LH115 Meet Your New Math(s) Teacher: a Robot Dog, with Jan Liphardt

This podcast is supported by World of Learning conference and Exhibition L and D professionals. You need to be at World of Learning, the leading L and D event happening this 7th and 8th of October at the NEC Birmingham, UK. An unrivalled opportunity to learn, discover, interact and network. Register for free today@learnevents.com this podcast is supported by Synthesia. Training videos shouldn't take weeks to make. Synthesia lets you turn text into professional multilingual training videos in minutes. It's fast, scalable and so simple, anyone can use it. You can write an email, you can make a video.

Hey there, my name is Alex and I'll be guiding you through this training session.

70% of Fortune 100 companies use Synthesia to modernize onboarding, simplify compliance and scale global training. Podcast listeners get an exclusive 30 day free trial. Go to synthesia IO greatminds. Don't miss it. This offer won't last.

Small kids come running when one of our robotic dogs goes to a kindergarten. They love the robot dogs. They wanted to sit and jump and they talk and engage and their eyes sort of light up. They're fascinated by it. And that's of course, very important for teaching.

Welcome to the Learning Hack, a podcast about the people and technologies creating the future of learning. I'm John Helmer.

Now guess what? Learning is cool. Learning is. That's cool. I'm learning.

Learning is fun and knowledge is power.

Knowledge Education. Welcome back. This episode is about one of the deepest wishes of humanity since the dawn of the industrial age. We are tool using creatures and not content with inventing machines to ease the burden of manual labor. We've long aspired to create physical world avatars of ourselves, entities with skills of thought and action like our own, subject to our will. When we dream of the future, we dream of robots. It's not a dream anymore. They're here. Modern manufacturing is more and more automated. We hear of dark factories in China and entirely staffed by robots to autonomous taxis flying the streets of San Francisco. And in the conflict in Ukraine this July, it was widely reported that Russian troops surrendered to the first entirely robotized fighting unit. The robots in this last example were pretty crude, not at all like something from Star Wars, a bit more like Wall E. And though robots are getting better very quickly, there are still limits to what they can do within human environments. However, thanks to recent developments in generative AI, the use as a tool for learning is closer than most people think. I wanted to talk to someone working on this important frontier of human knowledge. Someone who not only knows stuff, but does stuff. And who better than Jan Lipphart? Hack facts.

Jan Liephart is the founder of Open Mind, a platform designed to help people engage across divides through evidence based conversation training. A professor of bioengineering at Stanford University, his research spans data science, cancer biology and human behavior. He previously led teams at Google X and the Chan Zuckerberg Biohub. John's work integrates scientific rigor with social impact.

I was really knocked out by this conversation and can't stop talking about it to my friends. Jan completely changed how I think about the whole area.

John, you are completely wrong.

And the core of that changed awareness is to do with the phrase embodied AI. It's one thing feeding questions into your computer or your phone for an LLM, but what changes when that dialogue is happening with a humanoid Android or even a robot dog, say, in the room with you? What new vistas of fruitful interaction, of learning open up and what are the dangers? Jan, it's a great pleasure to have you on the podcast. Welcome to the Learning Hack.

Thank you very much. Looking forward.

So you're a teacher and academic, but also an entrepreneur. And as well as straddling that divide, your work with open mindsets are very interesting. Interesting place where AI intersects with robotics to build smarter robots. To begin with that, can you tell us what is the problem that you're trying to solve with open mind? What is the kind of, what is the wellspring of searching for a correct metaphor? What drives that company?

Well, a lot of us think about large language models and AI as a tool for helping us with our homework or helping us to code or helping us generate images or movies. But when you think about other implications of AI and generative AI, if something, if, if a technology is able to fluently write computer code, then it's likely that that computer code can then be used to directly control physical things like robots. If a computer is able to write computer code and you're able to translate what the computer is writing into physical movements, then you can start to ask yourself if large language models are good at controlling robots. And the answer is yes, they are. And what this means is that you can now literally shake hands with some of the large language models. We give them a physical shell that allows them to autonomously explore the world. So the large language model can now stand up literally and walk out the front door and go have a walk in the park and look around and talk to people, because it's now able to move around the physical world.

This is a Fantastically interesting area and we're all enthused by the details. So I hate to get into the negative so soon, but the AI side of that combination of AI and robotics on the AI side, it's really surged ahead in the last three years to the extent that we're all quite familiar with LLMs now. Massive take up of that, especially with ChatGPT for instance. So we know about both their fantastic potential and some of the pitfalls. Now Andre Karpathy was talking recently about Genai's jagged intelligence. He says it suffers from hallucinations and terror grade amnesia. I hope I'm pronouncing that right. Prompt injection risks and so on. So I'm curious about what he describes as cognitive deficits play out when you bring gen AI to bear on a physical world instantiation like a humanoid robot or a robot dog. In other words, when you put AI, well, specifically LLMs, all these drawbacks into a physical world thing, what are the potential risks of these cognitive deficits immediately raised when that happens?

Well, so you're absolutely right to bring that up. It's obviously a major challenge and it's also a really interesting topic to dig into. But before we go into that, let me tell you as a professor and teacher, I've been grading undergraduate homework for 20 years. And the number of problem sets and essays I've graded generated by humans that have suffered from incomplete sentences, incoherent arguments made up references is astronomical. Every single one of us on this podcast, I'm sure when we were undergraduates, we wrote at least one essay from for our English class at three in the morning. And when we woke up the next day, we have no memory whatsoever of what we wrote in that essay. And then our poor professors and teachers had to try to read that and make sense of it and grade it. So just from the perspective of someone who's graded thousands of undergraduate homeworks, the amount of like hallucinations I've had degrade generated by humans is, is, is, is very large. So getting things wrong is not specific to large language models. It's, it's a general challenge for all of us and of course people. Humans have built a very complex infrastructure for collectively deciding what is right and wrong. A system for putting you in jail if you get it wrong, juries for voting up or down on crimes you may have committed. So problems relating to decision making in the real world are not specific to large language models, but it's something that humans have had to contend with for, you know, thousands of years. But you're absolutely right.

It's a great point that LLMs are in some ways slightly crap, but humans are much crapper. This is true. On the other hand, what I'm getting at is when you come into the physical realm, you look at autonomous vehicles, for instance, one person gets killed by an autonomous vehicle. That sets back the public acceptability of autonomous vehicles for years. And all the time people are promising it's going to be we're going to have self driving cars next year and it doesn't happen. A lot of this is to do with the fact that the safety has to be much higher. Meanwhile, in the UK at least 10,000 people a year are killed by human drivers on the roads and nobody seems to remark on the fact at all regularly. And I'm sure it's even more in the States because it's a much bigger country. You know, doesn't the barrier have to be higher when it is a machine that is performing a function that has life or death implications?

You're absolutely right. It's fascinating to me. I was just looking at the road Deaths in the US and American human drivers kill 37,000Americans a year in car accidents. The most recent numbers from the robot car company Waymo that will drive you around San Francisco without a human driver is that they're typically 20 times safer than if a human were to be driving the car. So I as a parent am much more comfortable with the robot driving my son to the airport than I am with human driver because it's 20 times safer. And. But you're absolutely right, there's a big difference between safety as defined quantitatively and public perception. So you the news will jump on a situation where a bike, a cyclist has been hurt by a robot car. But we'll generally then not put that into context of after millions and millions of miles driven, it turns out that robotic cars are 20 times safer than human drivers. So I completely agree with you. It's from a parent or technologist perspective, it's very important to me that as we deploy large language models into robots, that we do so in the safest possible way. And that means spending a lot of time on things like how can we understand what the large language models are doing and how do we guardrail their behavior.

On the positive side, can you talk a bit about what OpenMind dropped in May, which is called multi agent endpoint, I believe.

Oh yeah. So one use case for humanoid robots in the home is to assist old people that live by themselves. And another use case we're looking into is to help blind people with seeing eye dogs. And both those use cases involve good decision making by the robot and they also involve dealing with sensitive healthcare data about that individual. And in the US at least, and I'm sure this is also true in the uk, dealing with healthcare relevant data brings with it a whole set of extra rules and regulations relating to data privacy and things like that. And so what we do for robots that are deployed into healthcare settings, we provide a endpoint that is HIPAA compliant. So it complies with the rules and regulations for computing on healthcare data. And it's designed to be a convenient endpoint for robots in healthcare settings to think and do so in a way that is compliant with the privacy regulations.

So by endpoint you're talking about what about the dog?

Oh, no, this is a cloud endpoint. So a lot of the compute we do is not necessarily taking place on computers inside the robot.

Okay. It's essentially.

On cloud computers. That's important because it allows us to reduce the price of the robot. It doesn't need a big supercomputer on its back and it allows us to do a lot more compute per unit time than we could do in an edge setting on the robot itself.

Okay. Also, what I saw in that announcement was it seemed to be different types of LLMs. You would have something which was doing a sort of spatial awareness thing and then another bit. Could you talk a bit about that? Because I thought that was quite interesting.

Oh, yeah. It's basically like a committee of LLMs.

Committee, yes.

And they're arguing and there's always at least three. There's one really little simple LLM who has complete command override authority. And we call that the emergency responder. So if the dog is aware that, for example, it's about to fall down a flight of stairs.

Yeah.

That information goes to this committee of large language models. And then the first responder LLM sees that, raises their hand and says, hey, I'm going to override everything and return tokens in 300 milliseconds. And that helps the robot quickly avoid important potential problems. And the core LLM in the middle, that works on a two second response time. That's the one doing things like speech. And then there's a mother LLM. And think of the mother LLM as your mother walking behind you, looking over your shoulder and narrating your interactions with humans.

It's slightly terrifying, no?

Well, so imagine, for example, your mother is walking behind you and every 30 seconds says, John, yeah, I noticed you're slouching I noticed you're not looking at humans. I noticed you keep repeating yourself and you start every sentence with hey or woof. Consider the following alternatives. And so the robot, every 30 seconds is getting that kind of mentoring input that's designed to improve gradually its interactions with humans. So you're dealing with a committee of large language models that all have different urgency and they have different cognitive capabilities because some need to be simple and fast and others have tens of seconds to think. And then in aggregate, the system is providing hopefully optimal outputs.

Yeah, so it's interesting, there's kind of analog to what happens in the brain and there's different parts of the brain which, you know, sounds a little bit like what the Freudians would talk about in terms of superego, ego and id, that you have different bits of the consciousness doing different tasks and some of us have got, you know, our mothers in our heads saying don't slouch. And that's very Freudian. Yeah, yeah.

This is hilarious. You've exposed my almost complete ignorance as it relates to Freud. But it's fascinating to hear that and your.

But you've rebuilt this, you've rebuilt the system.

You know, yes, I'm a proto Freudian and I hadn't realized it.

Yeah, yeah, yeah, yeah. It's so interesting. I want to talk a bit more about the hardware side of robotics, but first of all, could talk about this is a learning podcast. I feel as if I ought to raise the learning question pretty early in the interview. I'm interested in a few of the things you've said on that score. Firstly, that your 13 year old son, who we said hello to earlier, has conversations with a robot dog to help him this maths homework. And you've indicated that there's a huge potential for learning here. You said this is a great moment when robots can start to do this. But an objection to that would be, if you're going to be cynical, you know, you can do that kind of teaching with just text and images on, for instance, a tablet. Why go to that bother? So, but to put it more positively, what might an embodied AI give us in a learning context that screen based media can't? Well.

