell us a little bit more about that and also how that ends up impacting people's lives, not only in a technical sense, but in a living day-by-day sense?
00:17:00.399 --> 00:17:07.359
The most concerning thing that I think happens in this regard is a narrowing of our imagination of what's possible.
00:17:07.519 --> 00:17:19.519
And so normally when somebody doesn't mess with our collective agenda setting process in a democracy, that means that we're having a lot of debates in society, not just in the context of elections, by the way, right?
00:17:19.599 --> 00:17:25.039
So like democracy is something bigger than just like, oh, you go and like cast your vote every four years.
00:17:25.200 --> 00:17:29.039
That's like a really thin notion of democracy that I want to overcome.
00:17:29.359 --> 00:17:46.000
I think democracy happens daily when you talk to people, ideally also people that you disagree with, and you form opinions, you deliberate about different options, and ideally you come to a fuller understanding of what you and your neighbors and your co-citizens think is desirable and possible.
00:17:46.240 --> 00:17:58.880
But of course, if you have this massive concentration of power, this can lead to some policy preferences and some arguments gaining like a lot of traction also in the media and in the public imagination.
00:17:59.119 --> 00:18:08.559
This actually gets exacerbated by the fact that some tech CEOs are also doing a lot of very aggressive media company acquisitions.
00:18:08.720 --> 00:18:20.400
So Larry Allison comes to mind, a figure who now has really, really far-reaching control over a vast media ecosystem right before the midterm elections.
00:18:20.640 --> 00:18:24.240
And also, of course, this has global reach far beyond the United States.
00:18:24.480 --> 00:18:35.039
And so these are tech CEOs that don't just see themselves as executives, they see themselves as political activists who actively shape these narratives and these options.
00:18:35.279 --> 00:18:42.799
And so two examples of potentially misleading or artificially narrow options that I've seen recently are the following.
00:18:43.039 --> 00:18:52.400
One is this idea that comes from OpenAI, but also from Palantir that says, look, we're in an arms race with China and maybe other uh foreign adversaries.
00:18:52.559 --> 00:19:06.880
And so, therefore, if you want democratic AI, you gotta back American companies because, of course, these American companies are operating in uh ostensibly democratic structures, while uh, you know, competitor companies abroad are not doing that.
00:19:07.039 --> 00:19:13.440
And so this argument often gets fused with a bid for lesser regulation in the tech space.
00:19:13.599 --> 00:19:25.440
I noticed that, you know, Alex Carp, the Palantir CEO, wrote a book justifying this idea, also justifying the idea that actually we shouldn't separate corporations from government.
00:19:25.599 --> 00:19:27.599
Actually, we should have a deep enmeshment.
00:19:27.759 --> 00:19:30.559
We should welcome the idea of a military-industrial complex.
00:19:30.720 --> 00:19:40.559
And so, you know, for obvious reasons, I think this can quickly be a very narrow type of view that uh, you know, suggests to people that they only have two choices, right?
00:19:40.640 --> 00:19:43.519
When actually more choices might be available to them.
00:19:43.599 --> 00:19:51.680
Like maybe it's not just, you know, either we all die under authoritarian dictatorship or we we regulate some tech companies.
00:19:51.759 --> 00:19:54.160
You know, like maybe these are not the only two options.
00:19:54.400 --> 00:19:58.720
I also was very interested to see another type of narrative emerge.
00:19:58.880 --> 00:20:00.160
This is over the past few years.
00:20:00.319 --> 00:20:06.400
I noticed that Sam Altman went to a Senate hearing that I covered very extensively in my research.
00:20:06.480 --> 00:20:10.799
I kind of did a deep dive on this particular hearing in 2023.
00:20:11.039 --> 00:20:18.160
And I noticed that when Altman was asked by lawmakers, hey, how should we actually regulate this technology?
00:20:18.319 --> 00:20:21.599
They were kind of saying, like, oh, we're clueless, like, we don't even know where to start.
00:20:21.839 --> 00:20:23.279
He actually came prepared.
00:20:23.440 --> 00:20:27.359
So he said, Look, I have an idea, I think you should do it.
00:20:27.599 --> 00:20:35.039
And so while he was outlining this idea, he was also asked directly by these lawmakers, so who should implement this idea?
00:20:35.200 --> 00:20:38.000
Who should oversee the governance of AI in America?
00:20:38.160 --> 00:20:44.799
And so he just laughs and he just says, Well, we can send your office a list of the people we think are qualified.
