Human friendly robotics
Artificial intelligence entrepreneur Mark Palatucci talks to John Thornhill about the consumer robot revolution and his efforts to help create empathy between humans and their robot toys
Presented by John Thornhill. Produced by Fiona Symon
Transcript
You can enable subtitles (captions) in the video player
Hello and welcome to Tech Tonic, a podcast that looks at the way technology is changing our lives. I'm Hannah Kuchler, Financial Times San Francisco correspondent. Last week, we heard from Danah Boyd about the impact of social media on our political and cultural lives. This week, we talked to an artificial intelligence entrepreneur about the consumer robot revolution.
Where is that robot looking, and how often does that robot actually turn in your direction and look up and try to look in your eye? And, you know, that's something that we didn't think about and then ended up being a very powerful moment. And then we've really tried to kind of lean into that and we've added quite a lot in terms of those types of things that really helped create that empathy and the bond.
That's the voice of Mark Palatucci, who came into the studio recently to talk to the FT's innovation editor, John Thornhill, about his work in trying to create empathy between humans and their robot toys, and about how we should got about regulating the use of consumer robots.
So welcome to the FT studio, Mark. I wondered if you could start out by telling us your background. How did you get interested in robotics?
I started working in robotics in 2004. The US government had launched something called the DARPA Grand Challenge. The goal was to create a self-driving autonomous car that could navigate its way through the Mojave Desert in California. You know, I was lucky enough to join the team at Stanford and participated in that, and just found a deep love for robotics and realised the huge impact it was going to have and decided to focus the rest of my career on that.
And that was Sebastian Thrun, was it?
That's correct. Yeah, he--
And how did that challenge go?
So Stanford was lucky enough to win in 2005 and it was just an incredible experience. I remember being there when the car drove its first mile through the desert. It was 110 degrees and super hot and you know, everyone, Sebastian popped a bottle of champagne, and we were kind of having this little mini party in the desert.
How did you get from there to being a co-founder of Anki?
A lot of the folks that worked on that team had trained and studied at the Robotics Institute of Carnegie Mellon, and at the time, I applied and was very lucky enough to get in and made the move to Pittsburgh, and that's where I met my two co-founders of Anki, Hanns Tappeiner and Boris Sofman. We were all in the same incoming PhD class together, and a few years in, we got super excited about the idea of applying robotics to consumer applications.
Why consumer applications?
Well, at the time, there was almost nothing that was happening in consumer. Most of the things we were working on were related to government research and industrial automation, occasionally military type of applications. But nothing had made it yet into consumer products, and we realised that there were some really big trends that were causing that to change and that's where we saw opportunity.
Right. So tell us about Anki. What do you produce? Can you tell us about some of the products that you have--
Anki is a robotics and AI company, and we're focused on consumer products. We're focused on taking the sophisticated science of robotics, computer vision, machine learning, signal processing, and really bringing it to applications and consumer products that you could buy at retail for maybe 100 pounds or 200 pounds. We've started focusing on entertainment, initially, so we have two products in market-- Overdrive and our second is Cozmo, and we plan to continue to focus on consumers going forward.
So we're launching Cozmo. It's a smart, interactive character. We've tried to bring the types of character and personality that you would see typically in an animated movie or feature film, something from science fiction, and actually bring that to real life using robotics and artificial intelligence.
Right. So describe for us a bit, Cozmo. What does it do?
So Cozmo's a robot that you can play different types of games and activities with. He'll do different tricks for you. You can really just watch him on your tabletop and see what he does and you don't necessarily know-- every day is a little bit different. We've tried to make his personality really come through. We've hired folks from some of the best animation houses like Pixar and Dreamworks to really bring out, you know, his character and to create a relationship with the human player.
What does he look like? I mean, how big is he?
He can fit roughly in the size of your hand. He has a head that he can move up and down, he has a lift that he can use to pick up blocks and also push objects around, but roughly the size of the palm of your hand.
And what's particularly clever about Cozmo? You're saying that it uses AI-- in what respects?
