Jonathan Schaeffer: I go to academic conferences. I can honestly say that of the hundreds and thousands of people I've met over the years, as researchers, as scientists, etc., every one of them believes in AI for the good. And that's
Martin Wheatley: Mm.
Jonathan Schaeffer: what we're working for. And we all believe that the potential For AI to revolutionize in a good way the world, literally, everybody in the world to improve their quality of life. We believe in it and we believe it will happen. And yes, there are bad scenarios that could happen. But we shouldn't get dragged down and have hypothetical bad scenarios obstruct the real goal, which is to make this world better, safer. more equitable for literally everybody around the world.
Martin Wheatley: Hello and welcome to the Frontier Leader Show. I'm Martin Wheatley and each week I speak to leaders, coaches, practitioners and technologists working out what good leadership looks like in an AI-shaped world. While you still have the app open and in your hand on your phone, please hit follow or subscribe so you don't miss out on the next episode. And if you love the content in this conversation. Then if you head to www.theaileaderlab.com, I have a growing number of resources about how you can start to lead with AI. But today's guest we are speaking to Dr. Jonathan Schaefer. Now Jonathan has been at the forefront of artificial intelligence for more than 40 years, an AI researcher, entrepreneur, author, and inventor. He's dedicated his career to building practical, trustworthy AI systems that solve real world problems. He's the inventor of Kind, a privacy-first AI platform that helps professionals and organizations search, analyse and gain insights from their own curated knowledge base. And his accomplishments include leading the development of AI systems that mastered checkers or drafts, if you're in the UK, and competed against world-class poker players, both recognized by the Guinness Book of World Records, which is a feat in itself. So Jonathan, welcome to the show.
Jonathan Schaeffer: Thank you very much, Martin. It's a pleasure to be here.
Martin Wheatley: And I think this is genuinely the first time I've ever ever spoken to someone in the Guinness Book of World Records. It used to be it used to be there used to be a television programme on when I was a kid called Record Breakers. And I used to love it every week, you know, that they'd go out and watch people try and break records. So fantastic. Absolutely.
Jonathan Schaeffer: watching the break records wouldn't have been quite as exciting, I think, as some of the stuff
Martin Wheatley: Ha ha.
Jonathan Schaeffer: that made it to T V.
Martin Wheatley: Yeah, maybe. But just for for people who don't know you and obviously I I gave a bit of an intro there, but you've obviously, like we said, been in AI for over forty years. You know, how how how has your kind of career what's your career arc been and how have you seen the the overnight success of of AI develop over the last over the last forty years?
Jonathan Schaeffer: Well, I've been around since the dinosaur age. so I've seen AI evolve from systems that were heavily I'll call it curated by humans, where the humans would supply all the knowledge. These were very brittle systems because humans are not very good at explaining, for example. How do you do things? How do you drive a car? How are how, as a doctor, do you come up with a diagnosis? These systems called expert systems required humans to precisely specify all the information necessary for these systems to operate at a, well, by today's standard, a mediocre level. But over the 40 years, what's been happening is actually really. Amazing. We think of the human brain as this incredible machine, which it is, that has this vast amount of knowledge, and we've learned so much. And it seems natural to extract that information and give it to our computer underlings, our computer students, if you will, these programs that are trying to learn how to be good doctors, good drivers, good good authors. And over the 40 years, what's happened is that human knowledge has become less and less important. And in many fields
Martin Wheatley: Just because that's important to add in.
Jonathan Schaeffer: of AI right now is almost irrelevant because we let the the AI discover and learn on its own. And to me, that's been something I hadn't expected way back when, when I started. it's the path towards Computer discovery has has just slowly and methodically moved in that direction. And then the world changed. The world changed on November 30th, 2022, and you you're gonna say, what a geek. Why would you remember that date? Well, it's because
Martin Wheatley: I would have to.
Jonathan Schaeffer: I had seen ear earlier versions of Chat GPT, and quite frankly, they were amusing, but they, if you don't mind, I'll put in a technical word, they sucked. Okay. But on November 30th, 2022, it was like a technological revolution to me. Because these algorithms which didn't perform particularly well, like literally overnight, did amazing things. And I was stunned. I was professionally stunned, but also I was stunned at at the world's reaction because like literally overnight everybody was playing with Chat GPT. And in many ways it was predictable because it was following a clear path that we knew about. But It's this hockey stick phenomenon where something happens and it doesn't go in this straight linear line, it leaps
Martin Wheatley: Okay.
Jonathan Schaeffer: up, and that's exactly what happened. And of course, since then we're seeing something that is unprecedented in human history. We are seeing technological innovation over the last four years at an unprecedented rate. The amount of money that's going into AI. research that's happening, the infrastructure that's being invested in. It's all staggering. And there's no precedent for this kind of degree of innovation happening on this, relatively speaking, very small timescale. So I'm stunned at what's happened and yet as you'll see as we talk about it, I'm also very concerned about what's happening.
Martin Wheatley: Hmm. And the elements I think you might be well, you'll correct me if I'm wrong, but might be concerned about, which I think a lot of people are concerned about, is is the element of trust and privacy around how these models, certainly how the cloud models with, you know, Chat GPT, Claude, Gemini, whichever one to use. perhaps we can pivot slightly into to talking about what I obviously mentioned that you've you've the founder of Kind and and what you're building out there. 'Cause I think that answers quite a lot of the questions and how people can use it in a in a in a better, safer way.
Jonathan Schaeffer: Yes, so thank you for asking. when I started in AI as a student many, many decades ago, I had a vision for AI and this will date me because when I was young I got addicted to watching Star Trek. And in Star Trek there's a computer, right? Spock Captain Kirk would talk to the computer, and the computer had no personality. The computer never chatted with Kirk, right? You would ask it a question, or Kirk or Spock would ask it a question, and it would go off in compute and it would give a fact-based answer. And if it couldn't answer the question, it said so. And in my mind, that was my model for AI. That's not the AI that we have today. There's gaps. So when you look at what ChatGPT and all the other LLMs, large language models like ChatGPT, Gemini, Claude, et cetera, their fundamental algorithm is flawed. There's a certain amount of what I'll call coin flipping that goes into it, which means that answers are non-deterministic. You just can't You can't predict what the answer is going to be, and the answer will change. But because of this coin flipping that goes on, These programs make mistakes. They make factual errors. They fabricate. And it's really cute. We don't call them errors. Why don't we call them errors? We call them hallucinations.
Martin Wheatley: That's the one.
Jonathan Schaeffer: this is part of the anthropomorphization that's going on with these programs. We we chat with them. No, we're not chatting with them. We're we're asking questions. I don't believe we should be anthropomorphizing these programs. We should be using them as tools. We should be using them as tools to do a different kind of AI, not artificial intelligence, but augmented intelligence. We want to use these as tools, like Kirk and Spock did, to augment us. To enhance our decision-making ability. We're in charge, we're using the computers to inform us, advise us, and we ultimately take the decision as to whether to use it or not. And we have to take the responsibility. So these programs have problems and The problems go deeper than just accuracy. They're insatiable for data. The current models get better with more data. And so these companies have a vested interest in getting data from you. And data is multifold. One is that we put our files into the cloud where these companies can see them. And read them and use them to train their AI. I object
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: to that. We can use these tools to ask questions. Even if we
Martin Wheatley: We can use these
Jonathan Schaeffer: don't put our files up there, we can still go to ChatGPT or Claude and ask questions. Asking a question is data. It tells these companies how you're using the tool and what kind of information you're looking for. That's more data. There are things that I do that I don't mind putting into the cloud. There are things that I do that I don't mind asking questions of Claude, for example, and knowing that whatever I interact with this program, it's using it, it's gathering data on me. They say there's no free lunch, and that's exactly what's going on here. I can use these free tools,
Martin Wheatley: Now
Jonathan Schaeffer: but they're gathering data from me. They get value by giving me free access.
Martin Wheatley: Yeah.
Jonathan Schaeffer: But there are things that I don't think belong in the cloud. There are things that are personal, they're private to me. For example, I have a directory on my computer
Martin Wheatley: And go back.
Jonathan Schaeffer: here with my medical history. Do I really want to put that into the cloud? Do I really want one of these LLMs to see it and train off of it? No. I have another directory with all my financial information and investments. It's my information, it's not theirs. I have directories with my intellectual property, my ideas. I recently wrote a book. Do I want all that in ingested into these large language models? No. I believe in privacy. Some things don't have to be private, but some things do. So after 40 years of being a professor, I've had a great ride. I started talking the talk about the problems I see with AI today, with hallucinations, with lack of privacy, there are security problems, there are safety problems with chat. there are issues of advertising that I object to, there's issues with monetization of individuals that I object to. I also am have Concerns about all the data's data center mania that's going on. So I can talk the talk, but I decided to walk the walk. And so I left the university and I formed a new company. It's called Sincera Software Solutions. And we've released our first two products, which were called Kind. And what they are is desktop AI. Okay, that doesn't sound sexy, desktop AI. But the idea here is that. These companies want me to move my data to the cloud to get these AI processing. I'm flipping the script. I want to move the AI from the cloud to my desktop so that I can get my processing. So, what kind does is it's a walled garden on your desktop. And what you do is you just drag and drop files into kind. And what kind is, is it has sophisticated AI technology that does the analysis of your files on your desktop and allows you to do what the chatbots do. You can search your file, you can search your files, you can ask questions about your files, and it only works on the data that you give it. So, if you, for example, if I put in all my financial information into kind, there's a wall around it. And if I now ask a question about my financial information, the answer only comes from my data. There are no external influences that affect the answer. There's no hallucinations because everything is fact-based from my data. So that's in a nutshell what the idea is, is kind puts on your desktop your own private AI, which you can do ask questions and do searching. It's vendor agnostic. We don't train on your data at all. there's no internet resources used so you can turn off the internet and you still have high quality local AI processing. It's safe, it's secure, it's private, no hallucinations. That's what it's all about.