I have strong opinions about learning. I've always thought that the normal classroom setting where you have one teacher and then you have 30 students sit quietly and take notes. I've never been convinced that that's as effective as it could be. And probably for many people listening, some people, maybe you learn by building something, maybe you learn by watching a movie, maybe you learn through a conversation. And for me personally, the way I learn is to break things. I take them apart and I, like, go look inside, and then I try to fix them or build them. And so there's a big part of just, like manipulating things with my hands, for example, and in terms of just looking at a screen and sort of passively watching something on the screen, I've always wondered if there's great benefit to like, a physical or mechanical component to demonstrating things. Like, most obviously, when I go to the hospital and I see how doctors are trained when it comes time to train them, cpr, you do this by putting a plastic patient onto a bed, and then you literally walk them through the steps and get them to do that. And there's actually, there's two parts to your question. The first part of the question is the sort of intersection of AI and learning in general. And I think the real opportunity there is that you can personalize content specific to every child. Many children they may have, there may be one thing that they don't understand, and that makes it difficult for them to learn new material. And for most human teachers, it's very difficult. Just speaking for myself, it's very difficult to keep track of exactly what every student in my classroom knows and doesn't know. But with one thing the AI is good at is remembering exactly where every student is in terms of problems they've successfully solved and so forth. So that's. That's. I think a general advantage of AI is that you can personalize content more specifically, but I like to think that you can also make the content much more engaging. So instead of, for example, showing an equation on a screen, you could animate a physics equation, for example, by showing a bouncing ball or water flow downstairs, just to get at concepts better. And then, of course, if the AI is now controlling a physical object, it's able to jump around and bark, and it's animated and moving. That's probably very useful for smaller kids. The small kids come running when one of our robotic dogs goes to a kindergarten. They love the robot dogs. They wanted to sit and jump, and they talk and engage, and their eyes sort of light up. They're fascinated by it. And that's, of course, very important for teaching. You need the student to be to listen or at least look at you. And then for other areas, you can imagine, for example, the humanoid robot directly demonstrating things. A new company is just starting focusing on rehabilitation medicine for people who've injured knees and shoulders. And what the humanoid is doing is directly demonstrating the exercise they should be doing. So the human can has this model in front of them. So those are then extra opportunities unlocked by virtue of the large language model controlling a physical robot, as opposed to just being this sort of passive, dreary voice interface.

If you're looking to connect with the top minds in learning and talent, there's no better place than an iventive event that's I V E N T I V. Since 2009, they've hosted nearly 300 interactive conversational events exclusively for senior leaders like Global Hitler Learning. I've been a sponsor myself and found these events invaluable. Less about sitting and listening, more about real impactful conversations. It's an unmatched opportunity to collaborate with leading chief learning officers on the challenges they're tackling today. Join the conversation visit iventive.com learninghack that's iventiv.com learninghack I run another podcast called Great Minds on Learning, which is all about learning theory. As you're speaking, I'm kind of ticking off in my head, oh, yes, that's learning by doing. Oh, yes, that's. There is a lot of learning theory to support a lot of the things you're saying there. I'm interested how you might apply what you were talking about in terms of the dog for blind people with the multiple different types of intelligences. Because in learning you're also talking about kind of dialogue based learning, which goes back to sort of Socrates and the Greeks. And Socrates was a dialogue based kind of way of learning. And you could have your inner Socrates as part of one of those intelligence, rather than your inner mother, so to speak, prompting the dog to ask interesting questions that are going to kind of challenge the student and kind of unlock knowledge. So that is a really interesting area.

I think that's exactly what's happened for the educational robots. We've been changing the prompt to prevent them from immediately giving the answer. And instead the prompt is engage the human and emphasize the human's understanding of the concept, comma, through a series of engaging questions. So that's exactly what, what, what has been coming out of the just having kids interact with either the dogs or the humanoids.

Yeah. So you were saying that the challenge with the robot dog helping you some of this maths was not to get the dog to just do the homework for him.

Exactly. You know, it can win the math Olympics. And so obviously it's easy for the dog just to do middle school math homework. But as a parent, of course, that would make no sense. I expect Ben to actually learn the.

Material yeah, Another thing in what you said, it reminded me of something, I think in one of your other YouTube appearances. My learning style is YouTube and Wikipedia, as I very often say, on the great minds of learning. But there was a Richard Feynman quote. I can't remember exactly what it is. Perhaps you can prompt me, but if I can't make it, I don't understand it.

