Hod Lipson Interview Transcript

[music]

[laughter/voices]

Bruce McCabe: Action. Exactly, and I've got... I was reading your bio and it's an incredible, story you've got, multiple universities, your PhD in Israel...

Professor Hod Lipson: Well, that's actually not atypical, I would say. Most faculty at least in the us, many come from other places, this is sort of the mecca, and faculty tend to stay in the first, in the same university for a very long time, but we decided to move it from the rural area into the city. And there was a...

BM: And you're a professor of Computer Science? Would that be correct or...

HL: So that's mechanical engineering and ...

BM: Mechanical engineering and data science.

HL: Data Science. Data Science, which is an emerging area, it's not a classic discipline, but it's inevitable and it weaves its way into everything.

BM: And it speaks to what you are, you're combining these two areas, the hands-on, the physical, the experimental, the robotics side, also with the theoretical and the artificial intelligence and the software side.

HL: Exactly. Yeah, so when it comes to robotics, you really have to be a jack of all trades, you have to do everything because there's material science, there's manufacturing, there's design, there's mechanics, there's programming, there's AI, there's machine learning, there's electronics, there's a whole thing, and this is everything.

BM: So before we get onto this multitude of stuff you're doing in Creative Machines Lab, can we just talk about you for a sec because I didn't really get your story as to how you got here. Was there an inspiration? Was there a particular event? Was there something that started you down this path of being interested in robotics early, or was it an evolutionary process?

HL: Yeah, there's a lot of serendipity to it, I have to admit, it's not like I, as a child, I wanted to do all the...

BM: [laughter] It sometimes happens.

HL: Sometimes happens, not for me, but I think, to me, I've always been interested in this, maybe even a little bit of a hubris in this idea of, "Can we make life? Can we create, recreate a robot that is as intelligent, as conscious as a human?" And you see it in movies all the time, it's not particularly an original idea, but it was always almost a taboo topic in academia, so to...

BM: Really?

HL: Okay, so even the question of, "What is consciousness? Can you reduce it into a machine?" Is still...

BM: That's a thorny one.

HL: It's a very thorny one in academia, outside of academia, for many reasons, and people still, we have this I think this very human-centered belief that humans are special and, "No, it can't be done. A machine cannot have feelings. A machine cannot be creative." There's some things that are very, very deeply held that we humans feel, and for me, that's a call to action. When somebody says, "A machine cannot be creative, a machine cannot have feelings. A machine cannot self-reproduce." That was another one that is uniquely biological. You look at, one by one, these very tall walls that separate us from the everything else, that's for me a call. And so one by one, I'm looking at these things, and sentience is the ultimate one. Creativity has... We've been working on this for a long time. It is now falling left and right, and you're probably aware of all these things, and it is also surpassing even my wildest dreams and how machines can be created.

BM: Really? Already, it's surpassing your dreams?

HL: Yeah, already everything from art...

BM: All this art, yeah.

HL: Visual art is probably the easiest but everything from writing text to engineering creativity is just starting.

BM: Oh, those prototyping, where you see AIs have started to prototype a new lamp, it's given a set of instructions and it just comes up with a brief that begins the process for the engineer.

HL: Exactly. All kinds of things, I mean the... the self-application, which is work we did a while ago, we're still working on that, and, metabolism was another one, that machines can consume material to make themselves better. There's a whole list of them.

BM: This idea of sentience and all that, which seems so controversial, I have another way of approaching it, which is, if we can edit and create life, which we're now doing from that angle.

HL: For the biological good …

BM: Right, so if we're doing that and we're calling that "a bit of machinery," clearly it's machinery. And we're building it up, so what is the difference if you're coming from that angle?

HL: Well, there are people say, "Well, you can edit an existing living thing, but to create something that truly has emotions and... " It’s – the resistance I think is not logical, it's more...

BM: No, it's emotional.

HL: It's emotional, and it's self-preservatory, in a way but part of it, I think for me it's one of these big questions. I think humans... If you look at Biblical texts, which in Israel we studied, mandatory, so I know it inside out, even though I'm an atheist, the first chapter in Genesis is trying to answer the origin of the universe, the origin of life, and the origin of the mind in just one chapter. And the answers are unsatisfactory to me but scientifically, we've been able to pretty much give a pretty good answer to the origin of life and a pretty good answer to the origin of the universe but not to the origin of the mind. And what is consciousness? Where it is. So I have a very simple answer to, "What is consciousness?"

BM: Oh, go for it.

HL: Are you ready?

BM: Yeah. I'm ready [laughter].

HL: Alright. The answer.

BM: It's not 42. Right?

HL: Yeah, exactly, that's another question that that's the answer to. But it is... And I've read a lot of texts and philosophers, and they didn't have a good answer, so here's my answer. It's the ability to imagine yourself in the future. Alright? And it's not a black-and-white thing. A dog can imagine itself eating lunch, so it can drool before it... We can imagine ourselves in retirement. Alright? So we plan, we think about that. What are we gonna do so that we...

BM: Wow.