Martin Wheatley: Yeah, and that and that's so it it's is it an equivalent to being a personalized local, secure kind of notebook LM? So to speak, or so it's not called Notebook L now, Gemini Notebook, whatever it's called now. I'm assuming it it's it's that kind of it's that kind of idea that you can drop your files, your documents in there and then you can interrogate just those, but obviously it's it's secured locally. That's summarizing.
Jonathan Schaeffer: Correct. So No Notebook LM is is a good example, but the problem
Martin Wheatley: Mm.
Jonathan Schaeffer: with for me, the problem with Notebook LM is to use it, I have to put my files into the cloud. That means
Martin Wheatley: Mm, mm.
Jonathan Schaeffer: Google sees them. So Obviously, our tool is not as powerful as Notebook LM. We don't have, you know, a thousand software developers like Google, but what we do have is a powerful desktop AI that sits on your desktop. You don't move your files to the cloud, nobody sees them, and you get the benefits of leading edge AI.
Martin Wheatley: Yeah. And I guess I guess there's a lot of if you follow the the AI hype on YouTube and everything, there's all this noise around second brains and building things out with Obsidian. I'm sure you've I'm sure you've seen this in in in your world more more than more than once, maybe. But I guess the again the difference with this is you can connect Claude or whatever you can to Obsidian, which obviously is then your files on your drive, but They still get sent to and from to and from Claude, don't they? Whereas yours is purely local. Yeah.
Jonathan Schaeffer: Correct. And and that's the important part. it's
Martin Wheatley: Mm.
Jonathan Schaeffer: the privacy. One of the things that I've seen over the years is that We've lost the notion of privacy.
Martin Wheatley: Mm.
Jonathan Schaeffer: People, I I think it started with social media. People said, Wow, this is really cool. I have a social media account. I can upload all my photos. I can say things. And the companies are very happy for that because they have data that they can use and learn from and take advantage of. But When cloud computing came along, now it was hey, give us all your files. And yeah,
Martin Wheatley: Yeah.
Jonathan Schaeffer: guess what? You get free backups and free storage and all these other good things. But again, there's no free lunch. They're getting something from it, and they get to see your data. And when AI came along, sorry, not AI per se, but when ChatGPT came along and all of a sudden there was this rush towards massive investment in AI. Everybody wanted your data, so we're you're encouraged now to move your files even even more so now to the cloud, which again just it's a loss of privacy. And again, the point here is that some things are I'm okay with, but other things I'm not. And so I don't want to disadvantage the individual. By forcing them to move their files to the cloud. If you want something private, get many of the advantages or most of the advantages of internet-based AI, but put them on your desktop where
Martin Wheatley: Mm.
Jonathan Schaeffer: you maintain your privacy. The world has moved towards this
Martin Wheatley: Yeah.
Jonathan Schaeffer: view that we should be moving all of our data into the cloud. And there are only a handful of large companies who offer the kind of cloud infrastructure that that give you vast amounts of storage and access to leading edge tools. And that's making these companies richer
Martin Wheatley: He's amazing.
Jonathan Schaeffer: and more powerful. And I'm not sure that that's the right way to go.
Martin Wheatley: No, I think I think you're right with the the whole mass rollout of social media. I think it has desensitized people to to giving away their data. Constant status updates, photos, share and like you said, when the cloud came along and took away the problem of managing stuff locally, yeah, people are just used to doing it, aren't you? But this is a very a very secure way of of managing your your sensitive data, definitely.
Jonathan Schaeffer: But you know, it doesn't even have to be what you would call sensitive data.
Martin Wheatley: Mm, mm.
Jonathan Schaeffer: Th I I I I think that I have data which I wouldn't regard as sensitive, but there's still a sense of ownership and I I just don't want other companies or companies knowing about some of my personal stuff. I like to collect certain things. I have a
Martin Wheatley: Mm.
Jonathan Schaeffer: Lots of information that I've collected over the years about my particular interest and it's mine. So and I'll tell you one thing that's interesting about some of the information I I've collected, which is which caught me by surprise when I started developing kind. I collect books, okay? Very simple. And
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: in 1995, basically the World Wide Web sort of came along and people put things on the web. And I would download on on I would put onto my desktop book catalogs or auction results or interesting articles because I read the stuff, I collect the stuff, I use it to to enhance my knowledge. Well Companies come and go. Companies purge their website. I discovered that a lot of the knowledge of the stuff that I have is no longer on the web. In other words, I can ask questions on the internet and it can't find the answer.
Martin Wheatley: Mm.
Jonathan Schaeffer: But I have the answer on my laptop. And to use powerful AI to get those answers is amazing. But I don't wanna have to take this and put it in the on the web because now Google has access to it and all of a sudden this information ends up g going out there to the world. But in some sense, I wanna keep it. It's mine.
Martin Wheatley: Yeah, yeah. No, I g I understand that completely. Understand that completely. You you've mentioned a you've mentioned a couple of times we've been speaking about enhancing enhancing thinking and and you know, not replacing thinking. And one of the one of the big criticisms, especially with people who are now steadily relying on AI, is it's taken away a lot a lot of that critical thought process. You know, they put in a question, personal or work related, they grab the first answer that comes out of chat GPT, whatever, and then the pass it on up the chain of of whatever. W what's your kind of take on take on that and how products like kind can can help augment the the the human in the loop, so to speak.
Jonathan Schaeffer: So I'm a scientist, ideal of accuracy. If I write a paper
Martin Wheatley: Right.
Jonathan Schaeffer: and it has falsehoods in it, I'm gonna get shot down or embarrassed. The problem with the current large language models is, and I use them, I'll ask questions and they'll give me answers, but I never
Martin Wheatley: Mm.
Jonathan Schaeffer: use them as is. I have to invest the time to verify whether that information is correct. And so you can think of Chat GPT as wow, look at this fantastic time saver, because I ask a question, I've got the answer, I'm done. For me, and I I would hope for most people, particularly professionals. You're not done because you must understand that there is a reasonable probability that information in the answer given back by one of these chat programs is fabricated or incorrect. And therefore, you have to take the responsibility to address it. Now, I I would say that these are incredibly valuable tools for finding things, but you have to take the responsibility. What happens with kind though is I'll I'll give you an example.
Martin Wheatley: Mm.
Jonathan Schaeffer: I worked, I recently wrote a book, An Introduction to AI for non-technical people. And I could write on my my laptop. And when I found some information on the web that was really interesting that I liked, I put it on my laptop into my into my kind collection to do with this book. And over time, I had, here's all my writing, and here are all the references that are in my book, all the documents that I found and all the web pages that were interesting. And now I could ask questions that would only come from my data, nothing else out there, because there's a lot of garbage out there.
Martin Wheatley: Mm.
Jonathan Schaeffer: Therefore, I could ask questions about my book, and I could verify that the answers were correct. But I also discovered something interesting. So just as taking a step back, kind sits on your desktop. I I don't mean to be nosy, but how busy is your desktop? Your computer. During the day, maybe it's busy. It's busy right now. It's doing a video
Martin Wheatley: Mm.
Jonathan Schaeffer: recording. Is it working 24 hours a day? You know, maybe you're running hundreds of agents or doing big number crunching? Probably not. What's it doing on the weekends? It's probably idle. The point is you got a lot of compute horsepower on your desktop or your laptop.
Martin Wheatley: Yeah.
Jonathan Schaeffer: Well, We can harness it because you've put kind on your desktop and we do the AI processing. So in the case of my book, I put in all this information, and before I published the book, I started asking questions because I wanted to verify that everything was correct, et cetera. And I found mistakes. And so what we did is we've added a feature to kind that allows it to look within your data. For factual inconsistencies. Now you can't do that with Google. You can't say, hey Google, here are all my files. Tell me if there's any factual mistakes in them. But we can, because we're on your desktop.
Martin Wheatley: Hmm.
Jonathan Schaeffer: Your data set, you may think that your data, your medical files, your financial files, your book files, your collecting files, whatever your files are, your PowerPoint files. You maybe say, Wow, I got lots of data. You have small data. The amount of data that's on your machine, okay, how many gigabytes is it? Gigabytes is small. Google's dealing with petabytes, which is
Martin Wheatley: Mm.
