The following is a conversation between Dr. Louis Rosenberg, Founder and CEO of Unanimous AI, and Denver Frederick, Host of The Business of Giving on AM 970 The Answer in New York City.


 

Dr. Louis Rosenberg © unanimous.ai

Denver: Are you concerned about how human beings will adapt to a world of artificial intelligence, robotics and all the other technological changes that face humanity in the next 25-, 50- or a hundred years? Will we be eclipsed by artificial intelligence? One person who has given this an awful lot of thought is Dr. Louis Rosenberg, the Founder and CEO of a Silicon Valley startup called Unanimous AI, and he is with us now.

Good evening, Louis, and welcome to The Business of Giving!

Louis: Yeah. Thanks for having me!

 

… social species like fish and birds and bees have enabled another type of intelligence, which biologists refer to as “swarm intelligence,” where once they’ve developed their individual brains to a certain level, they start to work together in systems. And when they work in systems and solve problems in systems, they become much, much smarter.

Denver: This has been an issue for you for some, I don’t know… 25 years, whether humans would lose their position as the top intelligence in our environment. And you were pretty concerned that there might not be an answer for this until you looked at other species. What did you discover?

Louis: It’s really interesting that there are a lot of species who go through a period where they need to get smarter to overcome some challenge in their environment, and they need to get smarter fast. And normally, when we think about evolution, we think about the development of a brain. And a brain is a network of neurons and obviously, the larger and larger that a brain gets, the smarter the species can be. But that takes a really, really long time.

What happens is that social species like fish and birds and bees have enabled another type of intelligence, which biologists refer to as “swarm intelligence,” where once they’ve developed their individual brains to a certain level, they start to work together in systems. And when they work in systems and solve problems in systems, they become much, much smarter. And this is ultimately why birds form flocks and bees form swarms and fish form schools; they get smarter together. And the premise of the work that I do and our company, Unanimous AI, is to say, “Well, if birds and bees and fish can do this, why can’t people do it? Why can’t we amplify our intelligence by thinking together in systems?” And one of the motivations for wanting to do this is that artificial intelligence systems are coming right down the street and will soon challenge us.

Denver: Let me ask you this: How do bees do it? How do they communicate and interact in a way that allows them to converge on a good solution?

Louis: Each species has evolved its own mechanism doing it. Fish, for example, can detect vibrations in the water beside them, and birds do similar things. The species that actually has understood the best of all are bees, and particularly honeybees because biologists have been studying them for over 50 years. And what honeybees do to amplify their intelligence and the way they form a system is actually by vibrating their bodies, and biologists call this a waggle dance. Because to people, it looks like the bees are dancing, but really, they’re generating signals that represent their preference for various problems they are trying to solve. And when all the bees are vibrating their bodies, they’re basically engaging in a multidirectional tug-of-war. They’re pushing and pulling on the problem until they have converged on the one solution that they can best agree upon. It’s usually the optimal solution. It’s also the solution that’s best for the group as a whole. And so they’ve developed a method that not only amplifies their intelligence in solving problems better intellectually, they’ve also developed a method that points them to solutions that are best for the greater good of the colony.

A swarm – when a group converges together on an answer – is a better way to combine the knowledge and wisdom and insights and intuition than simply taking a vote.

Denver: That is incredibly cool. I know there’s a field of research called collective intelligence or crowd intelligence. How does swarm intelligence differ from that?

Louis: Collective intelligence is, I would say,  the umbrella concept that the general area of collective intelligence is saying:  groups can be smarter than individuals. And people have known about this property of collective intelligence for over 100 years. It goes all the way back to 1907. There was an early experiment of estimating the weight of an ox in a county fair in England. And what they discovered was that even though they had hundreds of farmers make this estimate… and what they found was even though the individuals were actually not so great at making these estimates, when you took the average, the average was really, really close. And so, the field of collective intelligence is kind of this broad idea of getting to where the group is smarter than the individual.

Crowd intelligence usually refers to treating each individual as a piece of data that you can collect in a poll or survey, and there is good evidence that if you take a poll of a large group of people, you will get an answer that is smarter than the individuals who make up that poll.