HL: Okay? So we think about our family after... So we think to the degree to which you can see the horizon, that's how self-aware you are.

BM: That's really interesting.

HL: And then with that... For me, this is a practical definition that I can build into a machine, and it's not only that. So we acquire that ability over time, so an infant can also probably see itself into a lunch. Alright? So the horizon is very short and it forms a model of itself, and as young children, they can see maybe a year ahead. You know? And as you grow older, you see further and further, you become more self-aware, and I think this is uh ... We are now having our robots... So our self-aware robots are not what you would think is a self-aware robot. It's not, ‘Does it look like a human?’ ‘Does it say "hello?"’ ‘Does it wanna take over the world?’ None of that stuff.

BM: And we don't care about any of those things, really. We want outcomes.

HL: It just models itself. It begins to imagine what it is, and right now they can predict their own future to about, let's say a few seconds. Alright? So they have the self-awareness of, I don't know what's their biological equivalent? But it's not more than a cockroach. But it's climbing and it's the first step on a long journey but with that definition, I have no doubt, with exponential growth, we're learning...

BM: That's quite powerful, because it's quite a simple concept.

HL: But you can see the evolutionary advantage of being able to imagine yourself in the future. You can make plans, you can decide, "Should I do A and B?” because I can see, imagine myself doing this.

BM: And also, the further you can imagine, perhaps it's quite analogous to ‘the deeper the consciousness.’

HL: Yes. Exactly so.

BM: Because it's quite a sophisticated process. If you're imagining further out, it's exponentially more complex things to imagine.

HL: Exactly. More uncertainty, more things to factor in.

BM: More variables. Yeah.

HL: But the more you can do that... And imagine multiple futures, it's complicated, it's not a simple answer. We also have different models of ourselves, so you can imagine yourself in the future financially, you can imagine yourself physically, you can imagine yourself in terms of your legacy and accomplishment. There's different angles.

BM: Relationships.

HL: Relationships, so it's a sophisticated thing. So again, if you look at children and humans and adults and so on, there's different levels of all of that, so we we're at the very preliminary stage, but that way of thinking about robotics also allows robots to do a lot more. So, for example, this is a robot here, maybe I'll... This is Johann's project, maybe we can turn it on. This is a robotic face, it can do all kinds of interesting things, can smile and stuff like that but it's not self-aware in any sense of the word. It can fake emotions alright? [laughter] Better, by the way, then all these other robots you may have been on stage with that have a canned conversation and all that stuff. I've had to hold conversations with... I forget what's the name of that famous robot, and pretend like it's spontaneous. "How are you doing, man?" But this robot, the one thing that is self-aware. Behind, see, this is a rubbery mask which we peel off and put different things, you can see behind the scenes, it's a very horror movie kinda thing.

BM: Oh, absolutely. I'm on the set of Westworld right now, pulling off the face.

HL: Alright, Exactly, so you know. But it has cameras in the eyes and the whole thing, but what's really hard about this, it has 25 different motors and they distort the face, but this face needs... But in order to smile, this robot spends about a day making random faces in front of a mirror, and it watches itself, it's like a horror show because it distorts in ways that you can't even... In horrible ways. Alright? And after it does this for a day, it can learn what it's gonna look like when it activates particular motors. Okay. This is a very simple self-model.

BM: Yeah, so it's experiential learning, feedback loops...

HL: But it's the self model, it can imagine itself in the future in terms of like a half a second in the future, so just like when your brain pulls certain muscles, you know what you look like, you know that you're gonna smile, you know you have a self-aware, and this is what we did. And once it can do it, it can start imitating YouTube videos and smiling like a human, and frowning or doing whatever, whatever it want. So that ability to model itself into the future even in a simple thing like this, is useful.

BM: That's fascinating because that's a whole other way of looking at it from when I saw one of your early videos, you had a robot that learned to move itself, maneuver itself around in a very unusual way because it had done it through trial and error, and that's how I often think of these things or the hand manipulating the Rubik's cube, the robot hand. So this is a whole other way of thinking about it, it's imagining the... One might have been imagining itself moving across a room, but now I’m imagining what the facial expression will be if I pull these servos.

HL: Yeah, yeah, exactly. Exactly, so self-simulation, so that's what we're trying to do, so our recent project was this robot here [HOD MOVES TO A DIFFERENT ROBOT], which models itself, and it's a robotic arm. Again, nothing special about this thing, and it moves around almost randomly in front of the mirror again for about three hours, and there's a camera, so the mirror is basically a camera that's watching, and then after thinking about itself for about a day, it has this almost like a cloud, that is, how it thinks about itself, it's body self-awareness. If you close your eyes and you think about where your body is, what part stays here your arms are...

BM: Where your arms are, what space you are occupying...

HL: Exactly. Exactly, you don't have a... It's not to a millimeter, you can roughly know where you are, it's almost like a cloud, where I think I'm roughly here and not there. And now if you need to reach, grab an apple, but not hit a thorn, so you can imagine where you're gonna... How are you gonna weave through and where your body's gonna be and whether or not we need to go through a door. That's what this robot for, so we can actually visualize this and you can see the cloud around the robot, And there are some videos which I can send, see if you can watch. And it's fascinating to see how the robot imagined itself, and sometimes it gets a little bit wrong. It thinks that it's slimmer than it really is.