Yeah, if you can't build it, you don't understand it. Exactly.

Yeah. Which is really interesting, I think. And I'm also thinking there again of experiments I've seen in kind of virtual reality learning, which is much more hands on, scenario based, your kind of inner world. Whereas what you're doing is rather than taking people into a virtual world, you are bringing the virtuality, the AI, into the real world. So just think.

Exactly. It breaks my heart when I see kids that are staring at a iPad during dinner and their parents are talking and laughing and the kid is staring at this iPad and not engaging. And it'd be very nice to, instead of having kids look at computer screens, to have the technology just bring them out of that a little bit and back into the physical world. And I'm much happier with a robotic dog running to the park with the kid and them playing with a tennis ball or something than I am about the kid staring at some computer screen.

With your other hat on, so to speak. What effect do you see AI having on higher education, on what is taught and how it is taught?

Well, if you have, I have yet to meet anyone on earth. Who is able to map out a road forward for all of education and universities generally. What I see is that most universities are in this business as usual mode where they still think of their primary responsibility as, you know, undergraduates come, come in the fall and we go through this material and three or four years later we give them this degree. But obviously things are changing extremely quickly, much more quickly than most universities are able or willing to change how they're organized. For example, one thing we see practically in computer science hiring is that three or four years ago, if you were a Stanford computer science graduate, you are almost guaranteed to have an amazing job immediately after graduation. And just in the last three years that's changed almost completely where the entry hiring for computer science engineers has contracted very, very sharply. So the world certainly is changing around universities, but many universities are not changing as quickly perhaps as they could. Just to give you a few examples, what would Stanford University do if a large language model applied to become a student? Is the role of universities to educate only humans? Or is the role of the university to educate all computers, regardless of their compute architecture? How do should universities set aside resources to help large language models learn? Those are just a few questions. And then the other thing we struggle with at Stanford, and this is probably true of all universities, is we have to come up with good policies for how our human students use large language models to, for example, do their homework or help them write essays or help organize materials. That question is incredibly more difficult to answer than something as simple as a ban. For example, my PhD students, as they think about what problems they want to solve in their PhD, I tell them to use large language models and deep research to summarize the literature because it's patently impossible for a human to read 37 million papers and digest 37 million papers and prioritize open research questions. So in 2025, the standard guidance for PhD student literature summaries is to have the computer do it, because the computer is just much better at it than a human. So all universities and all teachers and professors have to find some way of navigating this technology which is exceeding human capabilities. More and more in different areas.

Different but related question. I was at an AI conference a couple of weeks ago and people from Microsoft were presenting their vision of the future of of AI through Copilot and so on. And there was an interesting question that came in from a friend of mine actually at the end of the speech, which was it appears that graduate level jobs are kind of disappearing or there are less and less of them coming onto the market. And this may well be a result of AI. People are worried that because AI gets to do so much of the grunt work you'd normally give to an intern or something like that, that there is no longer a graduate appointment. So you get this kind of gap in the labor market with nothing to do. And what should you do about that? And the Microsoft response was, well, we think that higher education should pick up the slack here. They should teach people how to use AI. Now my podcast co host has we recently did a series, sorry, an episode on great minds on learning about higher education and critics of higher education, of which he's one. And I had him in the back of my head saying they're the very last people to teach people about AI. I think you'd probably be a bit more optimistic about he picking up the slack there. In other words, people are going to have to kind of actually study AI before they can go into a job, so maybe spend longer at university.

Well, unfortunately, AI is moving so fast.

And.

If anything, it will accelerate acquisition of new capability. And AI is very good at accommodating different languages and different backgrounds and different capabilities as it relates to the humans that are interacting with the AI. So this notion of prompt engineering as being a, as something that everyone needs to know, I think that's a short transient of one or two years. At which point the delta of being a good prompt engineer versus not even knowing what that is is simply too small. And from direct observation, you're absolutely right. A lot of the jobs that in the past were done by interns and were an essential part of them learning how to work with other people in teams and figuring out what a company does and how it works. That's all, unfortunately, under a lot of pressure. And there's no way to sugarcoat this. It's a big, obvious problem. And almost certainly businesses will pay the price for this. And not only businesses, but also society in general will probably pay the price for this more and more.

Could you think of a solution to that? Or is it something that other people other than you and I can have to scratch their heads about and come up with?

Oh, well, I don't think there are any simple solutions. I think the like. Imagine, look at the situation from an outsider perspective. The way we've organized, like our universities and society and jobs and taxes and democracy and all these things is built on a central notion which is that the things in this world that is running around and thinking and deciding and learning and so forth are people. And now all of a sudden you have a technology that is better and better and better at doing things that just a few years ago most people would have laughed at, like winning the math Olympics. Seriously. And so what you're touching on is just a much bigger problem of how, like, our society all across the system accommodates severe competition, severe, like cognitive competition. And there's a much overdue debate, broad debate and discussion about what we want this world to look like. So this is not simply a question of, oh, there aren't great internships for the summer for our graduates. It's a much, much bigger problem. Sorry to be negative, but it is. There's like, from my perspective, I'm actually very concerned about this because I see on the one hand a technology that's moving extremely quickly. And then on the other hand, I see a lot of institutions and businesses, universities, political systems that aren't having really difficult conversations right now about the best way to navigate this futurist.

Yeah, interesting. It'd be nice to continue that. But I Think it's another podcast. And I'd like to talk a little bit more about the healthcare side, which I've already touched on. You originally are a physicist. I think you corrected me there. I thought you're a biologist, but your work there brought you into kind of working on human biology and healthcare. And I think you've done work on the application of AI to healthcare. Well, I know you have. What do you see as the potential there of AI and healthcare? And do you have plans to be part of that in the future, or is your focus kind of solely on what you're doing with open mind?

Oh, well, healthcare, I think, is really important, I would hope, for every single child. Presumably they care about their parents, and it's also important for all parents. Presumably they care about their kids and things like that. So, I mean, healthcare is really important for all of us. And I've always been. It's always been awkward for me to see that some people have access to amazing care and other people don't. And so when we think about AI in healthcare, it's really important to keep in mind the difference between the platonic ideal of healthcare and healthcare in the real world. And in the US as you know, we do not have universal coverage. There's tens of millions of Americans who have no obvious way of paying to go see a doctor, which sounds incredible. And so for tens of millions of Americans, their healthcare is going to the hospital, to the emergency room because they don't have health insurance, for example, and the situation is not getting better. So when you think about the intersection of AI and healthcare, some people say, oh, Jan, everyone should have an amazing personal human doctor that they can talk to 247 and keeps them healthy. And I completely agree that is the ideal. But unfortunately, that's not the reality of most healthcare in the us and so when it relates to AI, the big opportunity there, I think, is to give access to good diagnosis, for example, at vanishingly low cost to every single human. I see no reason that exquisite diagnostic capability should somehow be limited to only some of us. No, this should be available to all of us. If I have a question about my health or want good advice for how I should stay healthy, I would certainly like to think that that's available to every single human on Earth for pennies. This should be ubiquitous. And what AI does is, at least in principle, give us a technology where we could make exquisite diagnostic capability available to every single human at essentially zero incremental cost. Once you've built these models, then what prevents you from making them available to 1 million people or 100 million people, or for that matter, everyone. So one thing that AI does, it solves this tremendously difficult problem of simply accessing care. A great example was during COVID I was working with the hospital system to build triage systems so that people with mental health questions could get matched to resources more quickly. And so we built this system and it was never deployed because the UN doctors said, yan, it is pointless for you to find more people who want mental health treatment because our human doctors are completely overloaded and the wait times are six months. So if you bring us another 15 million people who would like to talk to a human doctor, it's pointless because it will just double the wait times to a year. And so from my perspective, yeah, I completely emphasize, empathize with overworked human doctors. But then the solution is to invent other capabilities that allow humans to get access to some care. Another example is if you look at old people's homes. It turns out that the vast majority of people in old people's homes are never visited by their relatives. We all have this rosy picture of we age and we live in this beautiful place and people help us and make us food, and our grandchildren come to visit us. But sadly, for most old people, that's not true at all. Most people are by themselves with rudimentary, if any, care. And an Australian company has been deploying humanoid robots into memory care facilities. So these are old people's homes, specifically for people who are struggling to remember, for example, who they are or who their relatives are. And in that setting, you're dealing with people, 40% of whom have not been visited by a relative or a friend in the last year. So these are people that are very isolated. And when you deploy a humanoid into that setting and the humanoid looks at the person and listens to them and remembers their name and has an engaging conversation and is patiently listening to the 21st story about the same thing over and over again. Humans get so attached to the situation and they start to smile. And human nurses then have to rub lipstick off of the heads of the humanoids because people will kiss the humanoid. And it sounds completely dystopian.

It does in a way, yeah. And yet kind of moving, that story about rubbing the lips. My mother was in a at home towards the end of her life. She was suffering with vascular dementia. So I kind of witnessed. But I did visit her regularly. And it was the case that the nurses said that, you know, she's lucky that she has her family visiting her regularly because Most of the people in here don't. So that echoes, anecdotally echoes the, you know, she mentioned. Yeah, but there is a dystopic thing to it as well. It's like, you know, if I'm a kind of, I was born just after the Second World War and I'm starting to lose my marbles and my only visitor is a robot, isn't that going to freak me out?

I 100% agree with you. It sounds dystopian. But if the alternative is the old people, the old person sitting in a room by themselves and not smiling, like something just feels like if that's the alternative, then a spark in their eye and a smile and them wanting to engage with the technology, who am I to say that that's dystopian or bad? Like, who am I to remove a smile off of an old person's face who hasn't been visited by a relative in years?

Yeah, I think that's a really good point. On the hardware side, why is it that robotics seems to have developed relatively slowly compared to screen based AI, for instance? I mean, we've had this kind of big overnight reversions of opinion about AI. It's had AI winters, and then suddenly you get something like, oh, it just beat Garry Kasparov at chess, or oh, it just sold, just beat the best player in the world at go. Oh, yes, they just released this thing, ChatGPT, and it can parody the poetry of Robert Frost and all the rest of it. We don't necessarily get that in robotics. I mean, I know that there is development. You have these kind of dark factories in China where they have no lighting because it's all done by robots. But that isn't something that you see in everyday life. And you know, we've seen on social media those little videos by Boston Dynamics of kind of robots doing acrobatics and things. But that it seems to be few and far between that we get a real breakthrough in robotics. Perhaps it's because science fiction has shown us so many robots over the years. I don't know. But is that wrong, that subjective opinion, or is there a reason for it?

John, you are completely wrong.

Yeah. Good.

And that's because most people are not paying attention. A great example would be Waymo in San Francisco. Robotic cars. And so those are not cars. Those are robots with four wheels that are driving you to the airport very, very quickly. And your life is in their hands. And if you had told me a few years ago, Jan, you're going to be 20 times safer off having a robot drive you to the airport than you doing it, I would have been surprised. But now that's the daily just experience here. The other part of robotics that has been moving very quickly are of course military and defense applications of robotics. A great example would be the drone war in Ukraine.

Drone war in Ukraine, yeah.

So the ability of a drone to loiter and do autonomous target prioritization and its ability to localize itself based on visual data because GPS is being jammed, those are all examples of increasing autonomy and decision making in the military world. And likewise, if you visit factories, some factories, they're almost completely automated and you already gave an example of that. But it's unfortunately true that most people do not spend their time visiting, for example, the Tesla Gigafactory and noting the absence of many humans in the factory. So a lot of these, a lot of the intersection of AI and robots is taking place either in, is taking place in settings that many people fortunately or unfortunately do not spend a lot of time in. What you're really referring to is the lack of like a humanoid in most people's living room.

I don't have a robot butler or.

