Jim Freeze Hi and welcome. I’m Jim Freeze, and this is The ConversAItion, a podcast airing viewpoints on the impact of artificial intelligence on business and society.
The ConversAItion is presented by Interactions, a conversational AI company that builds intelligent virtual assistants capable of human level communication and understanding. In this episode, we’ll discuss the ways in which AI can amplify human intelligence. We’re joined by Dr. Louis Rosenberg, an accomplished technologist and prolific inventor.
In 1992 Louis’ doctoral work at Stanford resulted in the world’s first immersive augmented reality system. He later founded a company that developed the first desktop 3D digitizer to allow animators to create 3D computer models of physical objects. That technology was used in the production of movies like Shrek, Ice Age, A Bug’s Life, and one of my favorites, Titanic.
Today, Louis is the founder and CEO of Unanimous AI, a tech company that amplifies the intelligence of human groups using AI algorithms modeled after swarms in nature.
Louis, welcome to The ConversAItion.
Louis Rosenberg Yeah, thanks for having me.
Jim Freeze Yeah, we’re thrilled to have you here. To kick us off, can you take us through a little history of Unanimous AI? What inspired you to found the company back in 2014, and how has the solution evolved since then?
Louis Rosenberg Yeah, so my background is really in human computer interaction and how to use technology to make people perform better. And for a number of decades I really focused on technologies that could make individuals perform better. And as you mentioned, I worked in the fields of augmented reality and virtual reality.
And then about 10 years ago I started wondering, well, while technology can make individuals perform better, what happens when you start looking at groups of people? And that’s when I started asking the question, are we really following the best methods for harnessing the intelligence of groups? And like a lot of technologies, one of the best places to look is nature.
So I spent a long time looking at, well, how does nature access, and amplify, and harness the intelligence of groups? And I found that, A, nature is remarkable in how it does it, and B, it does it very differently than humans do it. And what nature does is it forms swarms. Biologists have been studying this for 50 years. They describe it as swarm intelligence.
And that’s the reason why birds flock, and fish school, and bees swarm. They become significantly smarter when they work together in these real time closed loop systems. And so Unanimous AI was founded on a really simple question, which was, hey, if birds and bees and fish can get so much smarter by forming these real time systems, maybe people can do it.
And so we started building technology to see if that was possible. And it’s five years later now, but even from the earliest prototypes we were amazed. We realized, yeah, nature’s right. We can make groups of people significantly smarter if we connect them together in systems. And if we use AI as the core of that system, it works really, really well.
Jim Freeze It’s really interesting. When people talk to me and ask me about AI and what AI really is, I characterize it, it’s a number of different things. It’s a lot of data, it’s algebra, and it’s statistics. And I’m just thinking about that definition that I give people in the context of what you’re doing.
I’ve read some interesting tests and applications of your technology—you’ve had some headline grabbing predictions at major events like the Oscars and elections. And in particular in 2016 you used your platform, you were able to predict the top four horses at the Kentucky Derby and in the right order, which is pretty amazing and maybe even a little scary.
Using that Kentucky Derby question as an example, could you walk us through how a group of people arrive at an outcome using your platform? And once again I’m thinking about that definition of data, of statistics, of algebra essentially. How does that all work?
Louis Rosenberg Right.
Jim Freeze And it’s fascinating.
Louis Rosenberg Yeah, so I agree with you, most AI that people are working on is really about having a large set of data in some database somewhere and then performing algorithms on that data, which is really statistics and finding patterns in that data. And that works for a lot of things. Traditional AI has all kinds of power and use, but it’s most useful when you have a dataset that is really, really big, and it doesn’t get out of date, like forecasting the weather, right?
You could have a massive database about weather data and it’ll slowly change over time. The problem is for a lot of issues that people are to solve, especially in business, these huge datasets don’t necessarily exist, or if they do exist, they aren’t necessarily up to date. You might have data about products from last season, but it doesn’t mean that data represents the new set of products that are coming out this season.
And, in fact, when the issues in question involve people, the data usually gets out of date even faster. And so, one of the things that we do at Unanimous AI is say, hey, there’s the data that exists in a big database somewhere, but there’s this other database that AI is ignoring, which is all the information that exists inside of people’s heads.
Inside of human heads there is massive amounts of data, and in fact we start with the premise that, hey, people are really smart. People have knowledge, and wisdom, and insight, and intuition that is really, really powerful, and it’s being ignored by AI. And, in fact, the amazing thing about people is that we’re out there continuously updating our database. We are out there in the world seeing, and collecting, and analyzing.