BM: Is it a continuous process so it can continually learn?

HL: Continually, and it's not...

BM: And it's not just one-time, three-hours, you can keep it going?

HL: No, it's a lifetime.

BM: Lifetime. Okay. That's really important.

HL: That's really important, so we do something, like we break it, we bend the arm, and the robots only bumps it to something where it thought it shouldn't, it's unexpectedly... And that's again a metaphor for life, you bump into something where you thought you wouldn't be. That means you need to, amongst other things, adapt your self image, and this is a... And then it goes into that cycle. It continuously adapts and you can see the robot’s, basically, self-image also deforming to match reality.

BM: Very interesting. This is a whole... So I'm looking at robots that learn, and always those different learning pathways and parameters. There's experiential learning, continuous learning. We want robots to learn like babies and learn forever, right? That's finally our goal.

HL: Life goal. Yeah.

BM: Yeah but it's one thing to say, "I wanted to keep reading material on medicine or manufacturing and keep learning that way,” but this is actually... It's also constantly learning about itself.

HL: Yes, exactly, and there's almost this blind spot where as roboticists, we always have robots learning about everything else. About the world, about interaction with people, about machines, cars, whatever, welding, but not about themselves.

[laughter]

HL: You take all that technology and now you did all reverses and a machine begins to learn about itself, and this is what happens in... And I think it happened to humans as well. So as primates, we learnt about others all the time, you have to... If you are a predator, you need to model the prey in order to catch prey, so animals model other things all the time, and thus eventually we started turning that and modeling ourselves, and then we're thinking about ourselves as a third person, and this is where this consciousness of awareness started, but I bet that our ancestors were mostly occupied with modeling other things. The competition, because...

BM: Have not been solved. So if we now push the boundaries. We were just chatting before we clicked on the recorder about the fact there's some philosophical goals and really big goals here, and then there's short-term practical ones that fall outta that, but if we push the boundaries and go, "where could we end up with this?" In my imagination, I might be thinking too small, I'm thinking there's a factory somewhere that makes lawn mowers or the furniture in this room, and the robots in that factory, maybe there's different product runs or different material being used and the chairs that are being manufactured, and the robots find that their grippers aren't appropriate and their maneuvering needs to be adjusted, and maybe they then go back and get some adjustments made to themselves, to do a better job.

HL: Exactly. Exactly. Exactly. So there's many practical applications. If you are in a self-driving car, you want your self-driving car not just to model the road, but to model itself to know, "My left-front wheel isn't working well today, and I'm gonna slow down," or, "I cannot drive in this snowstorm, I have not experienced this before, I don't know how to... I can't do this. I'm not equipped." This is all examples of self-modeling. Or, "I'm broken," or, "I'm not experienced enough," or...

BM: Or I put something on the roof rack, my height's changed now.

HL: "Now I can't... My dynamics are not right, I need to go more slowly.” Whatever it is. And you want your machines to do that.

BM: Definitely. Definitely.

HL: We want them to do that, and that's for smart cars, smart cities, smart factories. The more, the smarter they are, the more we want them to be self-aware. We want the financial systems to be self-aware, to know that something's going on, gonna slow down today. There's a...

BM: [laughs] That'd be good.

HL: Yeah. It would be good because, you see, the more we depend on this technology, we need it to have these checks and balances inside because we can't crawl around fixing these things all day long. I mean this is... We are already reaching a limit where maintenance of machines is a problem.

BM: Definitely.

HL: We rely more and more on these systems, supply chain and all that, but we don't have the human capacity to maintain it, so these machines need to take care of themselves, and that's it. And this is what biology figured out a long time ago.

BM: That line, "The machines need to take care of themselves." That's a really nice summary of that. because I'm constantly running up against the economics of sophisticated robots, it's all very well to say, "We can do machine learning with free tools and tool sets and data sets and effectively do it anywhere." But when we get to a machine, you can't run away from the cost of the servos and the steel and the cameras...

HL: And the cost of mistakes, and this is a big piece. When you have a physical system, a mistake can be disastrous. There's energy cost, there's risk, there's time, collecting data, you're wasting time, you can't do a million trials and see what works, so these machines need to take care of themselves, and this is what we're trying to do.

BM: Interesting. Yeah, and so...

HL: That's the short term. The long term is my philosophical interest in this, but I think from an evolutionary point of view, it's the same thing. Evolution gave us this gift of self-awareness to save us so we can take care of ourselves and make the right decisions. That's evolutionary advantage. But it's very expensive, so this is why it's not black and white, and some animals only need to predict a day, some animals can get away with predicting... If you can get away with predicting for an hour or for a minute, that's all you need, you don't need to carry this big brain. And this brain is unstable, I mean it can make the wrong decisions, it can do all kinds of things. So you can see... If you look at the animal kingdom, you'll see this struggle with how much self-awareness is the right level. [laughter] A cockroach has just enough to get by in, not too much. It's not  an obvious answer. More is not better.