Swarm intelligence takes that to another level, and it’s really inspired by nature. Because these natural systems like birds and bees and fish, they don’t take a poll, they don’t take a survey, they don’t run a focus group. What they do is they form a system – a real-time system where they’re all participating at the exact same time. They’re pushing and pulling off of each other that they can converge together on an answer. And it turns out,  at least from the research that we do and research that others do, is that nature has the better method. A swarm – when a group converges together on an answer – is a better way to combine the knowledge and wisdom and insights and intuition than simply taking a vote. And what we see is that if we allow people to do that, they can significantly outperform. They can make very, very accurate predictions by using swarms.

Denver: How does a swarm work? How many people do you need to make it effective? How long does it take?  And what are people doing during it?

Louis: The first challenge that we faced when we started working on this years ago is that we know how bees do it, but humans can’t waggle dance. So we need a different type of interface. And so, we developed an interface where people can log into a server from anywhere in the world, and we have a visual interface where they each can control what looks like a little magnet to convey their will on the swarm. And so each person controls a little magnet with a mouse or a touch screen, and they can basically pull the swarm in a direction that they want, but everybody else also controls a magnet. And so, they’re all pulling against each other to guide the system to an answer. And it was designed in a way to really emulate the same kinds of input that birds and bees and fish do when they form swarms. And what we found is that when we enable groups of people to connect from all over the country or all over the world… and we ask them to answer questions or make predictions as a swarm, they will converge together on much more accurate answers.

And so, one example of that—a lot of people have heard of this. It got a lot of press last year— was that we were challenged by CBS Interactive to predict the Kentucky Derby this way. And so we as a company don’t know anything about horse racing; we had never predicted something like that before…but we believe in the power of a swarm. And so what we did was that we got a group of 20 horse racing enthusiasts from around the country who just logged into our server at the same time, and we asked them together to predict the race. Now, CBS didn’t just want us to predict the winner. They wanted us to predict the first four horses in order, which in horse racing is called the “superfecta.” And last year, it went off at 540 to 1 odds.

So, it was a big challenge, and we were really just hoping to do as best as we could. So we had the swarm, come in, they made their predictions for the four horses. We gave that to the reporter that wrote an article. In fact, she went to the Kentucky Derby, and she placed a bet on the superfecta, and she tweeted out her ticket, which added to the pressure on us. And that turned out that the horses finished in exactly that order. So, anybody who had placed a $20 bet on those four horses would have won $11,000.

Denver: Which included you by the way, right?

Louis: Which included me. Fortunately, I placed a bet. I would have been kicking myself if I didn’t. The reporter placed a bet. A bunch of our readers placed bets. One of her readers reported winning $50,000 so it was a pretty amazing result. But to us, when we look at it scientifically, the more interesting thing is not just the final result, which always involves some luck because it’s horse racing, but the interesting thing was that the swarm got that answer perfect. All four horses correct. But if we look at the individuals, those 20 people who participated in the swarm, before they participated in the swarm, we asked for their individual predictions and so we could compare. And not a single one of those 20 people got all four horses correct on their own. And in fact, had they just taken a vote, a standard majority vote, they would’ve only gotten one horse correct out of the four. But when they worked together in real-time as a system, as a swarm, they converged on the correct four horses. And it really goes to this point that a swarm is really a much better way to aggregate the knowledge and wisdom and insights of a group than the traditional way people do, which is by taking a vote.

Denver: Sounds like a superintelligence to me in many ways. And you’re really amplifying the knowledge of those individuals. I don’t know to what degree. Maybe you do. How much smarter is a swarm than the people who make up the swarm?

Louis: It’s a great question, and it varies depending on the question that we’re asking and the population that we’re bringing into the swarm. But we just did a study with some researchers at Oxford University where we looked at predicting English Premier League Soccer games. And we did it actually over 50 games. So, we had a group with predictions over five weeks, predicting all 50 games during those five weeks. And what we found was that by comparing majority vote to a swarm, the swarm was 130% more accurate. And so it was really a significantly better way to tap the intelligence of the group than the way people traditionally try to get input from the population.

Denver: Without getting too deep into the woods on a subject that most of us don’t understand, Louis, how do these artificial intelligence algorithms work in relation to the people who are swarming?