And so, Unanimous AI, what we do is instead of performing AI on a database of static data collected somewhere, we instead connect groups of people together in real time. And so when I say real time I mean it. It’s not like a survey where you collect data and then you process it after the fact, when I say real time I mean if we’re going to bring a group of people together, they’re all going to connect at the same time.
And so I’ll use the example of the Kentucky Derby like you mentioned. One of the things that we do is we say we can take a group of people and we can connect them with AI and turn them into an artificial expert and make them smarter. And so journalists very often say, okay, that’s a bold claim, here’s a challenge. And so we get lots of challenges, whether it’d be the Kentucky Derby, the Oscars, or the Super Bowl.
And so a couple of years ago CBS Interactive challenged us to predict the Kentucky Derby. And they said, don’t just predict the winter, predict the first, second, third, and fourth. And in horse racing that’s called the superfecta. And it’s a very, very difficult thing to predict.
The year that we were looking at it, the odds were 540 to one of being able to predict the four horses in order. And so we said, okay, well, we’ll take that challenge. Now I’ll mention that we didn’t know anything about horse racing, we didn’t have any data about horse racing. All we had is the ability to tap the intelligence of a group and amplify it as a swarm. And so what we did is we found 20 horse racing enthusiasts.
Jim Freeze Well, you’re actually getting at what I was going to ask you next, which is how do you identify the folks that go into creating a reliable group. And there has to be some domain knowledge?
Louis Rosenberg Yes. So typically we want to find a group of people who know something about the topic in question. And the more they know the better, because what we found statistically is that if we take a group of people we can make them significantly smarter than that group would have done on their own. And, in fact, we typically, when we connect with people into a swarm, that swarm will usually beat 90% to 100% of the individuals.
Jim Freeze Wow.
Louis Rosenberg Which means if I have a group of people who are really smart, I can make a super expert. If I have a group of people who are novices, I can turn them into an expert. And I can give some examples of that. But in this example of the Kentucky Derby we found people who were enthusiasts. They weren’t professionals, they were enthusiasts, they all followed horse racing, and we found them just by putting an ad out online and getting 20 people connect, and responded.
And the way our system works is people can be anywhere, they just have to log into our platform which is called Swarm. And so we send out this invitation, they connected to Swarm, and now we’re going to ask them to predict each of the four horses. And what happens is we say, who’s going to come in first, who’s going to come in second, who’s going to come in third, who’s going to come in fourth?
And now each time we ask a question it appears on all their screens at the exact same time and they get an interesting user interface that pops up. A lot of people say it reminds them of a Ouija board because we’re giving each person the ability with their mouse to control a little magnet on the screen. And they’re moving the magnet basically to engage in this multi-directional tug of war.
And so some of the people are pulling towards one horse, some of the people are pulling towards another horse. They’re pushing and pulling and the AI algorithms are watching all of their behaviors and determining how this swarm should move. And in a lot of cases what it looks like on the screen is almost like a swarm of bees or a flock of birds. And so they’re basically pushing and pulling with the AI moderating them and they converge on an answer.
And so in this example they went and they predicted first, second, third, fourth. It took them about five minutes to do it. We gave the predictions to the reporter. And the reporter actually went to the Kentucky Derby and she placed a bet on the superfecta and she tweeted out her ticket, which put a lot of pressure on us. So now, she tweeted out our ticket.
And the reason that this got a lot of attention is that the swarm was perfect. And so anybody who had placed a $20 bet on those four horses would have won $11,000, which I did. I’ve placed a $20 bet.
Jim Freeze I was going to ask you if you made money off of this, but you should have done more and then you could have done your next round of funding.
Louis Rosenberg I know. So I placed a bet, the reporter placed a bet, a bunch of her readers placed bets. In fact, one of her readers reported winning $50,000, which is an amazing story. But what’s actually the most interesting thing is that you can then go back and look at those 20 people who predicted, because we had the data about how they would have done as individuals. And as individuals, not a single one of those 20 people would’ve gotten all four horses correct.
And in fact, had they just taken a vote, which is the traditional way people harness the intelligence of groups, they would’ve gotten one horse right out of four, so they wouldn’t have won anything. But when they worked together as a swarm they were perfect. And so that’s really the power of swarm intelligence, is we can connect groups of people together. And we can allow them to combine their knowledge, and wisdom, and insight, and intuition in an optimal way to get the most out of their perspectives.
And the interesting thing is that our AI algorithms are trained on human behaviors. We didn’t train on any information about horse racing, we trained on how people will interact when their various levels of confidence and conviction… It’s basically saying everybody is wrestling with this problem, they’re all interacting, and we’re watching their behaviors, and we’ve trained on their behaviors.
So it doesn’t matter if I ask them a question about horse racing or if I ask them a question about forecasting the stock market, or ask them a question about diagnosing a medical problem. The question can be anything as long as the participants are behaving like humans.
Jim Freeze Now, betting or predicting sporting events is interesting, but I’ve got to believe that there are some really profound and very important future applications or even maybe current applications of the technology that perhaps in medicine as an example.
Louis Rosenberg Sure. So we’ve looked at a really wide range of applications, and in fact, we work with large companies and universities across a variety of things from forecasting sales to analyzing and predicting which marketing messages are going to work, to, we work with hedge funds to predict the stock market. But the most interesting is in medicine where, can we take a group of medical professionals, a group of doctors, and make them smarter?
And so we formed a partnership with Stanford Medical School and performed a study where we looked at radiology. Can we have groups of radiologists look at chest X-rays? And this was the test example, look at chest X-rays and diagnose them either the traditional way, which is that a chest X-ray comes up on the screen and a radiologist looks at it and then makes a diagnosis, or have a group do that by vote, or have a group through that as a swarm.
And so what we did in this study was we had a group of, with small groups of four or five radiologists, they were all at their own workstations, some were at Stanford, some were at Duke, some were at other universities, and an X-ray would pop up on their screen. They would make the diagnosis together as a swarm.
And what we found was that when they work together as a swarm they were 33% more accurate than when they did it as individuals, which was really, really surprising to us actually, because these individuals are already experts, right? Unlike the horse racing enthusiasts, here we have individuals who are… They’ve gone to school for 12 years to become a radiologist. Can you even make them better? And yet when we connected these groups of people together their diagnostic accuracy went up by 33%.
And so it demonstrates that the exact same algorithms that work for a group of horse racing fans works for a group of radiologists.
Jim Freeze Yeah. And when you think about it, maybe the results were surprising but as I hear you talk about it, I guess it’s not surprising. Because you’re tapping in once again to that data source, which is the minds and the experience and the intuition of all these experts. And you’ve gotta believe that delivers a better result, and obviously it does. It’s interesting. You did a TED talk in 2017, and this is my final question. It really relates to using that human intelligence in your system maybe to solve some big societal problems. In the same way that a school of fish is able to optimize the way they work themselves through the ocean, maybe there’s a way that, and I think you talked about it or touched on it in your TED talk, that you can apply your platform to improve the human experience or maybe even solve some big societal problems.
Louis Rosenberg Absolutely. And that’s one of the things that we’re very, very interested in, which is how can swarms of people reach better decisions. And it’s very easy to test a problem where there’s a right or wrong answer, and can you predict who’s going to win the Super Bowl or not? But when you have a more political type of question where you have groups of people who have opposing views, and does a swarm enable them to reach a decision that satisfies the population in a better way?
And again, this goes back to nature, which is, in nature you don’t see a school of fish suddenly break into two because they can’t agree on which way to go. And maybe 100 million years ago that happened and all those schools of fish died out, and this process of swarming, it actually evolved so that it amplifies places where groups agree, it actually exposes where they agree, and it actually helps a group find common ground which helps reach decisions on political questions or societal questions that are best for the group as a whole.
And we actually just did a large research project, it was actually in the UK, where we’re looking at Brexit. And we did a study of can we have, which is, in the UK Brexit was this super polarizing issue. And what we found was that if you took a group and you had them just take a vote, they could never reach a decision and they would always reach an impasse, but when they worked together as a swarm, they would converge on solutions that would maximize their collective satisfaction.
And so it was a better way to get a group to find solutions that they can best agree upon. And it goes back to the traditional way that political groups reach a decision now, is by basically taking a poll. And a poll is polarizing. That’s what a poll is. I mean, what a poll does is it reveals the differences in opinion whereas what a swarm does is it actually reveals the commonality in a group, and it allows the group to follow that path.
And so we have a lot of hopes for swarming being used in political and societal decision making.
Jim Freeze Well, heaven knows we could use that in our country right now. So, Louis, this has been fantastic. I’ve learned a lot, and I’m sure our audience is really going to appreciate this episode. Thank you so much. It’s been a pleasure chatting with you.
Louis Rosenberg Yeah, it was fun. Thanks for having me.
Jim Freeze All right. Take care.
On the next episode of The ConversAItion, we’ll talk about how robots can co-exist with humans. We’ll speak with Dr. Michael Littman, a computer scientist and professor who co-directs Brown University’s Humanity-Centered Robotics Intitiative.
This episode of The ConversAItion podcast was recorded at the PRX Podcast Garage in Boston, and produced by Interactions, a Boston-area conversational AI company.
Well, that’s the end of today’s ConversAItion. I’m Jim Freeze, signing off, we’ll see you next time.