Jonathan Schaeffer: millions of times bigger, right? So we can run. During the evenings and weekends, and do things with your data that you can't do in the web. And therefore,
Martin Wheatley: Mm.
Jonathan Schaeffer: inside a collection where we've got boundaries on where it's looking, because it's only looking in your data, kind is able to do interesting things that you can't do elsewhere. But we have access to your CPU and you have the resources that allow us to do it. And what who cares if it happens overnight? It's still
Martin Wheatley: Yeah.
Jonathan Schaeffer: a value add, and it gives you more insights, more capabilities, a better understanding of the data that you have.
Martin Wheatley: Yeah. It's definitely an interesting shift definitely from from the cloud based solutions that are there. 'Cause like you said, using the capabilities or the the infrastructure if you like on on your laptop or device in the background when you're not using it as much is it's just an efficient way of using a device, isn't it, at the end of the
Jonathan Schaeffer: But let's think about it this way: it's a sunk cost. You've already
Martin Wheatley: Mm, mm.
Jonathan Schaeffer: purchased the machine. It's just sitting there, it's depreciating. Use your
Martin Wheatley: Yeah, yeah.
Jonathan Schaeffer: resources. But there's an important advantage about that. Because if you're doing your AI processing on your desktop, you're asking questions and getting answers, or you're analyzing some of your data all on your machine, you're not using. Claude, you're not using Chat GPT, which means you're not sending queries to these data centers. You're being environmentally responsible. I'd like to say, you know, the environmental movement says think globally and act locally. Well,
Martin Wheatley: Mm.
Jonathan Schaeffer: with kind, you're removing your queries from those data centers. You're putting them onto a device that you already own and has no cost. You're not paying for tokens. You're being environmentally responsible by not encouraging further growth of this data center mania.
Martin Wheatley: Yeah. Does it does it how does it I I guess what one of the key challenges I've seen when trying to do something, obviously not not solely on my my desktop per se, but is kind of agreeing that the structure of where should I put this note? How should I record this? Does does kind do that for you to some extent, or does the user still have to think, well, that this is this is my project on Books for argument's sake, just coming back to your example. Does does kind take care of that or is it still user driven in terms of the the setup?
Jonathan Schaeffer: So this is what happens. First of all, we never move your files. Your files,
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: they're on your machine. we never move them, we never remove them, et cetera. So all you you do is in kind you create a collection. I might create a collection called say financial.
Martin Wheatley: Yeah.
Jonathan Schaeffer: All you do is drag and drop into that collection any files that you think are relevant. I created
Martin Wheatley: Mm.
Jonathan Schaeffer: a collection for my book, and I just drag and drop into there anything that I think is interesting. I mean, you have curated the data. You're putting
Martin Wheatley: Mm.
Jonathan Schaeffer: into these collections the material that you think is important and relevant. And it could be material that's on your desktop, or it could be material that you get from the web, but you've curated
Martin Wheatley: Yeah.
Jonathan Schaeffer: it, you've put it in there. And kind automatically analyzes it and adds it into its own internal database. But you don't have to reorganize your files or or anything like that. Just drag and drop whatever you think is relevant to that topic and you're done.
Martin Wheatley: Hm. Okay. Interesting. How does it work for obviously 'cause it's not connected to the web. How does it work from does it is it built for businesses as well as personal or is it just personal at the minute in terms of knowledge base?
Jonathan Schaeffer: So originally it was envisioned for business
Martin Wheatley: Mm.
Jonathan Schaeffer: and we pivoted and so that the the two products right now that are out are for individuals. And the reason we pivoted to consumers is because we're a small company and the businesses started saying, Wow, we really like kind and we want it integrated into our Into Gmail. We want it integrated into our financial systems. We want to it integrated into our document management systems or content management systems. And we said, great, we can do it. Just you know, sign on the dotted line. And of course, companies wouldn't do that. They wanted us to build all this integration to all these integration tools. But without a contract, it's it's very hard for a small company to start doing all sorts of work without a guaranteed payoff at the end. So we started, we went to consumers so that we could at least start generating sales and revenue. But in the
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: fall, we're launching our first foray into business. It's gonna be called kind shared. That's where you create a collection and you can share it, say with A small group of people. So for example, I have a a collection of to do with my family genealogy. I'm not putting it up on Google, but I want
Martin Wheatley: Yeah.
Jonathan Schaeffer: to share it with my mother and my brothers, right? And my immediate family. And so kind will work where it will can create a collection that is shared and private only to the people that you permit. And we're working with two companies right now they're roughly ten to twenty people and it's the same thing where for example one is they're working on contracts and so create a collection this is the customer they've worked with this customer for many years in this collection. You can drag and drop in all of the contracts related to this customer. And now, when you ask questions, you get answers that are 100% guaranteed only to do with this customer.
Martin Wheatley: Mm.
Jonathan Schaeffer: None of the none of these contracts they want on the to put into the cloud. And
Martin Wheatley: Yeah.
Jonathan Schaeffer: so you can have multiple people who are working on this contract. with your own private collection getting answers only to do with with that particular data set. So we're going now that we've established ourselves, we're moving into the business world sometime
Martin Wheatley: Mm.
Jonathan Schaeffer: in the fall.
Martin Wheatley: I mean that that that's that's definitely, you know, having talked to businesses myself, it's it's definitely a key area. That ever everyone's having the same conversation. Where's the data going? How's it been secured? So so building that, like you said, if you can just hone in for the client example on a contract that you can work together and keep it use the local AI but not send it anywhere. That's that's it's a really interesting It's a really interesting product that you're developing, I think, in terms of, you know, letting people trust the process without giving away their their IP somewhere else, really.
Jonathan Schaeffer: Right. So I you've mentioned the word trust several times and I think I I've got to weigh in on the word trust.
Martin Wheatley: Mm.
Jonathan Schaeffer: You know, I'm I'm naive and again I grew up on Star Trek, so
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: blame Gene Roddenberry, but I I think AI is only gonna succeed if people have trust in AI.
Martin Wheatley: Mm. Mm-hmm.
Jonathan Schaeffer: And we know that trust is hard to earn, but it's easy to lose. And there have been multiple studies in Canada and the United States, and I I can't speak for the UK, but I'm sure that the studies are probably similar. That public trust in AI is pretty low. And it's not a surprise. There's been a lot of negative stories about AI. There've been dystopian stories. There's there's these you know giant wars going on between, you know, trillion dollar US companies jockeying for the lead in the AI leadership and billions and trillions of dollars being spent and environmental impacts and hallucinations and everything. There's just a lot of negativity about AI and people just Don't trust it. But if you want to adopt AI, people need to trust it. They won't take something that they have doubts about. And so most people are not knowledgeable about AI. The idea, part of the motivation behind kind, is to put something that's Safe, secure, private, accurate on your desktop so that you can build trust, that you can ask questions to get answers and say, hey, that's right. And not worry about
Martin Wheatley: Yeah.
Jonathan Schaeffer: hallucinations.
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: Trust is important. And I believe that AI. Is going to transform this world, and if we work together, AI can transform this world in amazingly positive ways. Just think of what it can do in healthcare as
Martin Wheatley: Yeah, yeah.
Jonathan Schaeffer: but one example. But there's a lot of resistance, and I don't think that the AI companies and governments and the educators are doing enough to help. reverse the negativity and start spinning a more positive light on what AI can do and I think it it y there's no simple solution, but starting at the grassroots, at the individuals and the small companies and working your way up is one strategy for trying to to build trust. And I'm hoping that kind is a product that in some way, small way, helps create trust in AI.
Martin Wheatley: Yeah. It it definitely the the the messaging could you just need to look at the messaging around data centers. Just that one one pie well obviously it's a large piece of the infrastructure, but how the whole messaging around that, certainly from the UK watching the US, is y you know, you can you can see how the messaging could be done could have been done a lot better. But I mean just just just quickly just kind of pivoting onto your your book. that you that you wrote, the games that computers and humans play. I it was that with this thought in mind about building that I don't want to say PR, but more of a positive message around around AI and the use of AI. I mean I've not I've not read it yet. I said every every time I speak to somebody like yourself, I always had a book onto my reading list. But is that part of the reason you wrote the book?
Jonathan Schaeffer: the reason I wrote the book is because over the years I've given dozens as well, dozens, maybe hundreds of lectures about AI, and
Martin Wheatley: Yeah.
Jonathan Schaeffer: many of them to to the public.
Martin Wheatley: Yeah.
Jonathan Schaeffer: And I believe that building trust in AI in part is involves educating people. We we need we as a society people need to understand AI. Not at the algorithmic level, not at the computer science level, but to understand what AI is and what it is not. And to understand that maybe all these dystopian scenarios that people talk about in the media are perhaps they need to be put into place as to, you
Martin Wheatley: Mm.