So we've really thought hard about the personality and we wanted this character to be incredibly lifelike. And we've built an entire engine from these feature film animation tools, all the way to the physical robot, and really tying those animations in with our AI system and what we call our emotion engine. So as the robot navigates the environment, he'll see different types of objects, he'll see people, he'll see pets, and what he expresses really depends on what happens to be in the environment that day and also, you know, what you've done with him in the recent past. And that's what ultimately will get expressed and it's something we don't necessarily even know ahead of time-- that isn't fully deterministic.
Can he speak?
He doesn't speak English per se, but he can attempt to say English words. So you can't have a conversation with him, but if he recognises your face, he will say your name. He'll get it out in a little bit of a awkward manner, but it's kind of endearing.
Right. And Cozmo's already been on sale in the US. I mean, how has the sales done?
So we're very lucky that Cozmo ended up, I think, being the number two bestselling toy in the US last year when it came out.
And how much was it retailing for in the US?
It was retailing for 180, and since then we've continued to grow the business. You know, we're launching globally in the UK, the rest of Europe, and also Asia as well.
And Anki is, as you were saying, a startup company, but you've got some pretty impressive backers. Could you tell us a bit about the corporate journey that you're on? How much have you raised and what are your ambitions next?
We've raised just about $200 million in capital. We're very lucky to work with some incredible investors. Andreessen Horowitz led our series A investment, followed by Index Ventures and also JP Morgan. And I think when we originally went on and started pitching for investment, really what we saw as an opportunity around the emerging industry of consumer robotics, and that's something that is much bigger than entertainment and that's only going to continue to grow. So our investors really believed and shared that vision, were willing to really take a long-term view of the company and recognise that this is not a three-year journey or a five-year journey, but in many cases, a 15- to 20-year journey and have been very committed to us and been great partners.
Right. And could you just tell us a bit about Anki Overdrive as well, the kind of first product that you launched? What is that?
So we launched what was first Anki Drive and then evolved into its kind of second generation called Anki Overdrive. And Overdrive is a smart racing game. So we try to take the Google autonomous self-driving car and then actually shrink it down to something that is 150th of the size and reinvent, really. A lot of folks are familiar with slot cars and, you know, very classic toy, very classic pay pattern, and we've tried to reinvent that with AI.
So the cars will come to life, they'll race against you, they'll try to knock you off the track or disable your car and really, you can imagine playing a video game but in physical form with a magnetic track that you can create, really, whatever type of environment that you want with. And we have people that put 20 pieces, 30 pieces, sometimes even 40 pieces and fill up the whole basement. So it's basically a racing game.
Right. And are there challenges? Do people challenge each other on these games or not?
Yeah, I think one of the things that's really exciting about it is the community that we've seen evolve around the game. And people create their own tournaments, they have their own tournament brackets. Multiplayer is a big part of it. But you'd also be surprised at a huge chunk of the pay is also people playing by themselves, so it's great because it works as a family game, but if your parents aren't available or if your brother's not available, there's still a tonne to do in the app even by yourself.
These products-- certainly the Anki Drive and Overdrive-- came about as a result, again, the smartphone revolution, because they obviously interact with the phone that you have in your hand. So is the consumer robotics revolution-- is it going to be very much enabled by integrated technologies such as that, or do you think they'll develop on their own?
I think you'll see both. I think smartphones definitely have had a huge impact on the space, not just for robotics, but more broadly in what we call the internet of things, and I think it's a big deal for a few reasons. Interface-- the fact that anyone who's used to using an iPhone or Android device can download an app and have an interface that they're already familiar with, that's a big component of it.
I think the second thing is really taking advantage of, in many cases, a very powerful processor that people already own and being able to offload the computation from the device-- from the robot or from the device to the phone really allows you to do much more sophisticated things but still maintain a very consumer mass market price point.
Where do you manufacture?
So we manufacture in a variety of locations. Our Cozmo robot is made in Asia and then we will ship it globally, so all over to Europe, US, and then other parts of Asia.
And you say this is a very young industry. As you're saying, robots have traditionally been used in a lot of industrial settings and so on. They're now being used with consumers. What new challenges does that present, do you think, to robot manufacturers?
I think if you look at industrial robotics, it's very, very different from any type of consumer application for a number of different reasons. So if you look to date, there's very little AI or learning in industrial application. If you think about the Tesla robotic manufacturing plant, there's perception and a number of other problems going on there, but the types of learning that those robots are doing, it's more about repeating the same action over and over and doing it with absolutely incredible precision. You know, that's very different from a robot like Cozmo that has to go into an unstructured environment, and what's on your kitchen table might totally change from Tuesday to Wednesday.
So really, I think the main difference with any type of consumer application is going to be dealing with the uncertainty of the world and adapting and changing.
Right. And there's a whole new academic discipline that's evolving off the back of this human robot interaction and a number of computer departments, like the former one that you were at at Carnegie Mellon and so on, right now employing all kinds of ethicists and philosophers or lawyers or anthropologists to try and understand how robots can interact with humans. How is that going to develop?
I think there are a lot of intersections between the fields that you just mentioned as well as psychology and you know, even in the products that we've developed, the team has learned quite a lot about what are the types of things that really resonate with a human being who's using it, and some of that is not necessarily obvious when you start.
And I'll give you an example just in Cozmo, like the eye gaze, and where is that robot looking, and how often does that robot actually turn in your direction and look up and try to look in your eye. And that's something that we didn't think about and then ended up being a very powerful moment. And then we've really tried to kind of lean into that and you know, we've added quite a lot in terms of those types of things that really help create that empathy in the bond.
A number of lawyers have also kind of weighed in on the whole robot debate, arguing that there's quite a difference now between robots as they have operated historically, where they've basically been kind of programmable tools with very predictable behaviour, to robots that are beginning to exhibit emergent behaviour and behave in unpredictable ways, ways that we might not even be able to explain or understand. How much of an issue is that?
I think it's certainly overblown a bit. All robots-- really, all AI-- operates under objectives and constraints. Just because a robot is able to deal with an unstructured environment and maybe the environment changing from day to day, and its behaviour might adapt to that unstructured environment, that doesn't necessarily mean that the robot's all of a sudden going to decide to do something nefarious one day and then the day after that, you know, he'll become a good Samaritan.
So that's where I think there's quite a lot of confusion in the mainstream press that I think will certainly change as people become more and more familiar with these types of devices, as they get smarter, when they realise that the sort of doomsday scenarios that people talk about are often significantly overblown.
They're not necessarily doomsday scenarios, though, are they? I mean, just some very practical issues with self-driving cars, for example, that when they interact with other cars or they have accidents or have to make life or death decisions, we want to understand why they've taken the decisions that they have taken. Is that not a real immediate issue for us?
Well, I certainly think that we need safety standards, right, and I think something like autonomous cars should certainly go through a very rigorous battery of tests to make sure that it performs to some sort of functional standard. But I would also argue that people don't really understand how the human mind works, and we're very content to get in the back of a taxi or get on an airplane and not really fundamentally understand the math of what's actually going on in the brain. And we're OK with that, because again, there's functional standards and people have to perform to some level of proficiency, and I think we can hold our robots to the same type of account.
Do you think there ought to be a federal robotics commission in the States?
I think yes, I do think that a primary benefit is ensuring uniform law. In a number of different areas, you'll find that in many cases, the law can get very fragmented and to give you an example, in computer vision, you know, there are states like Illinois, which happen to have their own regulations that are different from California and different from other states. And I think particularly as an emerging industry, we want to avoid a patchwork of laws and I think one of the benefits of thinking about this at the federal level is really ensuring that if you adhere to a standard, that that standard applies across the entire country and you're not worried about a human taking over when the car drives from California and Nevada.
Do you think there are privacy issues to do with consumer robotics as well?
I think so, and in the States, for example, we have a law called COPA, which is the Child Online Protection Act, and that is really targeted at kids under 13. And the privacy standards continue to evolve, and you've seen new products like Google Home Assistant or Amazon Alexa, where a lot of that competition is happening in the cloud and you need to have very, very strict guidelines around how that data is retained, when a user can choose to delete it, and fundamentally that they have full disclosure about what's going on. And ultimately, the consumer has to maintain control of their data.
So what have you done with Cozmo, could you tell us-- is that recording data? I mean, it's obviously analyzing--
So Cozmo is not online and, you know, that was a very explicit decision that we made, not only from a performance standard, but really, it helps the privacy question as well. And all data stops at the phone, so you make a direct connection between the phone and the robot-- the robot is not online-- and when the phone is talking to the robot, the phone is not online, either. And we also don't send any what we call PI, personally identifiable, information. So that's a big part of it, but that also helps from a security standard as well, because if the robot's not online, then no one can hack into the robot.
Now, you were saying earlier that you are just at the beginning of this consumer robot revolution and that you think you're on a 10, 15-year journey for how this is going to play out. Can you sketch out in some kind of futuristic way where this is going to end up? What do you imagine is going to be possible?
I think a few things. So I think that devices are going to continually get smarter. I think things that used to be hard problems, so take voice recognition, for example-- just a few years ago very, very few people were actually using it in any sort of daily capacity. In the last one to two years, it has made incredible advancements. So I think interfaces are going to be much, much more driven by natural voice, natural speaking. You know, there's a whole field called natural language processing, which works in concert with speech recognition. So I think that's going to be a big deal.
I do think that devices are going to become more autonomous as computation gets cheaper. You're not necessarily going to require the smart phone connection anymore. I think we're at a period where that makes a lot of sense, but I do see a future where we're going to want our devices to actually not require that and there will be Interface benefits to that, and we'll still be able to hit a price point that makes sense. I'm also very excited about what we call cloud AI or cloud robotics, which is an emerging area, and in that scenario, you have robots that are online that do have connection, in many cases, to very sophisticated AI hardware that is running in public clouds like Google, Amazon, and that will extend capabilities far beyond what you could ever do on just the robot alone.
And for Anki-- I mean, are you going to put most of your emphasis on consumer robots? Are you going to move into educational, domestic robots? Where are the limits for you as a company?
Our focus right now is consumers, and we've been building a platform to solve a lot of the fundamental challenges of robotics-- you know, specifically, positioning a robot in space, robot reasoning, as well as robot control, and we've been building that platform and we continue to expand on it, but really focused on consumer applications. We see many more robots around the home, not just for entertainment but really for more functional purposes as well.
I was reading an academic who thought that the ultimate test for robots would be if they could read the instructions on how to assemble furniture and then to assemble the furniture itself. How long do you think it's going to be before you get a robot to do that?
I still have trouble with some of my Ikea furniture, so we're probably-- we've passed the Turing test at that point and probably a few other tests. But I don't think it's crazy to think that within 10 years or 20 years. One area I forgot to mention earlier is really around robot manipulation-- having a robot with a hand pick up an object in an unstructured environment, where the robot doesn't know ahead of time what that object is. That is one of the most incredibly difficult challenges overall in the field of robotics.
However, in the last year or two, that's been a big area of improvement, and I think over the next five years I'd maybe go so far as to say it might be a solved problem, and if you have robots that can actually manipulate that easily, then I think you're bound to see incredible number of functional and utility applications that weren't possible today.
Right. And finally, I'd just like to ask about the kind of emotional intelligence of robots, as it were. I mean, that's one of the elements that you're talking about in Cozmo, that it can recognise facial expressions and respond to them. How far do you think that is going to go? How clever are robots are going to get in understanding human emotions and responding?
I think it will go just beyond human emotion. I think that robots will have much better, I want to say, contextual or semantic understanding of the environment. So being able to determine the mood or the vibe in a particular place by reading the expression not just off one individual's face, but really looking at a group of people, and what everyone else is feeling and expressing at the same time and really fitting into that.
A simple example, right-- if you can imagine people watching a sports game where everyone starts cheering and the robot's aware that there's a game going on and that people are excited, and he could get excited as well. And I think that's something that we can already do today, and I think we'll certainly see more and more of that contextually aware robotics occur over the next few years.
And where is Cozmo going to be launched? You're launching in the UK, but where else?
We're launching in the UK as well as broader Europe, so Germany, France, Nordics. We're also launching in Japan at the same time as well, so Tokyo and then other parts of Asia. So Australia is another market that we're very excited about.
Great. Well, Mark, thank you very much for the discussion.
Thank you so much. Great to be here.
We'll be back next week with another episode of Tech Tonic. In the meantime, if you'd like to comment on today's show or suggest any topics you'd like us to cover in future episodes, please email us at techtonic@ft.com. Don't forget to subscribe to our show on your favourite podcast app, and if you write a review that will help other people find us, too. Thanks for listening. This episode of Tech Tonic was produced by Fiona Symon.