BM: That brings to mind a scene in a wildlife documentary, watching penguins, and I'm trying to remember the producer. It's beautifully done. But in the Antarctic, and they had group behaviors and they all were quite sophisticated animals and this sort of thing, but every now and again, one penguin would wander off in a different direction. And they figured it was an evolutionary, exploratory response on behalf of [the colony]... But more often than not, that penguin would die. It would go off into the wilderness where there's no water, basically. And it was profoundly sad to watch, but that was an example perhaps in nature of having a limited ability to see itself in the future, but it just had to follow its response because that was a better evolutionary action for the species.

HL: Yeah, exactly. So I think there's that balance. And we humans are the ultimate extreme, which may or may not be a good thing. We'll see, we'll see how it ends.

BM: Yeah. Well, we always get to put our own checks and balances on things and people will get very quickly into dystopianism and, "Oh my god, the robots will take over," which I never buy into because we have such ability ourselves to design what we want in or out. But if we look at a word used a few seconds ago, "self-replication," and again, we stretch out just on a practical level, where do you see that... Just when you're—it's a dream, obviously, at this time, but can you see a practical scenario where robots are building themselves or building other robots as part of their functionality?

HL: Yeah, that's again inevitable for the same reason. So if you have a society that uses robots, you can't crawl around building robots all the time, and you have to have robots building robots, so there's no other way to do it, alright? You are not gonna build every... Robots build everything except themselves, practically, this doesn't work. Yeah.

BM: [laughter] That doesn't work.

HL: I mean this is obvious, so naturally, it's building...

BM: Yeah, they build cars, they build everything.

HL: So they're gonna build robots from robots...

BM: 3D houses, of course.

HL: Of course robots are gonna build robots and this is self-replication. That's it. It's not very complicated and it's not that magical. Now there are some really interesting engineering questions about, "How do you make a robot that can build itself?" And there's some design, recursive design constraints, but it ends up being the answer, that the robot has to be modular, like Lego, made out of almost standard pieces that it can pick up and place...

BM: Got it.

HL: And then that robot can also pick. So modularity is the answer. And then biology is the same thing. We are all... Plants, humans, chickens, everything are made out of the same 22 amino acids or whatever it is, and they just rearrange it different ways and you get different... Chicken or human. And this is why we can eat a chicken. We can eat a chicken because our body decomposes the building blocks and reuses it for itself. This is not magic, it's just reusing the same nuts and bolts. And once you think about it that way, it was a no-brainer for us to build that robot over there [HOD MOVES TO DIFFERENT ROBOT], which was already long dead but you can see a top-left cubicle, it's a robot made of blocks.

BM: This one here. Yes, that looks like...

HL: This one, another version, and these robots... You can assemble a robot out of them, and that robot given more blocks can put these blocks together, make a copy of itself.

BM: So this is like almost on a large scale, a swarm of particles that form a...

HL: Yeah. But they bond, they connect in a rigid way.

BM: And once they're connected, they're a system that's connected.

HL: Yeah, exactly, and then... So this robot, it's harder to build a copy of itself because it's made of these rigid, un-modular, non-modular components that are made for humans to bolt together. They're not very difficult for a robot to build. But these are modules that a robot can assemble, and in fact, I'll show you a project that we're working on right now [MOVES TO A DIFFERENT ROBOT].

BM: Yeah. Please.

HL: This is our metabolism project, robot metabolism.

BM: Robot metabolism, oh my god.

HL: Where we have a robot that walks up to another robot, takes it apart, and uses its parts.

BM: Oh, I love that.

HL: Okay. Basically eats...

BM: That speaks to recycling as well, so I'm loving this. Yeah. [laughter]

HL: Yeah, so it basically eats the other robot. This is the translation. And because of the part that it ingests, it can run faster or do something that it couldn't do before, so that's metabolism. And again, it sounds like you are very ominous, but the reality is, if you're building robots, you want the robots not just to be able to make more robots, but you want them to improve.

BM: Yes, exactly.

HL: You want them to be able to extend themselves. And if you're sending robots to another planet, you want the robots to be able to take care of each other, to replace parts, they need to be longer to reach something, then they can use some parts and extend themselves, so this is all... Again, biology does this all the time and it's based on this idea of modularity of components. This is why we eat plants and plants eat us, and we're all of us in one big recycling, closed economy. This is what these robots are gonna be, and so that's the other side from consciousness. This is the body and the mind, and I think these are, in the end, this is for sustainable robot ecology, if you like.

BM: A sustainable robot ecology. That's really nice.

HL: Yeah, they're robots that can build and program and understand what they are.

BM: I really like that because it speaks to the economics of it, as well as everything else.

HL: Right now, we just throw robots in the garbage after their done, and there's no recycling at all. And again, it's not sustainable. I'm not talking sustainable in terms of the planet, I'm talking about even our ability to keep manufacturing more and more of these things, it's just not...

BM: It doesn't work. Yeah, unless...

HL: Doesn't work. We don't have the mental capacity and we don't have the physical capacity to do this and...

HL: That is brilliant, and that segues into something that really inspired me way, way back, what first brought you to my attention, I first found you, was actually seeing some videos of soft actuators, and it's also mimicking life in another way. And I wanted to ask you about it because it's one of those things which I think we've made only slow progress on, but we've seen all the soft-botics with soft grippers and all of that. They don't excite me. What excites me is the idea of replacing musculature and making the entire robot much more analogous to animal life. Not anthropomorphic, but at least biological in the sense that it's soft and it can work very closely with us. Are you still doing stuff on soft actuators?

HL: So soft actuators, you know, we were looking at different types. That was a while, that was in... So this is again part of our quest to evolve robots, and there's only so much you can do with rigid materials, really, soft gives you so many more degrees of freedom to innovate with. Things can move in so many different more ways when they're soft, and so again, we do a lot of work in simulation but we are also trying to find out, "How can we actually make a motor that is soft?" And that was that project. Actually, these are mostly... Actually, there might be some here. So we eventually came up with a way to do it, but it's not quite there yet, so we can't actually...

BM: Yeah, early days.

HL: Yeah. This is I think...

BM: But when I imagine the long future, I really think this is a huge part of it at some level. And again, if we can do this in biology, if our muscles will work and be activated and have a certain level of strength, then it seems to me that it's a reasonable target for engineering.

HL: Yeah, I think it's going to be a interesting substrate to make robots in, so we'll move away from purely rigid to combination. I don't have the muscles here, they're in my office, but they're made of silk... So basic idea that was... It's kinda interesting story, we had this idea that if we can put alcohol into... And heat it up so it boils, it will inflate a balloon, is that the one?

BM: Yeah, I think that's the one I saw, this one was a white polymer.

HL: But any time we put the alcohol inside, it leaked out. So in frustration, the student mixed the alcohol with silicon and then suddenly it worked because what happened is that the alcohol became these tiny, tiny bubbles inside the silicone, so basically you have a block of silicone we can shape anywhere you want, with microscopic bubbles of alcohol inside, and you heat it just a bit, the alcohol expands and then works.

BM: So it's quite serendipidous.

HL: Yeah. Yeah. Like a lot of things here, there's a lot of serendipity in.

BM: You were saying actually before we started, that a whole bunch of... What's the ratio of experiments you start...

HL: Yeah. About 1 to 10.

BM: One to 10. So of 10 experiments started, probably nine of them failed, don’t go anywhere.

HL: Yeah. They don't go anywhere.

BM: That's the nature of...

HL: And that's hard for corporations to understand also, that they need to experiment at that rate and accept that rate of error. But the key is to do it in parallel, because if you do it one by one, you're gonna be fired after the third one. You're replaced. But if you do 10 in parallel, one of them works, you can talk about that one and not talk about the other nine, and all will be forgiven. So that's what we do in academia all the time. We publish the good ones and we forget about the bad ones.

BM: We need to remind corporations of that, definitely.

HL: Yeah, so that's the self-awareness. Lots of examples of different robots and different ways to do it, but that's the gist of it, and that's the robot's building robots, and a lot of these things have interesting byproducts. For example one way to make robots make robots is using 3D printing, so we've been working on 3D printing for a long time, including... For example, I wanted to... One of the things I really wanted to do and have not been able to do yet, been lots of failures, is have robots... A 3D printed robot that will basically be printed and walk out of the 3D printer. Batteries included, the whole thing end-to-end. The printer will just assemble it...

BM: [laughter] Oh, I want be here for that.

HL: Yeah, and I want the robot to just walk out, and I haven't been able to do that. We've been able to do almost all of it, but never the whole thing.

BM: But it's the right stretch goal, isn't it? because then you've got completion. That's a holistic solution.

HL: Yeah, and I've had many students. I said, "If the robot walks, you walk." But somehow, they did their time and they walked out, so they're almost there, so it's really...

BM: What's the hardest part of that?

HL: It's all hard.

BM: The battery?

HL: No, I mean it can do everything on its own but the hard part is the integration, it's to do the whole thing in one shot. All these different materials, they all have to talk, you have to have a brain, you have to print a brain, you have to print a... But the beauty of this...

BM: Oh my god.

HL: … you have to print the controller, you have to print the power, you have to print the muscles, you have to print the structure, the whole thing, and it has to detach and walk, so it can't be locked in either. And we’ve people, "As soon as it's out, can we do it in space? Can we do it in water?” Maybe we can do it downhill. Maybe it can walk out of there, all kinds of shortcuts." But it's really, really hard. "Can we do it with explosives? Can we do it... "

BM: No, please! I like the space scenario.

HL: Yeah.

BM: You mentioned Mars earlier, and I was talking to the guys at NASA a few years ago, and they were explaining to me why they were a great hotbed for commercial stuff, particularly environmental stuff because they're like, "We have more constraints than anybody, certainly in the world because we're not in the world, we're out in space! So we've got so many constraints. If we can make something work there, it'll work anywhere."