A like the robot butler. And that you're absolutely true. That is the one area this like direct daily contact with smart machines that most people don't have yet. But there's many other areas where the technology is moving very quickly. When it comes to humanoid robots for the home, we expect that end of this year people in San Francisco will be able to go to a store to pick up their humanoids to bring with them to their house, and that will only accelerate. There's one problem right now we're seeing that most people, when they think about what the humanoid will do in their home, they have a very old fashioned perspective on what humanoid robots are able to do. For example, the number one request for a thinking machine for your home is to get it to fold your laundry and do your dishes and mop your floor.

Empty the dishwasher.

Yeah, you want it like, oh, I want it to wet mop my floor and fold my socks. And from a robotics perspective, those are exactly the most difficult tasks. These fine motor skills involving fabrics or like soapy wet floors, that's exactly what you don't want. So the first wave of humanoid robots for the home, they're designed to, for example, be health companions or to educate or help you move things from your robotic car, like your Waymo, into your kitchen. So solving the last mile problem. And they're not designed to wet wipe your floor or fold your socks. So that future is still far away, where the $200,000 humanoid robot helps you by virtue of helping you with your socks. The first wave of that technology is going to be much more focused on things like health or education or solving the less mild problem.

Yeah, interesting. I deliberately say Robot Butler because I was thinking about the relationship we're having with AI and what it's like to have a relationship with a machine that's smarter than you and think of Robot Butler. I think of. Well, I know that you were at Cambridge for a while. You've got a bit of association with British culture. You may know of Jeeves and Wooster.

Yes, of course.

Yes, I think that may well be the kind of relationship that we're going to have with these robots. You know, Bertie Wooster, nice but dim, and Jeeves, much smarter and really the brains of the operation. That may well be the way it's going to be with me in my old people's home and my robot dog. You do paint some really fascinating possibilities for the future and it's possibly to get really excited about that as well as slightly worried. But unfortunately we had to bring this to an end. Lastly, who would you recommend that people follow if they want to know more about this space? I mean, if it is a. A few of the usual suspects, by all means, name them. And do you have any book recommendations?

Oh, I spend almost no time on social media because I have a robot world to build that is useful and helpful and works with humans. So I really don't spend any time on very much time on social media at all. But in terms of the thought provoking books, one obvious recommendation is Isaac Asimov's iRobot, which of course deals with the great difficulty of providing a code of conduct or a constitution to robots. And the entire book is about the unexpected difficulty and the unexpected consequences and all the difficulties robots have making the right decision as they interact with humans. So Isaac Asimov's iRobot and his Three Laws of Robotics and all the different ways thinking machines struggle to navigate their interactions with humans.

Okay, amazing. Of course, my recommendation would be the Jeeves and Worcester books and I hadn't.

Made the connection, but you're absolutely right.

Okay, thanks. That's absolutely fascinating and surprising, not least in your kind of citing Asimov, who I thought was really old school, but interesting to see that it's still relevant.

Well, Google just baked Asimov's Laws of Robotics into their most current model for robotics, which is Gemini Robotics and Asimov's Laws of robotics are baked into that, as they should be.

Oh, interesting. Okay. Thank you very much for taking the time. It's been a great interview. I've really enjoyed it. Thanks a lot, Jan.

Thank you very much, John. It's been a pleasure.

That's all on the Learning Hack podcast for this time, many things thanks to our guests and to our sponsors. The Learning Hack is completely independent and transparently funded by sponsorship. It is produced by John Helmar and hosted by him and also me, Natalie as Synthesia AI avatar. If you want to help others find us, please like follow, rate, review and subscribe on your podcast platform of Choice or on YouTube. Until next time, stay curious. Learning people.

Now I finally get it. The Learning Hack podcast is supported by Learning News, the Learning Sector's newswire. Rob and his team are good friends of the podcast and we really value the help and advice we've had from them and they do a great job. For the very latest news from around the Learning Sector, for interviews with learning leaders, the latest from Learning Sector vendors, and features on workplace learning, go to learningnews.com.