Jonathan Schaeffer: know, how realistic they are, for example. And so I've spent my whole career using games as an application for my AI. You mentioned the checkers program, And and poker, but I've done a lot more in games. And I built a reputation doing AI and always using games. Yes, I could have done other perhaps more interesting applications like medical applications or engineering applications, but I did games because I like to have fun in my research. And
Martin Wheatley: Mm.
Jonathan Schaeffer: so when it comes to explaining, when I I decided when I wrote the book, when it comes to explaining just the high-level concepts of AI, I used games.
Martin Wheatley: Yeah.
Jonathan Schaeffer: And so the purpose of the book is to go through and give you a gentle introduction, no formulas, no deep mathematical insights. As to how AI works in a high-level conversational way, using games like chess and checkers and poker and tic-tac-toe and other things like that, so that people can easily grasp the ideas. And at the end of the book, I hope people now have a high-level understanding of what AI is and that they realize when they get to the end that you know things like Chat GPT, when they see how it actually works. They're gonna be sh you know stunned that wow, I'm I'm amazed that it even does as well as it does given the fundamentals that are behind it. And that wow, now I actually understand why they get these stupid hallucinations. And
Martin Wheatley: Yeah.
Jonathan Schaeffer: so that they have a a much better appreciation of what AI can and can't do, and that a lot of what people are Focusing on in the doom and gloom literature is really off base. It's not that there, I'm not trying to say there aren't dangers with AI, there are. But many of these dangers are real, you know, really remote. It's it's like, you know, there's there's a chance the asteroid could you know hit Earth and destroy humanity. Yes, there are doom and gloom scenarios, but We have to put faith in humanity that there's a lot of people who believe that AI for the good is real and that we have an opportunity to do good and we are working in that direction. And the bad scenarios, well hypothetical, aren't real for the foreseeable future. And given time, I certainly hope that they're going to be completely unrealistic.
Martin Wheatley: Yeah. Yeah. I I'm an optimist, so I'm I'm very much Yeah. You've gotta be. You've gotta be.
Jonathan Schaeffer: I I'm an optimist too. We have to be optimistic, right? I I just wanna put this into perspective and that is You know, when nuclear technology came around in you know after World War Two, that's before your time and my time, I'm very happy to say. You know, there were a lot of doom and gloom scenarios. But the world got together and figured it out, and now we have nuclear regulation, and there's been no, you know, nuclear bombs gone off in the last what seventy-five years, whatever it is.
Martin Wheatley: Mm.
Jonathan Schaeffer: And okay, so there's two or three bad actors in the world, but the rest of the world works together to ensure that this technology is deployed in a safe way. way. And AI isn't quite as extreme as blowing up the world, but I believe we will work together in the same way to create a safe environment for deploying AI to the benefit of humankind.
Martin Wheatley: Mm. Yeah, good. Fantastic. Right. We shall move on to the kind of second second part of the show with three three key takeaways. So this is where I just ask any all all the guests just to give three takeaways that the listener can can take away. it could be anything. It could be a book recommendation, could be a yoga pose, could be anything you want. What what what
Jonathan Schaeffer: No yoga poses, sorry. the first takeaway is obvious to me. I already said it. I want people to think of AI as augmented intelligence. Think
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: of AI, and that includes Chat GPT and Claude and Gemini and and whatever, as being your assistant, your apprentice. If you're a lawyer, it's your articling student. If you're a doctor, it's your intern. If you're a professor, it's your graduate student. You're in charge. You have a job to do, personal or professional. Use AI as a tool to inform you, but always regard the AI as your underling, your your student. And you're in charge, which means you make the decisions. If you
Martin Wheatley: Mm.
Jonathan Schaeffer: like what the AI says, use it appropriately. If you disagree, do so. And don't use it. And if you watch Star Trek, I hate to come back to it, but you know, Captain Kirk did not always take the computer's advice. There's a human element. There's a we have things that these AIs don't, and we need to take advantage of who and what we are. Humans are empathetic. Computers are not. They're just cold, impersonal calculating machines. Use our strengths. We should be in charge and use AI as an incredible tool to make us more productive to. And that may mean getting things done faster to a higher standard. That does not mean going to Chat GPT, ask a question, copy and paste it into a document, and you're done.
Martin Wheatley: Mm.
Jonathan Schaeffer: If you're a professional, you have professional standards, professional ethics, you're in charge, you must take responsibility. Anyway, AI is augmented intelligence in my books. A second takeaway. Is that The world of AI is in a fraught situation now right now where all the money is going into a investing in AI technology that is fundamentally flawed. And I'm worried about that. Now
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: this is not a doom and gloom s scenario, but The AI gets better with the current technology we're using. The AI gets better as you feed in more data. And therefore, there's this gold rush of a handful of major, I guess, all US companies, to build bigger and better models, which means getting more data, building bigger. more powerful computing facilities. And the flaws that are in the current models are fundamental. They're not going to go away. A lot of that money I feel should be invested into research, into better ways of doing it, where perhaps there are no fabrications, there are no hallucinations, where the AI
Martin Wheatley: Yeah.
Jonathan Schaeffer: Better understands and is better better able to interact with humans and add more value. So I think that the trillion dollars we're seeing a year, something like that, going into data centers, is a poor investment. We all know the history of technology says that when you get economies of scale. Things get smaller, cheaper, more powerful, and that a lot of the data center builds might in a few years become, you know, obsolete facilities because we we perhaps don't need all that those resources. So I'm a little concerned, I more than a little concerned, that a lot of the AI power is being concentrated into a very a few very large companies and we should all be concerned about that. AI has to deployed
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: for the democratization of knowledge to the benefit of the world, not just to the benefit of a small number of of companies that happen to get there first. So that concerns me.
Martin Wheatley: Yeah, it's an interesting take. I watched an interview with Ilya Sutziva, if that's how you pronounce his name, and he he mentioned he mentioned the same thing rather than building out the infrastructure and the masses of data centers and everything like that, that the money should pivot to research, which I think is probably the reason why he left OpenAI back in where it was a couple of years ago now, wasn't it, and set up his own his own his own lab or whatever, but yeah, it's it's it's definitely that that perspective is is echoed, I think. By minority I think at the minute, because everyone, like you said, is is rushing for the gold rush on it.
Jonathan Schaeffer: what I like to s to say is that we There's a race and there's a handful of big players in the race. And the only thing that
Martin Wheatley: Mm.
Jonathan Schaeffer: matters with those big players is who finishes first. Because when
Martin Wheatley: Yeah.
Jonathan Schaeffer: you get across the finish line, there's this mythical thing called AGI, Artificial General Intelligence. And it's the pot
Martin Wheatley: Mm.
Jonathan Schaeffer: of gold at the end of the rainbow. And whoever gets there first supposedly is going to get these trillion dollar payoffs, right? But the right way of thinking about this race is these companies that are racing to the finish line, they're on steroids. Because they're not addressing the accuracy problem. They're not addressing the safety problems. They're not addressing the explainability problems. These AIs, they'll give you an answer, but they can't explain how they got the answer. Explainability is a big issue. Transparency is a big issue. But the only thing that's important for these companies right now is to go faster. to get to that finish line and a lot is being left undone and
Martin Wheatley: Yeah.
Jonathan Schaeffer: I think that's the mistake. And the last, the third takeaway is I go to academic conferences. I can honestly say that of the hundreds and thousands of people I've met over the years, as researchers, as scientists, etc., every one of them believes in AI for the good. And that's
Martin Wheatley: Mm.
Jonathan Schaeffer: what we're working for. And we all believe that the potential For AI to revolutionize in a good way the world, literally, everybody in the world to improve their quality of life. We believe in it and we believe it will happen. And yes, there are bad scenarios that could happen. But we shouldn't get dragged down and have hypothetical bad scenarios obstruct the real goal, which is to make this world better, safer. more equitable for literally everybody around the world. And and we need to be positive about it. And yes, a healthy dose of concern and action on those concerns is absolutely appropriate. But the media in particular love to pick up the negative stories and don't talk about the positive stories. And so We need to be every chance we get, we need to be positive and reinforce that this is AI for the good.
Martin Wheatley: That is a fantastic note to end on. Optimism at the end is always a good way to end. Thank you very much for that, Jonathan. That was that was very interesting. It i it's fascinating to hear what you're building, the the the thought process behind it and the benefit that it that it can do. So thank you very much. Where where's the best place for people to reach out to contact you to find you?
Jonathan Schaeffer: I'm on LinkedIn, so it's I'm easy to find. And of course you can also go to Sincera S Y N S I R A dot com and read more about us. W you can see my manifesto that's on our our page and you can see the the the products that we have and I hope that if nothing else out of this people are now a little bit more aware about privacy and maybe maybe moving everything to the cloud isn't necessarily in your best interests.
Martin Wheatley: Yeah, no, it's been very useful. Thank you very much. And we will link the the your website, your LinkedIn, and the link to your book on on Amazon in the in the in the description and everything. If you have found this useful hit follow or subscribe if you haven't already and a review or rating genuinely helps the show and I do appreciate it and it keeps my mum happy. There is a growing number of resources at theaileaderlab.com ⁓ for leaders who are looking to start implementing AI in their workplace. The link will be in the show notes. ⁓ Until next time. Bye bye for now.
Martin Wheatley: Mm.
Jonathan Schaeffer: what we're working for. And we all believe that the potential For AI to revolutionize in a good way the world, literally, everybody in the world to improve their quality of life. We believe in it and we believe it will happen. And yes, there are bad scenarios that could happen. But we shouldn't get dragged down and have hypothetical bad scenarios obstruct the real goal, which is to make this world better, safer. more equitable for literally everybody around the world.
Martin Wheatley: Hello and welcome to the Frontier Leader Show. I'm Martin Wheatley and each week I speak to leaders, coaches, practitioners and technologists working out what good leadership looks like in an AI-shaped world. While you still have the app open and in your hand on your phone, please hit follow or subscribe so you don't miss out on the next episode. And if you love the content in this conversation. Then if you head to www.theaileaderlab.com, I have a growing number of resources about how you can start to lead with AI. But today's guest we are speaking to Dr. Jonathan Schaefer. Now Jonathan has been at the forefront of artificial intelligence for more than 40 years, an AI researcher, entrepreneur, author, and inventor. He's dedicated his career to building practical, trustworthy AI systems that solve real world problems. He's the inventor of Kind, a privacy-first AI platform that helps professionals and organizations search, analyse and gain insights from their own curated knowledge base. And his accomplishments include leading the development of AI systems that mastered checkers or drafts, if you're in the UK, and competed against world-class poker players, both recognized by the Guinness Book of World Records, which is a feat in itself. So Jonathan, welcome to the show.
Jonathan Schaeffer: Thank you very much, Martin. It's a pleasure to be here.
Martin Wheatley: And I think this is genuinely the first time I've ever ever spoken to someone in the Guinness Book of World Records. It used to be it used to be there used to be a television programme on when I was a kid called Record Breakers. And I used to love it every week, you know, that they'd go out and watch people try and break records. So fantastic. Absolutely.
Jonathan Schaeffer: watching the break records wouldn't have been quite as exciting, I think, as some of the stuff
Martin Wheatley: Ha ha.
Jonathan Schaeffer: that made it to T V.
Martin Wheatley: Yeah, maybe. But just for for people who don't know you and obviously I I gave a bit of an intro there, but you've obviously, like we said, been in AI for over forty years. You know, how how how has your kind of career what's your career arc been and how have you seen the the overnight success of of AI develop over the last over the last forty years?
Jonathan Schaeffer: Well, I've been around since the dinosaur age. so I've seen AI evolve from systems that were heavily I'll call it curated by humans, where the humans would supply all the knowledge. These were very brittle systems because humans are not very good at explaining, for example. How do you do things? How do you drive a car? How are how, as a doctor, do you come up with a diagnosis? These systems called expert systems required humans to precisely specify all the information necessary for these systems to operate at a, well, by today's standard, a mediocre level. But over the 40 years, what's been happening is actually really. Amazing. We think of the human brain as this incredible machine, which it is, that has this vast amount of knowledge, and we've learned so much. And it seems natural to extract that information and give it to our computer underlings, our computer students, if you will, these programs that are trying to learn how to be good doctors, good drivers, good good authors. And over the 40 years, what's happened is that human knowledge has become less and less important. And in many fields
Martin Wheatley: Just because that's important to add in.
Jonathan Schaeffer: of AI right now is almost irrelevant because we let the the AI discover and learn on its own. And to me, that's been something I hadn't expected way back when, when I started. it's the path towards Computer discovery has has just slowly and methodically moved in that direction. And then the world changed. The world changed on November 30th, 2022, and you you're gonna say, what a geek. Why would you remember that date? Well, it's because
Martin Wheatley: I would have to.
Jonathan Schaeffer: I had seen ear earlier versions of Chat GPT, and quite frankly, they were amusing, but they, if you don't mind, I'll put in a technical word, they sucked. Okay. But on November 30th, 2022, it was like a technological revolution to me. Because these algorithms which didn't perform particularly well, like literally overnight, did amazing things. And I was stunned. I was professionally stunned, but also I was stunned at at the world's reaction because like literally overnight everybody was playing with Chat GPT. And in many ways it was predictable because it was following a clear path that we knew about. But It's this hockey stick phenomenon where something happens and it doesn't go in this straight linear line, it leaps
Martin Wheatley: Okay.
Jonathan Schaeffer: up, and that's exactly what happened. And of course, since then we're seeing something that is unprecedented in human history. We are seeing technological innovation over the last four years at an unprecedented rate. The amount of money that's going into AI. research that's happening, the infrastructure that's being invested in. It's all staggering. And there's no precedent for this kind of degree of innovation happening on this, relatively speaking, very small timescale. So I'm stunned at what's happened and yet as you'll see as we talk about it, I'm also very concerned about what's happening.
Martin Wheatley: Hmm. And the elements I think you might be well, you'll correct me if I'm wrong, but might be concerned about, which I think a lot of people are concerned about, is is the element of trust and privacy around how these models, certainly how the cloud models with, you know, Chat GPT, Claude, Gemini, whichever one to use. perhaps we can pivot slightly into to talking about what I obviously mentioned that you've you've the founder of Kind and and what you're building out there. 'Cause I think that answers quite a lot of the questions and how people can use it in a in a in a better, safer way.
Jonathan Schaeffer: Yes, so thank you for asking. when I started in AI as a student many, many decades ago, I had a vision for AI and this will date me because when I was young I got addicted to watching Star Trek. And in Star Trek there's a computer, right? Spock Captain Kirk would talk to the computer, and the computer had no personality. The computer never chatted with Kirk, right? You would ask it a question, or Kirk or Spock would ask it a question, and it would go off in compute and it would give a fact-based answer. And if it couldn't answer the question, it said so. And in my mind, that was my model for AI. That's not the AI that we have today. There's gaps. So when you look at what ChatGPT and all the other LLMs, large language models like ChatGPT, Gemini, Claude, et cetera, their fundamental algorithm is flawed. There's a certain amount of what I'll call coin flipping that goes into it, which means that answers are non-deterministic. You just can't You can't predict what the answer is going to be, and the answer will change. But because of this coin flipping that goes on, These programs make mistakes. They make factual errors. They fabricate. And it's really cute. We don't call them errors. Why don't we call them errors? We call them hallucinations.
Martin Wheatley: That's the one.
Jonathan Schaeffer: this is part of the anthropomorphization that's going on with these programs. We we chat with them. No, we're not chatting with them. We're we're asking questions. I don't believe we should be anthropomorphizing these programs. We should be using them as tools. We should be using them as tools to do a different kind of AI, not artificial intelligence, but augmented intelligence. We want to use these as tools, like Kirk and Spock did, to augment us. To enhance our decision-making ability. We're in charge, we're using the computers to inform us, advise us, and we ultimately take the decision as to whether to use it or not. And we have to take the responsibility. So these programs have problems and The problems go deeper than just accuracy. They're insatiable for data. The current models get better with more data. And so these companies have a vested interest in getting data from you. And data is multifold. One is that we put our files into the cloud where these companies can see them. And read them and use them to train their AI. I object
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: to that. We can use these tools to ask questions. Even if we
Martin Wheatley: We can use these
Jonathan Schaeffer: don't put our files up there, we can still go to ChatGPT or Claude and ask questions. Asking a question is data. It tells these companies how you're using the tool and what kind of information you're looking for. That's more data. There are things that I do that I don't mind putting into the cloud. There are things that I do that I don't mind asking questions of Claude, for example, and knowing that whatever I interact with this program, it's using it, it's gathering data on me. They say there's no free lunch, and that's exactly what's going on here. I can use these free tools,
Martin Wheatley: Now
Jonathan Schaeffer: but they're gathering data from me. They get value by giving me free access.
Martin Wheatley: Yeah.
Jonathan Schaeffer: But there are things that I don't think belong in the cloud. There are things that are personal, they're private to me. For example, I have a directory on my computer
Martin Wheatley: And go back.
Jonathan Schaeffer: here with my medical history. Do I really want to put that into the cloud? Do I really want one of these LLMs to see it and train off of it? No. I have another directory with all my financial information and investments. It's my information, it's not theirs. I have directories with my intellectual property, my ideas. I recently wrote a book. Do I want all that in ingested into these large language models? No. I believe in privacy. Some things don't have to be private, but some things do. So after 40 years of being a professor, I've had a great ride. I started talking the talk about the problems I see with AI today, with hallucinations, with lack of privacy, there are security problems, there are safety problems with chat. there are issues of advertising that I object to, there's issues with monetization of individuals that I object to. I also am have Concerns about all the data's data center mania that's going on. So I can talk the talk, but I decided to walk the walk. And so I left the university and I formed a new company. It's called Sincera Software Solutions. And we've released our first two products, which were called Kind. And what they are is desktop AI. Okay, that doesn't sound sexy, desktop AI. But the idea here is that. These companies want me to move my data to the cloud to get these AI processing. I'm flipping the script. I want to move the AI from the cloud to my desktop so that I can get my processing. So, what kind does is it's a walled garden on your desktop. And what you do is you just drag and drop files into kind. And what kind is, is it has sophisticated AI technology that does the analysis of your files on your desktop and allows you to do what the chatbots do. You can search your file, you can search your files, you can ask questions about your files, and it only works on the data that you give it. So, if you, for example, if I put in all my financial information into kind, there's a wall around it. And if I now ask a question about my financial information, the answer only comes from my data. There are no external influences that affect the answer. There's no hallucinations because everything is fact-based from my data. So that's in a nutshell what the idea is, is kind puts on your desktop your own private AI, which you can do ask questions and do searching. It's vendor agnostic. We don't train on your data at all. there's no internet resources used so you can turn off the internet and you still have high quality local AI processing. It's safe, it's secure, it's private, no hallucinations. That's what it's all about.