00:20:44.960 --> 00:20:51.440
And so he gets this response in the hearing of people on both sides of the aisle saying, Well, thank you.
00:20:51.599 --> 00:20:53.599
We really want your recommendations.
00:20:53.759 --> 00:21:04.079
And so I think this is very indicative of an artificially narrow option space where all of these other possible voices aren't really part of the conversation anymore.
00:21:04.240 --> 00:21:11.839
And all of these other ideas that they might have about how to regulate AI aren't really heard in this discourse either.
00:21:12.000 --> 00:21:19.839
And so it was no surprise then that at that hearing, Altman floated this idea of, you know, you don't need to regulate all that much.
00:21:20.000 --> 00:21:22.480
You just need to implement a licensing regime.
00:21:22.640 --> 00:21:32.799
The idea that in order to keep deploying AI and to reject democratic agendas, you should just be a big company that is able to apply for a license.
00:21:32.960 --> 00:21:34.880
And so the government oversees that.
00:21:34.960 --> 00:21:38.640
And so, you know, this handful of companies can then get that license.
00:21:38.880 --> 00:21:54.319
That sounds kind of responsible and nice, but what he didn't say was that actually only a small number of corporate actors would be able to apply for that kind of license successfully under the conditions that he had articulated.
00:21:54.559 --> 00:21:59.680
And so it would have actually solidified the immense market power that layers.
00:22:00.079 --> 00:22:09.599
Like OpenAI and Anthropic have, and artificially keep out any new competition, like any new startup, any small research team, any foreign competitor, anything.
00:22:09.839 --> 00:22:20.480
And so this is a very sneaky narrative because it sounds really ethical, but it actually really misleads us about what else might be possible.
00:22:20.799 --> 00:22:24.319
So this is what I have in mind when I think about AI exceptionalism.
00:22:24.720 --> 00:22:43.839
Whenever, and and for people who are listening this episode, I actually, we already talked before, and when you told me about the concept of AI exceptionalism, I started seeing this everywhere, everywhere in the narratives, everywhere around me, how artificial intelligence is seen completely different, how AI CEOs are seen in a completely different way.
00:22:44.000 --> 00:22:46.480
And it's something that I started sharing with my friends.
00:22:46.640 --> 00:22:54.160
It's, you know, we're not asking the sugar CEOs if they recommend people eating sugar or not, because we know that they are biased.
00:22:54.240 --> 00:22:58.480
We know that whatever they say will be fully biased in order for us to eat more sugar.
00:22:58.559 --> 00:23:03.680
But then suddenly, when it comes to AI CEOs, they are invited into Congress hearings.
00:23:03.759 --> 00:23:20.559
They're invited in these interviews to tell us what we should be thinking about in terms of regulating them, which seems to me that their agenda in big terms is also bringing this sort of neutrality in how they benefit from this discourse.
00:23:20.799 --> 00:23:34.559
So that narrow way of thinking also comes a lot with their ideologies and what they see is right, what they see is wrong, and what they see as to what our society should be striving for.
00:23:34.720 --> 00:23:38.799
So I think that this sort of conversation, we need to start talking about ideologies.
00:23:38.960 --> 00:23:42.799
We can't just talk about this without understanding what they actually believe in.
00:23:42.960 --> 00:23:47.839
So could it be possible for us to start talking about what are these ideologies that they're following?
00:23:48.079 --> 00:23:49.279
What is going around?
00:23:49.359 --> 00:23:54.640
And I know that effective altruism comes to play here, also effective accelerationism.
00:23:54.799 --> 00:24:00.079
Could you tell us a little bit more about that and how that gets implemented into the narrow narratives that we see today?
00:24:00.400 --> 00:24:12.960
Yeah, I started getting really interested in this question because uh, you know, I'm a philosophy professor and I started noticing that a bunch of these tech CEOs are suddenly saying, oh, you know, like I wrote a philosophical manifesto.
00:24:13.119 --> 00:24:16.400
That's usually an indicator that something has gone terribly wrong.
00:24:16.559 --> 00:24:19.200
Uh it's not easy to like just do philosophy.
00:24:19.359 --> 00:24:27.440
And so surprise, surprise, a lot of this stuff isn't, well, let's put it this way, it wouldn't pass uh in a philosophy class.
00:24:27.519 --> 00:24:34.960
Uh, because there's often actually many logical flaws in uh in these manifestos and treatises that we get from uh CEOs.
00:24:35.119 --> 00:24:41.599
I think it's also important to not put all of these pseudo-philosophical outputs in just one bucket.
00:24:41.759 --> 00:24:45.759
So it's not the case that there is one Silicon Valley ideology.
00:24:46.000 --> 00:24:50.319
They have actually splintered out into all of these different and weird camps.
00:24:50.480 --> 00:24:58.319
We started out earlier, like the older Silicon Valley that is still, you know, very much present in people's imagination and in the media.
00:24:58.559 --> 00:25:10.400
That is a kind of simpler story where nerds in Silicon Valley used to be kind of economically libertarian, like they like to tinker, they like to like, you know, deploy a cool product in the market and like get massively rich with that.
00:25:10.559 --> 00:25:16.240
But they also used to have this brand of being kind of socially liberal, at least in a kind of superficial sense, right?
00:25:16.400 --> 00:25:19.279
But so things have really shifted in recent times.
00:25:19.440 --> 00:25:26.799
Uh, you already mentioned effective accelerationism and effective altruism as two really important movements in this space.
00:25:26.960 --> 00:25:37.039
Uh, to that, let's add a third, and that is right-wing libertarianism, which is increasingly turning into dark enlightenment kind of alt-right authoritarianism.
00:25:37.200 --> 00:25:41.599
So that's a kind of different ideological trajectory that that works in parallel there.
00:25:41.759 --> 00:25:46.640
Let's maybe start with that last bit because I think it's it's increasingly gaining a lot of traction.
00:25:46.799 --> 00:25:58.400
So I'm thinking in particular of people like Peter Thiel here, who even like back in the day, like decades ago, would write these essays about how you know freedom and democracy are no longer compatible.
00:25:58.480 --> 00:26:04.880
And so if you really care about capitalism, then uh, you know, you gotta toss democracy to the side.
00:26:05.119 --> 00:26:14.240
Because ever since states decided to enfranchise women and poor people, that really messed things up for the capitalists, and so democracy stopped working, according to him.
00:26:14.480 --> 00:26:19.519
He later apologized for making that claim in the essay, but it was a pretty lukewarm apology.
00:26:19.680 --> 00:26:25.039
And actually, he seems to have doubled down on this democracy skepticism since.
00:26:25.119 --> 00:26:45.519
And instead, writing further philosophical essays in which he talks about the decline of the Western world and threats from mysterious foreign places against that Western world, that coincides with a more general political ideological drift to the hard right that has recently happened, especially over the past decade.
00:26:45.680 --> 00:26:52.319
And so one of Peter Thiel's mentees, Curtis Jarvin, is the figurehead of part of this movement.
00:26:52.559 --> 00:27:02.240
Curtis Jarvin is a blogger who is known for publicizing explicitly racist views in his writings.
00:27:02.400 --> 00:27:09.759
He's also known for articulating the view that actually we should abolish democracy, we should be run by a CEO monarchy.
00:27:09.920 --> 00:27:15.119
So the democracy skepticism actually is really key to understanding this whole ideology.
00:27:15.359 --> 00:27:27.440
There's a whole bunch of other kind of socially conservative influences in this ideology too, but it would be a mistake to think of them just as, oh, these are Republicans or social conservatives.
00:27:27.680 --> 00:27:41.119
These are people who go much further to the right and who basically think that the Enlightenment, the idea of giving people equal rights and like having a transparent discourse in society, that was a big mistake.
00:27:41.200 --> 00:27:43.839
And so that's why it's called Doric Enlightenment ideology.
00:27:43.920 --> 00:27:48.640
So they want to backtrack on all of these ideas of human rights, civil rights, and so on.
00:27:48.799 --> 00:27:53.920
Let's pivot a little bit to the other two twin theories that you mentioned before.
00:27:54.160 --> 00:28:03.359
So we have effective altruism, which started out as a kind of niche philosophical view that was originally much more concerned about like non-AI issues.
00:28:03.440 --> 00:28:15.839
So like back in the day, 10, 15 years ago, this was a group of people who were committed to the moral theory of consequentialism, the idea that the morally right thing to do is to maximize good outcomes.
00:28:16.000 --> 00:28:24.079
And so with this logic of maximizing good outcomes, they started thinking about well, how can we do good things in a better way?
00:28:24.240 --> 00:28:42.319
But so as a community, at some point recently-ish, they decided actually, if we want to maximize good outcomes, then we should be very concerned about a potential scenario in which AI could kill us all, because blocking that doomsday scenario is then absolutely important.
00:28:42.480 --> 00:28:45.440
It kind of trumps all of these other nice charity considerations.
00:28:45.599 --> 00:28:53.119
And so this movement really pivoted to this idea of being singularly focused on this one particular threat model.
00:28:53.359 --> 00:29:03.359
At the same time, the movement has been criticized by, first of all, being led by people who all tend to fit the same kind of narrow sociodemographic criteria.
00:29:03.440 --> 00:29:08.400
So it doesn't seem to be a very diverse movement, especially when we look at the leading figures.
00:29:08.640 --> 00:29:18.160
And this movement tends to de-prioritize a lot of issues that people in fact care about right here and right now, not even just thinking about doomsday scenarios.
00:29:18.319 --> 00:29:22.079
So, like right now, society has racism and sexism.
00:29:22.319 --> 00:29:25.599
Right now we have massive political division and unrest.
00:29:25.680 --> 00:29:31.119
And like there's so many real-world problems, we don't have to wait until this doomsday scenario to fix them.
00:29:31.279 --> 00:29:38.240
It seems like effective altruists don't really want to pivot their attention to these shorter term considerations.
00:29:38.559 --> 00:29:47.680
We can define the last one of those three super quickly, effective accelerationism, is just the backlash movement against that, uh, against effective altruism.
00:29:47.839 --> 00:30:04.240
And so this happened because a lot of people in the valley started getting really annoyed with all of this AI doomer discourse, and they started saying, no, that's instead of stopping AI, we actually need to pump the gas and just like really accelerate and keep deploying way more AI.
00:30:04.400 --> 00:30:06.319
And that's the thing that's gonna save humanity.
00:30:06.559 --> 00:30:16.319
And so these people, for instance, represented by venture capitalist Mark Andreessen, who wrote the so-called techno-optimist manifesto, they just want to push forward, right?
00:30:16.480 --> 00:30:24.240
So they their kind of a salvation narrative is something like well, the best possible AI is the thing that makes all of our lives better.
00:30:24.480 --> 00:30:43.359
What's interesting about this movement is that they too don't seem to be particularly diverse, and they also seem to center this idea of a few very, very smart people being able to save everyone if we just stop regulating them and annoying them so much.
00:30:43.519 --> 00:30:50.799
And so this idea of the hero genius who just needs to be let loose looms really large in this movement too.
00:30:51.119 --> 00:31:01.200
And why do you think that everyday people need to understand these ideologies to understand artificial intelligence today and how our democracy is being impacted by them?
00:31:01.519 --> 00:31:09.200
I think it's really important to be able to clock a narrative that might not be the only analysis of what's actually happening to us right now.
00:31:09.440 --> 00:31:24.799
So when a lot of people hear AI might kill us all, uh they understandably start getting upset and scared because here we have so-called experts telling us, oh, you know, like we might all die unless we follow this movement or like this preferred policy solution.
00:31:24.960 --> 00:31:34.319
And so often things are presented as facts when they're actually just more like opinions or at least kind of contested predictions about the future.
00:31:34.559 --> 00:31:51.519
Here it's important to notice that it's very hard to make very rigorous, scientifically defensible predictions of the future, especially when we look at the kinds of doomsday forecasts that we get from some people who are behind these ideological movements.
00:31:51.759 --> 00:32:00.720
So very often, an argument that sounds a lot like it's inevitable that X will happen is actually just one possibility.
00:32:00.880 --> 00:32:22.240
And so it's important for us as normal citizens and also as members of the academic community, as activists, as members of nonprofits, and so on, to think about okay, what are some alternative future paths that we want to pay attention to in order to contest this narrative that some specific course of events is actually inevitable.
00:32:22.480 --> 00:32:30.799
And so that's why it's really important not to just let policymakers try to ask themselves these questions, because this actually affects all of us, right?
00:32:30.880 --> 00:32:42.559
Like it's gonna affect my decision about whether I support AI-powered surveillance, if I think that I have to accept AI-powered surveillance because otherwise the free world will die, right?
00:32:42.640 --> 00:32:55.359
Like if I just accept that meta-narrative, then a whole bunch of my other opinions that somehow intersect with AI will be totally distorted by whatever these tech CEOs are telling us.
00:32:55.519 --> 00:33:06.640
And so being able to identify that and to say, oh, actually, that sounds a lot more like an opinion or a preference rather than an unshakable fact is really important in order to be an empowered citizen.
00:33:07.039 --> 00:33:22.000
It seems like understanding and clocking out and not trusting AI CEOs at face value is one of the ways that we're able to democratize AI in a way that I understand how this impacts my life, and I can clock that out because that's an ideological assertion.
00:33:22.240 --> 00:33:26.000
I might not need that, not want that, and I think that there's a different way for that.
00:33:26.079 --> 00:33:27.279
So maybe that is one.
00:33:27.440 --> 00:33:36.079
But are there more ways that we as a society, we everyday people, can push for this democratization of AI that clearly is not happening today?
00:33:36.400 --> 00:33:47.680
I think one very interesting opportunity that can be important in some cases, at least, is to actually say, hey, we don't actually need AI for this particular task.
00:33:47.920 --> 00:33:58.160
And so this is an argument that often gets vilified right now by the media and by tech industry, and it gets called something like AI populism by some.
00:33:58.240 --> 00:34:16.000
So there have been a number of recent think pieces that, for example, commented on the fact that like young people at graduation ceremonies started booing Eric Schmidt and other speakers who said, Oh, you know, I know you guys are worried about never getting a job, but you just need to adapt to AI and like this is the future.
00:34:16.159 --> 00:34:17.280
And so people weren't having it.
00:34:17.360 --> 00:34:19.199
Like people were really upset about that.
00:34:19.360 --> 00:34:38.960
And I think it's important not to dismiss that as just populism, but just because it happens to be a negative opinion about AI in domains like education or like in algorithmic hiring, which plenty of studies have shown to be very, very biased against women and people of color.
00:34:39.199 --> 00:34:40.079
And so people know this.
00:34:40.159 --> 00:34:42.239
People have good reasons to be very concerned about this.
00:34:42.320 --> 00:34:48.880
Like upon graduation, they'll be interviewed by some AI interviewer, they'll be unfairly assessed in many cases.
00:34:49.119 --> 00:34:55.519
Nobody keeps these systems in check, and so they'll just not get a fair shake after spending years and like a lot of money on their education.
00:34:55.760 --> 00:35:02.159
And so it's it's really key to allow that as like part of the democratic process.
00:35:02.320 --> 00:35:10.159
It can't just be that we deliberate about, oh, we should regulate AI this way or that way, but we have to accept it nonetheless.
00:35:10.320 --> 00:35:11.840
So this is like one important thing.
00:35:12.079 --> 00:35:17.599
Then I think the question is well, how do we actually translate this into collective decision making?
00:35:17.760 --> 00:35:21.599
And so here it really depends what kind of constituency we're talking about.
00:35:21.760 --> 00:35:27.360
I've seen some very positive examples of very local decision making about AI.
00:35:27.440 --> 00:35:33.360
So like often we kind of think, oh, democracy is just about like state-level elections and so on.
00:35:33.519 --> 00:35:35.679
But actually, democracy can happen in the city.
00:35:35.840 --> 00:35:46.239
Democracy can happen in like a federal state, democracy can even happen transnationally when different local groups decide to push for some agenda and they inspire each other.
00:35:46.480 --> 00:35:52.239
One concrete example I can think of is actually even before this generative AI wave.
00:35:52.400 --> 00:35:58.639
And that's the public debate in some cities on facial recognition technologies used in policing.
00:35:58.880 --> 00:36:23.840
I remember that in 2019, in San Francisco, no less, right in the middle of the valley, people locally decided, and this was very much led by local activists working on criminal justice issues, not just technical experts, they decided, hey, actually, we're very skeptical of using facial recognition tools for police surveillance because we're concerned about racial bias in this context.
00:36:24.000 --> 00:36:31.519
At that time, plenty of academic experts had already mapped many important risks for the population and their civil rights.
00:36:31.679 --> 00:36:34.079
And so they were able to successfully push this through.
00:36:34.159 --> 00:36:54.400
In 2019, San Francisco passed the stop secret surveillance ordinance, and that actually triggered this fascinating snowball effect across the globe, where like multiple American cities, but also multiple cities in other countries decided to also impose a local deployment moratorium until these risks can be better understood.
00:36:54.559 --> 00:36:58.719
And many of them have actually upheld that block, some of them have reverted it.
00:36:58.880 --> 00:37:18.800
But nonetheless, we see there is like a lot of space for bottom-up grassroots activism, even if it doesn't end up being this like national movement, it can be extremely transformative for local populations, and it can have this really inspirational effect across like different constituencies.
00:37:18.960 --> 00:37:33.440
And so I think we should look at that example and think about well, where else can normal citizens who see a problem in their city, in their local community, where can they push for a decision that makes a small difference?
00:37:33.679 --> 00:37:40.239
And so this can be the beginning point of a much larger push for a positive change.
00:37:40.480 --> 00:37:48.320
And so, of course, at the same time, it's really important to have like legitimate, democratically legitimate, top-down regulation, right?
00:37:48.400 --> 00:37:55.039
Where we get our elected officials to make some better choices that don't just accept an industry narrative.
00:37:55.199 --> 00:38:04.000
And so we do need like legal change, we need like good regulatory frameworks, but that can't be all that there is to democratizing AI.
00:38:04.159 --> 00:38:13.119
We also need these spaces for more flexible, more bottom-up kinds of activism inspired by these positive use cases that we've already seen.
00:38:13.360 --> 00:38:44.079
So I guess one message that I'd really want people to remember from this conversation is that even though it seems that democracy is really messy and like really hard and can often go wrong, and like it might also seem like it's impossible to do anything against these powerful corporations, actually, the data we have shows that democracy is possible even in a highly complex technical space like this, and we've seen it working multiple times.
00:38:44.320 --> 00:38:53.920
And so that should actually give us much more motivation to keep learning about these topics and to think about well, how does this apply to me and my local community?
00:38:54.239 --> 00:39:06.800
Is there something or different things that people should be paying attention to that maybe AI CEOs or maybe even policymakers are saying that this is what good democratization of AI should look like?
00:39:06.960 --> 00:39:09.440
But they're actually not good whatsoever.
00:39:09.519 --> 00:39:15.119
And they're just a way to hide the real good ways that we should be fighting against big tech.
00:39:15.199 --> 00:39:21.519
Is there something that you could tell us to protect us from trying to push democratization of AI in a bad way?
00:39:22.000 --> 00:39:30.639
Yeah, I think that's that's exactly the right question to ask, because uh I think there's actually a lot of democracy washing going on right now.
00:39:30.800 --> 00:39:35.599
And so this is a very important danger that I look at in my research.
00:39:35.840 --> 00:39:45.679
And so, what I mean by democracy washing are these kinds of approaches where some company or some government decides to just do like a little survey with a bunch of people.
00:39:45.840 --> 00:39:49.920
OpenAI and Anthropic did this as did a few governments.
00:39:50.079 --> 00:40:03.519
And so often that just means you told people about their current views, you do this like one time, so there isn't like a kind of continuous sense in which you engage with citizens, like you kind of just study citizens.
00:40:03.599 --> 00:40:14.480
So you look at citizens and you try to map their views, but you're not really engaging on the same level in terms of like equal engagement with these people.
00:40:14.639 --> 00:40:21.119
So I think a lot has already gone wrong with these approaches just in terms of like mindset and like how they view citizens.
00:40:21.280 --> 00:40:28.239
And so then, of course, an additional problem is that often these exercises are merely consultative, they're not binding.
00:40:28.400 --> 00:40:34.800
So, what that means is that companies and governments might get a better picture of what normal citizens want.
00:40:34.880 --> 00:40:38.400
And so they might realize, oh, I guess it's very different than what we want.
00:40:38.639 --> 00:40:40.079
And so, guess what?
00:40:40.239 --> 00:40:42.559
Nobody forces them to actually implement that.
00:40:42.800 --> 00:40:56.800
But what makes democracy a unique form of government is that it endows us with binding decision-making power, and that's actually the thing that keeps justice and transparency as intact values, right?
00:40:56.960 --> 00:41:04.639
Otherwise, if we don't have any tools to enforce that, then we have no chance against powerful actors in the system.
00:41:04.800 --> 00:41:17.280
And so the problem I see with these consultative approaches, they might treat citizens like, oh, you know, uh, let's understand your wish list, and then maybe we'll give you this one wish, but nobody's gonna like force us to follow through on that.
00:41:17.440 --> 00:41:23.199
And like, best case, you're gonna feel validated, but you're never gonna have any real control over us.
00:41:23.440 --> 00:41:38.800
And so I think this is just a way of pacifying this dissent and resistance that I mentioned before, while at the same time diminishing it as inherently populist or inherently fear-driven or misguided.
00:41:39.039 --> 00:41:47.760
And so the better way to deal with these kinds of problems actually would be to think about well, how do we get binding decision-making powers in the hands of citizens?
00:41:47.920 --> 00:41:58.239
And so this can happen either via a referendum, which is very bottom-up, uh, or it could mean that citizens successfully pressure their elected officials who then enforce their preferences.
00:41:58.400 --> 00:42:06.239
And so this also means probably really rethinking the kind of guardrails that we have against things like regulatory capture.
00:42:06.400 --> 00:42:15.599
So if we have too much lobbying and too much money in politics, well, we can't really enforce our elected officials doing what we're asking them to do.
00:42:15.760 --> 00:42:21.519
So we might need fundamental legal change on a lot of these anti-lobbying protections.
00:42:21.679 --> 00:42:25.519
So some countries are doing a better job with that currently than others.
00:42:25.679 --> 00:42:37.440
There's a lot of potential to pressure elected officials to make sure that these general democratic guardrails that actually intersect with AI but also go beyond it, that they're intact.
00:42:37.679 --> 00:42:40.639
Otherwise, we're always gonna play catch up, right?
00:42:40.719 --> 00:42:50.480
So if democracy is already hurting at this kind of structural level, it's always gonna be really hard to ask for accountability when it comes to AI policy and governance.
00:42:50.719 --> 00:42:55.119
If we could maybe rapid fire, what are the AI?
00:42:55.440 --> 00:43:05.119
It's gonna be hard to say that AI democracy washing that they should be paying attention to to make sure that they're not sidetracked now that they understand what should be happening.
00:43:05.360 --> 00:43:13.760
What are the agendas or the titles or the just the things that they are seeing in the media that can sidetrack them in terms of AI democracy washing?
00:43:14.000 --> 00:43:15.840
Yeah, maybe let's focus on three.
00:43:16.000 --> 00:43:17.679
First, the doomsday trap.
00:43:17.840 --> 00:43:24.159
So this idea that like we're all gonna die unless we do what corporations want us to do and like please don't regulate too much.
00:43:24.320 --> 00:43:26.960
The second, let's call it the wish list trap.
00:43:27.119 --> 00:43:34.639
So the idea that as long as companies just keep delivering what we want them to deliver subjectively, then actually there isn't a democratic problem.
00:43:34.800 --> 00:43:38.880
Clearly, this is misguided because power still remains outside of civic hands.
00:43:39.039 --> 00:43:42.719
And then finally, let's maybe say the arms race trap.
00:43:42.880 --> 00:43:52.800
So this is a little bit different than the doom state trap because we don't actually need to conjure up this fear about existential risk posed by AI itself.
00:43:53.039 --> 00:43:59.199
What we just need is this idea that another state could develop more powerful AI.
00:43:59.280 --> 00:43:59.760
And so that.
00:44:00.079 --> 00:44:09.440
That's why here, back home, wherever we are, we shouldn't stifle corporate power too much because otherwise we're just gonna lose and there's gonna be war and other conflict.
00:44:09.679 --> 00:44:12.800
And so these three traps I think are highly misleading.
00:44:12.960 --> 00:44:28.639
In all cases, I can imagine mechanisms, or we've already seen examples of mechanisms in which companies claim to include civic preferences, but they have actually not succeeded in fundamentally altering the decisional structures that are at play.
00:44:28.960 --> 00:44:29.920
I think that's so important.
00:44:30.000 --> 00:44:45.280
And I hope that everyone who is watching this or listening to this just pay attention to this agenda that they're trying to set, pay attention to the ways that they're going to try to make you think that yes, they respect our democracy, yes, they respect our preferences when they actually don't.
00:44:45.440 --> 00:44:56.960
And then lastly, I would like to ask you with everything that you know and everything that you have studied, do you really think that it is possible for us to create democratic AI?
00:44:57.360 --> 00:45:02.719
And if we really want to pursue this, what will we need to make that happen?
00:45:03.039 --> 00:45:06.639
I think that's fundamentally an open question, actually, and that's kind of the point, right?
00:45:06.880 --> 00:45:12.159
So we have to decide together what we want to do and what the democratic AI really means.
00:45:12.320 --> 00:45:18.079
If it means something like, okay, we develop a tool that, you know, we control locally in some way.
00:45:18.159 --> 00:45:29.840
So we're not dependent on, say, five US companies, but instead, like we have our own resources and we're able to, you know, design this tool in a way that best fits our preferences.
00:45:30.079 --> 00:45:36.480
This is something that some people think it's a kind of popular approach in the EU and in some other local contexts.
00:45:36.719 --> 00:45:45.760
A lot of people in that discourse talk about sovereign AI, so they're really concerned about dependency of the US and other geopolitical players.
00:45:46.000 --> 00:45:53.840
I think this could be an interesting approach in the sense that it could be more responsive to what local citizens actually want their tools to do.
00:45:54.079 --> 00:46:00.079
But I think it would be dangerous to assume that this is the only possible solution or even the best possible solution.
00:46:00.400 --> 00:46:08.320
Because I can imagine cases in which we build our own sovereign tool, uh, but we still actually have a concentration of power on our ground.
00:46:08.480 --> 00:46:22.079
And so I think the kind of value problem that we need to pay attention to is exactly what we started with oligopolistic concentration of power, authoritarian decision making, and then distorted agenda setting, right?
00:46:22.159 --> 00:46:30.400
So as soon as civic control over this trajectory of our future is diminished by any of these forces, that's the problem.
00:46:30.559 --> 00:46:36.480
It's not really just about, you know, have we solved the dependence on this company or that company?
00:46:36.639 --> 00:46:38.480
So we should resist easy answers.
00:46:38.559 --> 00:46:44.239
And I think we should also be open to the idea that different local solutions can and should look different, right?
00:46:44.320 --> 00:46:53.199
Because like different constituencies are different and they might have different local needs, they might have different ecological constraints, they might have different priorities.
00:46:53.360 --> 00:46:56.239
And so those priorities are actually currently getting ignored.
00:46:56.400 --> 00:47:03.440
One way of democratizing this space is to bring those priorities back into the conversation and to allow people to articulate them.
00:47:03.679 --> 00:47:30.320
It is so important to talk about this localization of artificial intelligence because it's exactly what we have been talking throughout the past hour, which is right now we are using the ideologies of Silicon Valley, we are using the technologies from Silicon Valley, and they have such a specific way of seeing the world, such a specific culture that at the end of the day, in Brazil, in Germany, in uh Japan or wherever country it is, people are using these technologies that are not made for us.
00:47:30.480 --> 00:47:36.400
They're not taking into consideration our culture, our language, our ideologies, the ways that we think, the ways that we don't think.
00:47:36.639 --> 00:47:40.639
We're just using something that was created there and pretending that it works for everyone.
00:47:40.800 --> 00:47:53.039
So big part of democracy is understanding that we are all different, that we have different ways of thinking, and that we should be considering these different ways of thinking in order to create a technology or different technologies that work for everyone.
00:47:53.199 --> 00:47:57.119
So I think that that is such a great way to end this conversation.
00:47:57.280 --> 00:48:06.239
And Leslie, is there anything that you would like people, everyday people, to understand about artificial intelligence that everyone should take away from this conversation?
00:48:06.480 --> 00:48:12.320
I guess uh the main thing I'd like to emphasize is resist this idea of learned helplessness.
00:48:12.559 --> 00:48:19.440
So whenever somebody says it's impossible to learn about this super complex space, you should question it.
00:48:19.679 --> 00:48:27.920
Because in very many cases, the real heart of these decisions is something that we can all understand and engage with.
00:48:28.079 --> 00:48:33.679
We might not always agree, but it's something that is achievable for all of us citizens.
00:48:33.920 --> 00:48:35.760
So, everyone, take this takeaway.
00:48:36.079 --> 00:48:37.039
Learn about AI.
00:48:37.119 --> 00:48:50.880
It doesn't matter if you like it, it doesn't matter if you hate it, you need to learn how these technologies work, what are the ideologies behind them, because that is the only way that we will be able to make artificial intelligence democratic for everyone and create technologies that work for us.
00:48:51.119 --> 00:48:53.280
So thank you so much, Annette, for joining us today.
00:48:53.440 --> 00:48:56.079
Don't forget to pick up their book on democratizing AI.
00:48:56.159 --> 00:48:59.599
And I'll see you next Tuesday for a new episode of Bots Don't Cry.
00:48:59.679 --> 00:49:09.519
And if you have heard the podcast until the end, please write down what was the biggest takeaway that you're taking away from this conversation today and also post the emoji of the planet.
00:49:09.679 --> 00:49:09.920
Why?
00:49:10.079 --> 00:49:15.519
Because it is really important for us to understand that everyone needs to be part of this conversation in order to democratize AI.
00:49:15.599 --> 00:49:19.199
So if you heard it until the end, let me know that you're part of the Stay Until the End gang.
00:49:19.360 --> 00:49:19.840
Bye everyone.
00:49:20.000 --> 00:49:20.639
See you next Tuesday.
00:49:20.719 --> 00:49:21.280
Chao, ciao.