Louis: So, to the people who participated, it looks and feels very simple because they’re just expressing how they feel,  but not just voting. They’re controlling their little magnet and pulling on the system, and they’re continuously doing that from across the entire question. Now when we ask a question like “Who’s going to win? The Cubs or the Dodgers?” for example. That question will get answered in 60 seconds or less by the swarm. But the participants are varying their opinions. They’re moving their magnets every millisecond during that response. And so our algorithms don’t just look at a single vote. They are looking at the behaviors of the people. And so, while a survey captures single data points, we’re capturing behaviors of every participant; and we can, from those behaviors, really know which people are confident and which people are not confident, which people have high conviction in an answer, which don’t. And if there are multiple options, we can know which of those options do people have confidence in, and which don’t they have confidence in. And because when we ask somebody a question: “Who’s going to win the World Series?” and you have eight teams in contention, people have pretty complicated feelings about that. You can ask them for a single data point of who they think is going to win, but if you really capture their behaviors, you start to know the trade-offs that they’re thinking about. They might think “Yeah, these two teams have strong chance. The other teams have a lower chance.” And so, the inner feelings that people have, their inner knowledge and intuition, is much more sophisticated than a survey can capture. So, what a swarm does is it really allows the people to express their true knowledge and their true wisdom in a way that gets captured.

Denver: Yeah. You really get the weights of how they feel about these things rather than just the absolute answer. Well, aside from the sporting world, who out there right now is using swarm intelligence?  And what has the impact been?

Louis: We do a lot of predictions in sports because it’s a great way to just test the power of the technology because you can get an answer the next day– to lose or not, the swarm was correct or not. That said, we also do a lot of projects for large companies that want to tap the intelligence of, for example, consumers to understand how consumers would react to a new product or a new feature or a new marketing message. And the thing that we do that’s really different is that we’re not asking the consumers for their opinion. We’re asking the consumers for their intelligence. And so we can have a group of 50 movie fans watch a movie trailer, and we’re not going to ask them, “Would you go see this movie?” We’re going to ask them, “Do you think that this movie trailer will drive people that you know to the movie theatre?”

So what we really do is we’re building an intelligence that acts as an artificial expert, that has better intuition, better insight than you can have if you were just asking an individual marketing executive for their intuition as to whether or not that movie trailer would work or not. And what we find is that by pulling together a group of 50 movie fans, we can build an artificial expert that has very, very strong intuition about how a movie will perform. We can do the same thing for groups of sales forecasters, people who will predict whether a product will sell. They can amplify their intelligence when it comes to doing a sales forecast, and we could do the same thing for really any type of decision where there’s human judgment and knowledge and intuition that can play a role; we can amplify those things and extract the highest input we can.

… what we find is that because everyone is equal in a swarm and everyone is anonymous, that people feel that their opinion counts, and they feel more open and honest. They feel like they don’t have to second guess what their boss wants them to say. They can actually express their opinions honestly.

Denver: One of our favorite topics on The Business of Giving is Organizational Culture. And I’d be curious if swarm intelligence has influenced and helped shape the corporate culture at Unanimous AI and whether you think there are principles and lessons from that that could be applied to other organizations’ work cultures?

Louis: That’s actually a great question. And one of the things that we try to do a lot of is actually use swarm intelligence to make decisions and make predictions for our own business. And what we find is that when a group of us get together online and answer questions as a swarm, we get new insight into how the whole group thinks and feels because of the very equal and flat process. And what we hear from our own group or from other companies that we do this with, what we find is that because everyone is equal in a swarm and everyone is anonymous, that people feel that their opinion counts and they feel more open and honest. They feel like they don’t have to second guess what their boss wants them to say. They can actually express their opinions honestly.

We actually did a project last year with the US Navy where they were looking at some issues, and they obviously have a very strict hierarchy, but they still want to be able to get the intelligence of a whole group. And what we found in getting feedback was that the fact that people felt like rank didn’t matter, they were all equal, they were all anonymous, they could give more honest input than they could have in other manners.