HL: Yeah, yeah, exactly.

BM: Okay, that's cool.

HL: But the 3D painting is a good example because as we do this, we have lots of different offshoots, so one is bioprinting. So as we're printing with batteries, which is five different groups and gels and things like that, we said, "Okay, let's print live cells and see what happens." And we're able to print the live cells, and the bio printing field was born. That was about 20 years ago. And the student that led that founded a company, I don't know if you've seen on the front page of New York Times just about two months ago, there was a ear that was printed...

BM: Okay, I didn't see that.

HL: And it was implanted in a human from their own cells. There was a lady that was missing an ear, born without an ear. There was a little stub, nub. He took that, cultured the cells...

BM: Wonderful.

HL: Printed a new ear, implanted it.

BM: Was that here, was it?

HL: Yeah. Yeah, I mean it was a company that's spun out of here.

BM: Yeah, that came out of here, that's interesting.

HL: And that's an example of an offshoot of robots printing robots.

BM: Absolutely. It is, and it does relate now. I've spent time at Wake Forest Regenerative Medicine lab down there with Anthony Atala, and so now I can connect the two worlds. I see it. If you can print those things, it just brings you closer to printing robots, you get better-printing robots, you get better... That actually relates.

HL: Yeah. So it was just printing with multiple materials, and bioprinting has its own constraints and so on, but it's related. And you can see the food printing... So we have this thing, once something goes commercial, I cannot work on this in the lab anymore. There's a conflict of interest, you can't have students working on stuff I have investments outside. Which is for good reason. So we don't do bio printing anymore in the lab, but the next thing is food printing, and you're still working on that in the lab. We're still working on the...

BM: I wanted to ask you about it today.

HL: Yeah, this is the connection but you can see the connection to robots making robots.

BM: Yes. It all relates.

HL: It relates. And food printing I hope to some at some point, commercialize that. Looking for the right student to take it to the commercial world. But I think it's a revolution waiting to happen.

BM: Well, I'm kind of learning that space at the moment, more about it, and I want you to educate me on where you see the early opportunities, because I'm thinking in my head that the early opportunities are in printing better … because plant-based protein substitute's a very exciting space, and printing a better steak because you can actually put the fats or the proteins in the right places and give it more textures, and so that idea of doing better meat substitutes through printing seems to be stage one. That's it. I want you to correct me if I'm wrong. And then looking past that, I'm like, Okay, that's probably in one factory, but then perhaps you could distribute that to restaurants and say, "We're giving you the recipe and the raw materials, the feedstock to do that." And then you want to protect your IP if you're really good at designing steaks. And then maybe stage three is in the household where you're doing it at home? Is that kind of a... Are you seeing …

HL: I think we printed our first thing in chocolate in 2005. We did it before, but it was frivolous and we thought, "Printing chocolate, it's not serious for academics, we need to print a cleaning room, I mean the real stuff. But it turns out that actually printing with food is harder than printing engineering materials.

BM: It's gotta be really hard.

HL: It's very... Like even butter, cream cheese, depending on three degrees difference in the room, it behaves completely differently, and the plastic doesn't.

BM: And we haven't even got into the hygiene factors here.

HL: Exactly, and also, the plastic things are made of one or two materials. Foods have many ingredients, and they involve cooking. So while you mix them, you can't just mix everything and then cook it, you usually have to do steps in-between, so.

BM: Hod, I watched the video of pureed chicken being laid down by a printer in your lab, and then laser cooked...

HL: The laser cook. So the laser cooking is our latest thing, which is... It's funny, in terms of media, I can talk about the robot self-awareness all day long but when there's a new way to fry chicken, now that's a revolution [laughter].

BM: That's all I wanna talk about.

HL: That's all! Like self-awareness? Whatever [laughs]. You can see how... We're just animals looking to new ways to fry chicken.

BM: Eat the chicken. But I interrupted your flow a little bit. Sorry.

HL: So the laser was to me the missing piece because with 3D printing of food, you can arrange all kinds of ingredients in amazing ways but if you can't cook it, you always end up with this...

BM: Mess.

HL: … pudding that has things in it, but it's never appetizing. But if you can actually cook it while you print, now all bets are off because now you can do incredible things and it can come out ready to eat. Now some people think, we... I didn't frankly didn't think about the meat surrogates as the avenue, but it turns out that this is …

BM: Okay!

HD: … this is the area where indeed most people are interested in, there is the novel... I thought it was gonna be more the confectionery sweets, pastries, things like that. We've done original experiments with food printing, where we found that interestingly, people are very conservative about food, for good reason, and people will not eat stuff they're not very familiar with. Very conservative! Most people are happy eating the same three things their whole entire life.

BM: Familiarity, that's the keyword there.

HL: It's the opposite of innovation. So we look for innovation in literature and art and technology, but when it comes to food, we are happy eating the same thing, the same three things forever. And it's for good reason, if you are a hunter-gatherer you're not gonna try a new new berry just because you're gonna stick to the same berries you've eaten before.

BM: The ones that worked, yeah?

HL: And those who have tried other things are not with us. Right?

BM: Yep, yep.

HL: So I think it's like that penguin. Right? So we're very, very conservative. So we made amazing things, broccoli, jello, purple cubes made with broccoli, and they taste like... And people were revolted. But even though it's all good and it's healthy and safe, people don't wanna touch anything.

BM: It just triggered that unfamiliarity button.

HL: Yeah. And somebody else should eat it first, and no, nobody is gonna try anything like that. So we are back, so the only place we can have a little bit of innovation is pastries and sweets.

BM: Back to that.

HL: Back to that, so there we're accepting, "Okay, chocolate new shape, okay, I'll go with that." That's a cake that looks in a different shape, still I'm uncomfortable biting into a cake that's shaped like a horse or octopus, but I'll do it. A green cake like an octopus? Okay. Maybe I'll do it.

BM: Just make sure you putting enough sugar and then it'll work [laughs] So dump 10 tablespoons of sugar in.

HL: Exactly, so it's the most forgiving. So people were happy with that, so actually steak is really hard, so we actually have a project called Redefined Meat, where we're trying to figure how to model cuts. If you look at all the meat surrogates right now, they're all patties because it is very forgiving.

BM: Yeah, it's boring though.

HL: It's boring but it's very forgiving, but if you wanna print a cut of something equivalent to a cut of steak, if it's not perfect, it is revolting, it's the same thing, we are very... We have 18 sensors to detect bad meat.

BM: Really?

HL: It doesn't look good, it doesn't feel good, it doesn't... I just made that up, okay? [laughter] Lots, okay? But in many ways, it doesn't smell right, it doesn't look like... The color isn't right, the shape isn't right.

BM: Before we even get to taste.

HL: … The marbling isn't right ....

BM: Texture! Yeah, a little slimy. Yeah.

HL: Is it slimy? So we're working on trying to model what a real steak looks like, and I was asking people in this area, "Why do you wanna model a raw steak? After all, most people don't handle a raw steak, they want the final product to be right." And they said, "No, if the raw steak doesn't feel right, even if it's just veggies inside, if it doesn't feel right, they will not cook it. The chefs, even though that they know this is a meat surrogate it's not supposed to be... "

BM: Interesting.

HL: … If it doesn't flex in the right way, they're not gonna touch it because it's harder.

BM: That's even harder, you're gotta get it right at two phases.

HL: Yeah, exactly, and it has to cook at the same time and it has to... In-between, it has to be the same.

BM: Well, we're moving into almost impossible territory.

HL: Yeah. There are so many constraints on this, because we are so locked in, from evolutionary point of view, with checks and balances about what the meat needs to look like in-between.

BM: But then you see all the milk substitutes and... I feel very bullish about this world of meat surrogates, I really feel bullish about it …

HL: No, I think it is, there's no.

BM: … I think it's gonna be massive.

HL: There's not a single bad thing about it. Right?

BM: Yeah.

HL: It's all good but it's just a hard engineering problem, it's a lot harder to print a steak than to print a titanium rocket because a steak is so complicated and there are checks and balances on it. But there's a such a long list of things that we are looking for. And that's why we... So we have a project here, which is the 3D printing and the laser cooking, but also modeling real steaks, so that we can get it to look exactly right. And this is a challenge I think the industry is facing, going from a patty to anything but a patty.

BM: Got it, you got it.

HL: Oh okay. This is really hard.

BM: So Professor Lipson, is there anything that you'd like to correct out there? Any messages that we haven't covered that you'd love people to know more about or...

HL: Well, we did talk about creativity, I think that's, I think, an area where people have the biggest blind spots. I can't tell you how many times we've had people here at the university give a commencement speech about how creativity is... As long as you're creative, your job is safe, and you can't be replaced because machines can't be creative. That's a such a deeply-held belief, and actually that's not our biggest attribute as humans, I'm afraid, because AI is incredibly good at being creative. Everything from art to pros and the staggering speed that this is moving forward, I mean this lab is called the Creative Machines lab because of that, so we really are looking... We have machines that make oil paintings, and there are some more in my office, and at home I have one that creates art and things like that, but also robots that design robots, and this is a little bit harder to see. Antenna is designed by machine. And so machines can design and create new things, I don't know about you, but this talk I gave in Brazil two days ago, all my slides had art that was generated by an AI, I don't need to create my own, I don't need to steal images from the web anymore.

BM: I'm doing that as well.

HL: I just say, "Okay, I need a cloud that looks like a robot." Okay, and I got a cloud that looks like a robot. I don't need to google for that for three hours in the hope that somebody else made one of these things.

BM: It's incredible.

HL: It's incredible.

BM: And the participative process where you get it to generate design, but then you say, "But actually, I wanted a different style," and it'll redo it for you.

HL: Exactly, I wanted a pink cloud.

BM: "Let's just change this part," And it'll redo that for you. Oh my lord.

HL: Yeah, so that's a tiny example, and now videos...

BM: It's a blind spot.

HL: I want a dog licking your ice cream and a video. Yeah, I saw videos now. It just came out this week. Videos, you type in the description, I don't know if you saw it, I want...

BM: I didn't try it.

HL: I want a dog licking ice cream on a beach and there you go. There's a video of a dog leaking ice cream on the beach generated for you, I mean, and then I want a recipe involving pineapple, and the AI generates a new recipe. Look, this is just mind-boggling. I want a dialogue, two page dialogue, so I think people don't understand that it's creativity. You know I had... this is in Mexico three days ago, somebody pushed and said, "Can a robot write a novel like Agatha Christie?" And I'm thinking to myself, "The bar is moving higher and higher, then how many humans can write a novel like Agatha Christie? Right?

BM: Not many.

HL: Not many. And now this has become the obvious challenge... I mean this is a really high bar. Can a robot be Picasso? And we have these machines that discover...

BM: I'll say "Yes, already, no question, it can be Picasso at some level." Yeah.

HL: But maybe not Picasso but I give a talk on my painting robot in Brooklyn, where all the artists live, I was booed off the stage, booed off the stage. This is so emotional to people. Creativity. It's the one thing we have over animals, Is the one thing we have. And now they say, "You're taking this away."

BM: And it's a blind spot ...

HL: And we cannot see it. We don't see it. We refuse to see it. We okay cars driving themselves, and factories, but not a machine that's creative. That is really, really, really troubling. And the truth is that we humans are really good at physical, at fixing things, that's not a happy conclusion. Right?

BM: Yeah. Yeah.

HL: But this is … I think this is what we're good at. So the other thing we humans are very good at, this is another blind spot, is conversation. So there is no AI that can have a conversation, and maybe next year there will be …

BM: It’s getting close!

HL: But I can tell you from the inside of academia, it's not, I mean there's AI that can... AI can generate essays and one-way things and monologues and reports and poems, but not have a conversation like you and I are having now. There's so much in it. That it's not just words.

BM: Yeah. Free-ranging conversations.

HL: There's free-ranging, there's empathy, there's body language. There's a million things that the shared background experience that we bring into this, that is hard for a machine. Now, will that happen in 10 years? Maybe. But right now, nobody in academia has any clue how to do this. This is not like around the corner or anything like that. This is really hard. And so I tell my students, you know, "Be sure you know how to talk to other people because that's really the one thing we have for a long time."

BM: I think though, we're gonna blow through that. I know so many people working on conversational AI, and they're all very optimistic! They look at me like you're looking at me on the other things we’ve been talking about! And they say I have a blind spot!

HL: You're right, so maybe, it could be. So maybe it's coming, but I think actually that is the danger, that's the biggest danger because it's not, robots gonna take over the world like Arnold Schwarzenegger-type robots shooting people, it's gonna be a conversational robot that can be easy to talk to and fun and then you don't need to talk to other people, and that's it, that's the end of the world because...

BM: Social media was nothing, Facebook was nothing compared to that! Now I... Just talking to...

HL: Just talk to... I mean yeah, if you're talking... People are concerned about their kids having online friends, just wait until they have synthetic friends, and this is the end, because maintaining relationships with other humans is hard, it's a lot of work and they are...

BM: Oh no! [laughs] Come on, you strike me as an optimistic person, I don't know, you've taken me down a negative path.

HL: That's the one thing I'm worried about. I'm worried about conversational robots. Really worried about that.

BM: Okay, the social implications of that.

HL: Social implication of really good conversational robots that just take away your need to talk to other people. And that's to me the... It's not guns, it's not self-driving cars taking away jobs. It is...

BM: Definitely not jobs.

HL: Yeah, you're right.

BM: Yeah. Wow. Well that's given me a whole lot to think about that I wasn't thinking about before, so I appreciate that.

HL: Okay, good. But I am optimistic, and people say, "Why are you working on self-aware robots? It always ends badly in the movies, right? It never, never ends in a happy way, so why?" So I think the answer is because, again, there are so many good things that can happen when machines take care of themselves. The benefits outweigh the risk. But it is a dangerous technology, and I'm not in any illusion that this is a walk in the park, I mean we are entering a very treacherous area, lots of benefits but we have to play it right.

BM: Yeah. Yeah. Yeah. Well, Professor Lipson, I've wanted to meet you for a long time, so thank you for having me in your lab.

HL: My pleasure.

BM: I have two things now on my list I'm going to watch for that are gonna be milestones. The first is that first 3D-printed steak that comes out of this lab, so you have to let me know when that happens, I'll fly here and taste it and eat it, I don't care if it's...

HL: I'm sure you have a high bar. I'm gotta take some of those with a lower bar. You probably know your steaks.

BM: Well, and then the second one is when you print that robot the walks out of the 3D-printer, that I need to be here, I need to see it, I need to be live streamed to it ... Somehow I need to be connected to that!

HL: Okay.

BM: So we'll watch closely. Thank you for your time today. I really appreciate it.

HL: My pleasure.

[music]