Martin Wheatley: Yeah, and that and that's so it it's is it an equivalent to being a personalized local, secure kind of notebook LM? So to speak, or so it's not called Notebook L now, Gemini Notebook, whatever it's called now. I'm assuming it it's it's that kind of it's that kind of idea that you can drop your files, your documents in there and then you can interrogate just those, but obviously it's it's secured locally. That's summarizing.
Jonathan Schaeffer: Correct. So No Notebook LM is is a good example, but the problem
Martin Wheatley: Mm.
Jonathan Schaeffer: with for me, the problem with Notebook LM is to use it, I have to put my files into the cloud. That means
Martin Wheatley: Mm, mm.
Jonathan Schaeffer: Google sees them. So Obviously, our tool is not as powerful as Notebook LM. We don't have, you know, a thousand software developers like Google, but what we do have is a powerful desktop AI that sits on your desktop. You don't move your files to the cloud, nobody sees them, and you get the benefits of leading edge AI.
Martin Wheatley: Yeah. And I guess I guess there's a lot of if you follow the the AI hype on YouTube and everything, there's all this noise around second brains and building things out with Obsidian. I'm sure you've I'm sure you've seen this in in in your world more more than more than once, maybe. But I guess the again the difference with this is you can connect Claude or whatever you can to Obsidian, which obviously is then your files on your drive, but They still get sent to and from to and from Claude, don't they? Whereas yours is purely local. Yeah.
Jonathan Schaeffer: Correct. And and that's the important part. it's
Martin Wheatley: Mm.
Jonathan Schaeffer: the privacy. One of the things that I've seen over the years is that We've lost the notion of privacy.
Martin Wheatley: Mm.
Jonathan Schaeffer: People, I I think it started with social media. People said, Wow, this is really cool. I have a social media account. I can upload all my photos. I can say things. And the companies are very happy for that because they have data that they can use and learn from and take advantage of. But When cloud computing came along, now it was hey, give us all your files. And yeah,
Martin Wheatley: Yeah.
Jonathan Schaeffer: guess what? You get free backups and free storage and all these other good things. But again, there's no free lunch. They're getting something from it, and they get to see your data. And when AI came along, sorry, not AI per se, but when ChatGPT came along and all of a sudden there was this rush towards massive investment in AI. Everybody wanted your data, so we're you're encouraged now to move your files even even more so now to the cloud, which again just it's a loss of privacy. And again, the point here is that some things are I'm okay with, but other things I'm not. And so I don't want to disadvantage the individual. By forcing them to move their files to the cloud. If you want something private, get many of the advantages or most of the advantages of internet-based AI, but put them on your desktop where
Martin Wheatley: Mm.
Jonathan Schaeffer: you maintain your privacy. The world has moved towards this
Martin Wheatley: Yeah.
Jonathan Schaeffer: view that we should be moving all of our data into the cloud. And there are only a handful of large companies who offer the kind of cloud infrastructure that that give you vast amounts of storage and access to leading edge tools. And that's making these companies richer
Martin Wheatley: He's amazing.
Jonathan Schaeffer: and more powerful. And I'm not sure that that's the right way to go.
Martin Wheatley: No, I think I think you're right with the the whole mass rollout of social media. I think it has desensitized people to to giving away their data. Constant status updates, photos, share and like you said, when the cloud came along and took away the problem of managing stuff locally, yeah, people are just used to doing it, aren't you? But this is a very a very secure way of of managing your your sensitive data, definitely.
Jonathan Schaeffer: But you know, it doesn't even have to be what you would call sensitive data.
Martin Wheatley: Mm, mm.
Jonathan Schaeffer: Th I I I I think that I have data which I wouldn't regard as sensitive, but there's still a sense of ownership and I I just don't want other companies or companies knowing about some of my personal stuff. I like to collect certain things. I have a
Martin Wheatley: Mm.
Jonathan Schaeffer: Lots of information that I've collected over the years about my particular interest and it's mine. So and I'll tell you one thing that's interesting about some of the information I I've collected, which is which caught me by surprise when I started developing kind. I collect books, okay? Very simple. And
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: in 1995, basically the World Wide Web sort of came along and people put things on the web. And I would download on on I would put onto my desktop book catalogs or auction results or interesting articles because I read the stuff, I collect the stuff, I use it to to enhance my knowledge. Well Companies come and go. Companies purge their website. I discovered that a lot of the knowledge of the stuff that I have is no longer on the web. In other words, I can ask questions on the internet and it can't find the answer.
Martin Wheatley: Mm.
Jonathan Schaeffer: But I have the answer on my laptop. And to use powerful AI to get those answers is amazing. But I don't wanna have to take this and put it in the on the web because now Google has access to it and all of a sudden this information ends up g going out there to the world. But in some sense, I wanna keep it. It's mine.
Martin Wheatley: Yeah, yeah. No, I g I understand that completely. Understand that completely. You you've mentioned a you've mentioned a couple of times we've been speaking about enhancing enhancing thinking and and you know, not replacing thinking. And one of the one of the big criticisms, especially with people who are now steadily relying on AI, is it's taken away a lot a lot of that critical thought process. You know, they put in a question, personal or work related, they grab the first answer that comes out of chat GPT, whatever, and then the pass it on up the chain of of whatever. W what's your kind of take on take on that and how products like kind can can help augment the the the human in the loop, so to speak.
Jonathan Schaeffer: So I'm a scientist, ideal of accuracy. If I write a paper
Martin Wheatley: Right.
Jonathan Schaeffer: and it has falsehoods in it, I'm gonna get shot down or embarrassed. The problem with the current large language models is, and I use them, I'll ask questions and they'll give me answers, but I never
Martin Wheatley: Mm.
Jonathan Schaeffer: use them as is. I have to invest the time to verify whether that information is correct. And so you can think of Chat GPT as wow, look at this fantastic time saver, because I ask a question, I've got the answer, I'm done. For me, and I I would hope for most people, particularly professionals. You're not done because you must understand that there is a reasonable probability that information in the answer given back by one of these chat programs is fabricated or incorrect. And therefore, you have to take the responsibility to address it. Now, I I would say that these are incredibly valuable tools for finding things, but you have to take the responsibility. What happens with kind though is I'll I'll give you an example.
Martin Wheatley: Mm.
Jonathan Schaeffer: I worked, I recently wrote a book, An Introduction to AI for non-technical people. And I could write on my my laptop. And when I found some information on the web that was really interesting that I liked, I put it on my laptop into my into my kind collection to do with this book. And over time, I had, here's all my writing, and here are all the references that are in my book, all the documents that I found and all the web pages that were interesting. And now I could ask questions that would only come from my data, nothing else out there, because there's a lot of garbage out there.
Martin Wheatley: Mm.
Jonathan Schaeffer: Therefore, I could ask questions about my book, and I could verify that the answers were correct. But I also discovered something interesting. So just as taking a step back, kind sits on your desktop. I I don't mean to be nosy, but how busy is your desktop? Your computer. During the day, maybe it's busy. It's busy right now. It's doing a video
Martin Wheatley: Mm.
Jonathan Schaeffer: recording. Is it working 24 hours a day? You know, maybe you're running hundreds of agents or doing big number crunching? Probably not. What's it doing on the weekends? It's probably idle. The point is you got a lot of compute horsepower on your desktop or your laptop.
Martin Wheatley: Yeah.
Jonathan Schaeffer: Well, We can harness it because you've put kind on your desktop and we do the AI processing. So in the case of my book, I put in all this information, and before I published the book, I started asking questions because I wanted to verify that everything was correct, et cetera. And I found mistakes. And so what we did is we've added a feature to kind that allows it to look within your data. For factual inconsistencies. Now you can't do that with Google. You can't say, hey Google, here are all my files. Tell me if there's any factual mistakes in them. But we can, because we're on your desktop.
Martin Wheatley: Hmm.
Jonathan Schaeffer: Your data set, you may think that your data, your medical files, your financial files, your book files, your collecting files, whatever your files are, your PowerPoint files. You maybe say, Wow, I got lots of data. You have small data. The amount of data that's on your machine, okay, how many gigabytes is it? Gigabytes is small. Google's dealing with petabytes, which is
Martin Wheatley: Mm.
Jonathan Schaeffer: millions of times bigger, right? So we can run. During the evenings and weekends, and do things with your data that you can't do in the web. And therefore,
Martin Wheatley: Mm.
Jonathan Schaeffer: inside a collection where we've got boundaries on where it's looking, because it's only looking in your data, kind is able to do interesting things that you can't do elsewhere. But we have access to your CPU and you have the resources that allow us to do it. And what who cares if it happens overnight? It's still
Martin Wheatley: Yeah.
Jonathan Schaeffer: a value add, and it gives you more insights, more capabilities, a better understanding of the data that you have.
Martin Wheatley: Yeah. It's definitely an interesting shift definitely from from the cloud based solutions that are there. 'Cause like you said, using the capabilities or the the infrastructure if you like on on your laptop or device in the background when you're not using it as much is it's just an efficient way of using a device, isn't it, at the end of the
Jonathan Schaeffer: But let's think about it this way: it's a sunk cost. You've already
Martin Wheatley: Mm, mm.
Jonathan Schaeffer: purchased the machine. It's just sitting there, it's depreciating. Use your
Martin Wheatley: Yeah, yeah.
Jonathan Schaeffer: resources. But there's an important advantage about that. Because if you're doing your AI processing on your desktop, you're asking questions and getting answers, or you're analyzing some of your data all on your machine, you're not using. Claude, you're not using Chat GPT, which means you're not sending queries to these data centers. You're being environmentally responsible. I'd like to say, you know, the environmental movement says think globally and act locally. Well,
Martin Wheatley: Mm.
Jonathan Schaeffer: with kind, you're removing your queries from those data centers. You're putting them onto a device that you already own and has no cost. You're not paying for tokens. You're being environmentally responsible by not encouraging further growth of this data center mania.
Martin Wheatley: Yeah. Does it does it how does it I I guess what one of the key challenges I've seen when trying to do something, obviously not not solely on my my desktop per se, but is kind of agreeing that the structure of where should I put this note? How should I record this? Does does kind do that for you to some extent, or does the user still have to think, well, that this is this is my project on Books for argument's sake, just coming back to your example. Does does kind take care of that or is it still user driven in terms of the the setup?
Jonathan Schaeffer: So this is what happens. First of all, we never move your files. Your files,
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: they're on your machine. we never move them, we never remove them, et cetera. So all you you do is in kind you create a collection. I might create a collection called say financial.
Martin Wheatley: Yeah.
Jonathan Schaeffer: All you do is drag and drop into that collection any files that you think are relevant. I created
Martin Wheatley: Mm.
Jonathan Schaeffer: a collection for my book, and I just drag and drop into there anything that I think is interesting. I mean, you have curated the data. You're putting
Martin Wheatley: Mm.
Jonathan Schaeffer: into these collections the material that you think is important and relevant. And it could be material that's on your desktop, or it could be material that you get from the web, but you've curated
Martin Wheatley: Yeah.
Jonathan Schaeffer: it, you've put it in there. And kind automatically analyzes it and adds it into its own internal database. But you don't have to reorganize your files or or anything like that. Just drag and drop whatever you think is relevant to that topic and you're done.
Martin Wheatley: Hm. Okay. Interesting. How does it work for obviously 'cause it's not connected to the web. How does it work from does it is it built for businesses as well as personal or is it just personal at the minute in terms of knowledge base?
Jonathan Schaeffer: So originally it was envisioned for business
Martin Wheatley: Mm.
Jonathan Schaeffer: and we pivoted and so that the the two products right now that are out are for individuals. And the reason we pivoted to consumers is because we're a small company and the businesses started saying, Wow, we really like kind and we want it integrated into our Into Gmail. We want it integrated into our financial systems. We want to it integrated into our document management systems or content management systems. And we said, great, we can do it. Just you know, sign on the dotted line. And of course, companies wouldn't do that. They wanted us to build all this integration to all these integration tools. But without a contract, it's it's very hard for a small company to start doing all sorts of work without a guaranteed payoff at the end. So we started, we went to consumers so that we could at least start generating sales and revenue. But in the
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: fall, we're launching our first foray into business. It's gonna be called kind shared. That's where you create a collection and you can share it, say with A small group of people. So for example, I have a a collection of to do with my family genealogy. I'm not putting it up on Google, but I want
Martin Wheatley: Yeah.
Jonathan Schaeffer: to share it with my mother and my brothers, right? And my immediate family. And so kind will work where it will can create a collection that is shared and private only to the people that you permit. And we're working with two companies right now they're roughly ten to twenty people and it's the same thing where for example one is they're working on contracts and so create a collection this is the customer they've worked with this customer for many years in this collection. You can drag and drop in all of the contracts related to this customer. And now, when you ask questions, you get answers that are 100% guaranteed only to do with this customer.
Martin Wheatley: Mm.
Jonathan Schaeffer: None of the none of these contracts they want on the to put into the cloud. And
Martin Wheatley: Yeah.
Jonathan Schaeffer: so you can have multiple people who are working on this contract. with your own private collection getting answers only to do with with that particular data set. So we're going now that we've established ourselves, we're moving into the business world sometime
Martin Wheatley: Mm.
Jonathan Schaeffer: in the fall.
Martin Wheatley: I mean that that that's that's definitely, you know, having talked to businesses myself, it's it's definitely a key area. That ever everyone's having the same conversation. Where's the data going? How's it been secured? So so building that, like you said, if you can just hone in for the client example on a contract that you can work together and keep it use the local AI but not send it anywhere. That's that's it's a really interesting It's a really interesting product that you're developing, I think, in terms of, you know, letting people trust the process without giving away their their IP somewhere else, really.
Jonathan Schaeffer: Right. So I you've mentioned the word trust several times and I think I I've got to weigh in on the word trust.
Martin Wheatley: Mm.
Jonathan Schaeffer: You know, I'm I'm naive and again I grew up on Star Trek, so
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: blame Gene Roddenberry, but I I think AI is only gonna succeed if people have trust in AI.
Martin Wheatley: Mm. Mm-hmm.
Jonathan Schaeffer: And we know that trust is hard to earn, but it's easy to lose. And there have been multiple studies in Canada and the United States, and I I can't speak for the UK, but I'm sure that the studies are probably similar. That public trust in AI is pretty low. And it's not a surprise. There's been a lot of negative stories about AI. There've been dystopian stories. There's there's these you know giant wars going on between, you know, trillion dollar US companies jockeying for the lead in the AI leadership and billions and trillions of dollars being spent and environmental impacts and hallucinations and everything. There's just a lot of negativity about AI and people just Don't trust it. But if you want to adopt AI, people need to trust it. They won't take something that they have doubts about. And so most people are not knowledgeable about AI. The idea, part of the motivation behind kind, is to put something that's Safe, secure, private, accurate on your desktop so that you can build trust, that you can ask questions to get answers and say, hey, that's right. And not worry about
Martin Wheatley: Yeah.
Jonathan Schaeffer: hallucinations.
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: Trust is important. And I believe that AI. Is going to transform this world, and if we work together, AI can transform this world in amazingly positive ways. Just think of what it can do in healthcare as
Martin Wheatley: Yeah, yeah.
Jonathan Schaeffer: but one example. But there's a lot of resistance, and I don't think that the AI companies and governments and the educators are doing enough to help. reverse the negativity and start spinning a more positive light on what AI can do and I think it it y there's no simple solution, but starting at the grassroots, at the individuals and the small companies and working your way up is one strategy for trying to to build trust. And I'm hoping that kind is a product that in some way, small way, helps create trust in AI.
Martin Wheatley: Yeah. It it definitely the the the messaging could you just need to look at the messaging around data centers. Just that one one pie well obviously it's a large piece of the infrastructure, but how the whole messaging around that, certainly from the UK watching the US, is y you know, you can you can see how the messaging could be done could have been done a lot better. But I mean just just just quickly just kind of pivoting onto your your book. that you that you wrote, the games that computers and humans play. I it was that with this thought in mind about building that I don't want to say PR, but more of a positive message around around AI and the use of AI. I mean I've not I've not read it yet. I said every every time I speak to somebody like yourself, I always had a book onto my reading list. But is that part of the reason you wrote the book?
Jonathan Schaeffer: the reason I wrote the book is because over the years I've given dozens as well, dozens, maybe hundreds of lectures about AI, and
Martin Wheatley: Yeah.
Jonathan Schaeffer: many of them to to the public.
Martin Wheatley: Yeah.
Jonathan Schaeffer: And I believe that building trust in AI in part is involves educating people. We we need we as a society people need to understand AI. Not at the algorithmic level, not at the computer science level, but to understand what AI is and what it is not. And to understand that maybe all these dystopian scenarios that people talk about in the media are perhaps they need to be put into place as to, you
Martin Wheatley: Mm.
Jonathan Schaeffer: know, how realistic they are, for example. And so I've spent my whole career using games as an application for my AI. You mentioned the checkers program, And and poker, but I've done a lot more in games. And I built a reputation doing AI and always using games. Yes, I could have done other perhaps more interesting applications like medical applications or engineering applications, but I did games because I like to have fun in my research. And
Martin Wheatley: Mm.
Jonathan Schaeffer: so when it comes to explaining, when I I decided when I wrote the book, when it comes to explaining just the high-level concepts of AI, I used games.
Martin Wheatley: Yeah.
Jonathan Schaeffer: And so the purpose of the book is to go through and give you a gentle introduction, no formulas, no deep mathematical insights. As to how AI works in a high-level conversational way, using games like chess and checkers and poker and tic-tac-toe and other things like that, so that people can easily grasp the ideas. And at the end of the book, I hope people now have a high-level understanding of what AI is and that they realize when they get to the end that you know things like Chat GPT, when they see how it actually works. They're gonna be sh you know stunned that wow, I'm I'm amazed that it even does as well as it does given the fundamentals that are behind it. And that wow, now I actually understand why they get these stupid hallucinations. And
Martin Wheatley: Yeah.
Jonathan Schaeffer: so that they have a a much better appreciation of what AI can and can't do, and that a lot of what people are Focusing on in the doom and gloom literature is really off base. It's not that there, I'm not trying to say there aren't dangers with AI, there are. But many of these dangers are real, you know, really remote. It's it's like, you know, there's there's a chance the asteroid could you know hit Earth and destroy humanity. Yes, there are doom and gloom scenarios, but We have to put faith in humanity that there's a lot of people who believe that AI for the good is real and that we have an opportunity to do good and we are working in that direction. And the bad scenarios, well hypothetical, aren't real for the foreseeable future. And given time, I certainly hope that they're going to be completely unrealistic.
Martin Wheatley: Yeah. Yeah. I I'm an optimist, so I'm I'm very much Yeah. You've gotta be. You've gotta be.
Jonathan Schaeffer: I I'm an optimist too. We have to be optimistic, right? I I just wanna put this into perspective and that is You know, when nuclear technology came around in you know after World War Two, that's before your time and my time, I'm very happy to say. You know, there were a lot of doom and gloom scenarios. But the world got together and figured it out, and now we have nuclear regulation, and there's been no, you know, nuclear bombs gone off in the last what seventy-five years, whatever it is.
Martin Wheatley: Mm.
Jonathan Schaeffer: And okay, so there's two or three bad actors in the world, but the rest of the world works together to ensure that this technology is deployed in a safe way. way. And AI isn't quite as extreme as blowing up the world, but I believe we will work together in the same way to create a safe environment for deploying AI to the benefit of humankind.
Martin Wheatley: Mm. Yeah, good. Fantastic. Right. We shall move on to the kind of second second part of the show with three three key takeaways. So this is where I just ask any all all the guests just to give three takeaways that the listener can can take away. it could be anything. It could be a book recommendation, could be a yoga pose, could be anything you want. What what what
Jonathan Schaeffer: No yoga poses, sorry. the first takeaway is obvious to me. I already said it. I want people to think of AI as augmented intelligence. Think
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: of AI, and that includes Chat GPT and Claude and Gemini and and whatever, as being your assistant, your apprentice. If you're a lawyer, it's your articling student. If you're a doctor, it's your intern. If you're a professor, it's your graduate student. You're in charge. You have a job to do, personal or professional. Use AI as a tool to inform you, but always regard the AI as your underling, your your student. And you're in charge, which means you make the decisions. If you
Martin Wheatley: Mm.
Jonathan Schaeffer: like what the AI says, use it appropriately. If you disagree, do so. And don't use it. And if you watch Star Trek, I hate to come back to it, but you know, Captain Kirk did not always take the computer's advice. There's a human element. There's a we have things that these AIs don't, and we need to take advantage of who and what we are. Humans are empathetic. Computers are not. They're just cold, impersonal calculating machines. Use our strengths. We should be in charge and use AI as an incredible tool to make us more productive to. And that may mean getting things done faster to a higher standard. That does not mean going to Chat GPT, ask a question, copy and paste it into a document, and you're done.
Martin Wheatley: Mm.
Jonathan Schaeffer: If you're a professional, you have professional standards, professional ethics, you're in charge, you must take responsibility. Anyway, AI is augmented intelligence in my books. A second takeaway. Is that The world of AI is in a fraught situation now right now where all the money is going into a investing in AI technology that is fundamentally flawed. And I'm worried about that. Now
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: this is not a doom and gloom s scenario, but The AI gets better with the current technology we're using. The AI gets better as you feed in more data. And therefore, there's this gold rush of a handful of major, I guess, all US companies, to build bigger and better models, which means getting more data, building bigger. more powerful computing facilities. And the flaws that are in the current models are fundamental. They're not going to go away. A lot of that money I feel should be invested into research, into better ways of doing it, where perhaps there are no fabrications, there are no hallucinations, where the AI
Martin Wheatley: Yeah.
Jonathan Schaeffer: Better understands and is better better able to interact with humans and add more value. So I think that the trillion dollars we're seeing a year, something like that, going into data centers, is a poor investment. We all know the history of technology says that when you get economies of scale. Things get smaller, cheaper, more powerful, and that a lot of the data center builds might in a few years become, you know, obsolete facilities because we we perhaps don't need all that those resources. So I'm a little concerned, I more than a little concerned, that a lot of the AI power is being concentrated into a very a few very large companies and we should all be concerned about that. AI has to deployed
Martin Wheatley: Mm-hmm.
Jonathan Schaeffer: for the democratization of knowledge to the benefit of the world, not just to the benefit of a small number of of companies that happen to get there first. So that concerns me.
Martin Wheatley: Yeah, it's an interesting take. I watched an interview with Ilya Sutziva, if that's how you pronounce his name, and he he mentioned he mentioned the same thing rather than building out the infrastructure and the masses of data centers and everything like that, that the money should pivot to research, which I think is probably the reason why he left OpenAI back in where it was a couple of years ago now, wasn't it, and set up his own his own his own lab or whatever, but yeah, it's it's it's definitely that that perspective is is echoed, I think. By minority I think at the minute, because everyone, like you said, is is rushing for the gold rush on it.
Jonathan Schaeffer: what I like to s to say is that we There's a race and there's a handful of big players in the race. And the only thing that
Martin Wheatley: Mm.
Jonathan Schaeffer: matters with those big players is who finishes first. Because when
Martin Wheatley: Yeah.
Jonathan Schaeffer: you get across the finish line, there's this mythical thing called AGI, Artificial General Intelligence. And it's the pot
Martin Wheatley: Mm.
Jonathan Schaeffer: of gold at the end of the rainbow. And whoever gets there first supposedly is going to get these trillion dollar payoffs, right? But the right way of thinking about this race is these companies that are racing to the finish line, they're on steroids. Because they're not addressing the accuracy problem. They're not addressing the safety problems. They're not addressing the explainability problems. These AIs, they'll give you an answer, but they can't explain how they got the answer. Explainability is a big issue. Transparency is a big issue. But the only thing that's important for these companies right now is to go faster. to get to that finish line and a lot is being left undone and
Martin Wheatley: Yeah.
Jonathan Schaeffer: I think that's the mistake. And the last, the third takeaway is I go to academic conferences. I can honestly say that of the hundreds and thousands of people I've met over the years, as researchers, as scientists, etc., every one of them believes in AI for the good. And that's
Martin Wheatley: Mm.
Jonathan Schaeffer: what we're working for. And we all believe that the potential For AI to revolutionize in a good way the world, literally, everybody in the world to improve their quality of life. We believe in it and we believe it will happen. And yes, there are bad scenarios that could happen. But we shouldn't get dragged down and have hypothetical bad scenarios obstruct the real goal, which is to make this world better, safer. more equitable for literally everybody around the world. And and we need to be positive about it. And yes, a healthy dose of concern and action on those concerns is absolutely appropriate. But the media in particular love to pick up the negative stories and don't talk about the positive stories. And so We need to be every chance we get, we need to be positive and reinforce that this is AI for the good.
Martin Wheatley: That is a fantastic note to end on. Optimism at the end is always a good way to end. Thank you very much for that, Jonathan. That was that was very interesting. It i it's fascinating to hear what you're building, the the the thought process behind it and the benefit that it that it can do. So thank you very much. Where where's the best place for people to reach out to contact you to find you?
Jonathan Schaeffer: I'm on LinkedIn, so it's I'm easy to find. And of course you can also go to Sincera S Y N S I R A dot com and read more about us. W you can see my manifesto that's on our our page and you can see the the the products that we have and I hope that if nothing else out of this people are now a little bit more aware about privacy and maybe maybe moving everything to the cloud isn't necessarily in your best interests.
Martin Wheatley: Yeah, no, it's been very useful. Thank you very much. And we will link the the your website, your LinkedIn, and the link to your book on on Amazon in the in the in the description and everything. If you have found this useful hit follow or subscribe if you haven't already and a review or rating genuinely helps the show and I do appreciate it and it keeps my mum happy. There is a growing number of resources at theaileaderlab.com ⁓ for leaders who are looking to start implementing AI in their workplace. The link will be in the show notes. ⁓ Until next time. Bye bye for now.