Denver: Just like the bees. Let’s turn to the social sector for a moment. Recently, Jeff Bezos of Amazon sought advice on how to spend part of his fortune to do social good in the world. And Unanimous AI formed a swarm to see what it thought. What were the recommendations to Mr. Bezos on where he should direct his philanthropy?

Louis: Jeff Bezos had put that request out, and lots of people responded on Twitter. In fact, he collected, I think, 47,000 responses on Twitter. And so what we did was we took those 47,000 responses… which were basically ideas that came from the population, and we filtered those down… because there were a lot of overlaps and a lot of duplicates… to a smaller set.  And in a group of 100 randomly selected Americans, we had them form a swarm and we had them go through and rank the suggestions with the goal of trying to find what would be the single philanthropic cause that the population could best agree upon.

And what was interesting was that the group converged very, very strongly– after going through lots and lots of options– on universal access to clean drinking water for everybody in the world as the number one best use of the philanthropy. And it actually surprised us because there were a lot of options on the list that would have appealed maybe stronger to Americans. Again, the population was all American citizens, and there were options on the list that did well… like lower drug costs, or lower healthcare costs, which would obviously affect Americans directly. Clean drinking water is actually one of the few on the list that really is not a problem in the US, or very rarely a problem in the US, and yet that’s what the swarm converged on. And it actually goes to one of the really powerful things about a swarm is that it evokes a far more selfless decision from a group than if you take a vote. And we’ve done studies that show that you can take the same population, have them take a vote, and they will converge on answers that reflect their own personal interests, stronger than interests of the group as a whole; whereas a swarm will actually evoke the answers that are best for the group as a whole and drive a greater sense of selflessness. And that’s exactly what we saw in this example where clean drinking water is a global issue, and it was the answer that had emerged very, very strongly. It wasn’t even close compared to the next highest option.

Denver: Well, Louis, do us all a favor and get some swarms going in Washington D.C., okay? Finally, Louis, the social sector tackles some of the most intractable and vexing problems on the planet — health and poverty and sustainability and inequality. Is anyone doing anything, or are there any plans to utilize swarm intelligence to come up with new and better solutions to these problems?

Louis: There are. We do projects that are in the social space. In fact, we just did a project actually with the XPRIZE Foundation where they obviously do a lot of thinking about: what are the big problems that face humanity, and there’s a lot of problems. And one of the things that they work hard to do is to make sure that they devote resources on the most significant problems that could have the biggest impact. So, we’ve started working with them on that and basically had swarms of their experts ponder these issues and converge on the important problems to solve in future XPRIZE competitions.

Denver: Yeah. I know the next one is going to be the early detection of Alzheimer’s.

Louis: Yes.

Denver: Well, that’s great. Well, Louis Rosenberg, the founder and CEO of Unanimous AI, I want to thank you so much for joining us this evening. Tell us about your website and what listeners will find there. I know it comes with a lot of sports predictions. What else is on it?

Louis: So our website is unanimous.ai. On our website, we do weekly sports predictions for baseball and football and English Premier League Soccer. We’re starting up NHL and we’ll also do NBA. And that’s really a fun way for us to demonstrate the power of a swarm intelligence, which is really a way to demonstrate the power of how smart people are out in the population. So, that’s on our website.

Also on our website, we do do interesting political swarms where we tap the sentiment of the population on various political issues. Also on our website, we have information about how companies that want to use swarm intelligence to do things like marketing forecasts or sales forecasts, how we work with those companies to do that. And there’s also academic papers that show the actual nuts and bolts of how we can amplify intelligence and what kind of results we’re seeing.

There’s also a TED Talk that I did on the power of swarm intelligence, and it’s a good way to get an understanding of how it works because it shows visually what these swarms look like.  And it’s always much easier to understand what we mean when we say: a swarm when you can actually just see what it looks like.

Denver: Thanks, Louis. It was a real pleasure to have you on the program!

Louis: Yeah. Thank you. It was fun.


The Business of Giving can be heard every Sunday evening between 6:00 p.m. and 7:00 p.m. Eastern on AM 970 The Answer in New York and on iHeartRadio. You can follow us @bizofgive on Twitter, @bizofgive on Instagram and at http://www.facebook.com/BusinessOfGiving

Share This: