Artificial intelligence is all the rage, but using swarm intelligence might be the best way to solve the world's biggest problems. Dr. Louis Rosenberg is the Founder & CEO of Unanimous AI, an artificial intelligence company that amplifies human intelligence by building "hive minds" modeled after biological swarms. Learn how swarm intelligence can combine the brainpower of humans and computers to solve humanity's biggest problems. Additional resources discussed: What is Swarm AI? Sports Predictions using Swarm Intelligence Business uses of Swarm Intelligence New hope for humans in an A.I. world | Louis Rosenberg | TEDxKC Waggle Dance Distances as Integrative Indicators of Seasonal Foraging ChallengesIndependence and interdependence in collective decision making: an agent-based model of nest-site choice by honeybee swarms | National Center for Biotechnology Information That "Old Book Smell" Is a Mix of Grass and Vanilla | Smithsonian.com Follow Curiosity Daily on your favorite podcast app to get smarter withCody Gough andAshley Hamer — for free! Still curious? Get exclusive science shows, nature documentaries, and more real-life entertainment on discovery+! Go to https://discoveryplus.com/curiosity to start your 7-day free trial. discovery+ is currently only available for US subscribers.
Artificial intelligence is all the rage, but using swarm intelligence might be the best way to solve the world's biggest problems. Dr. Louis Rosenberg is the Founder & CEO of Unanimous AI, an artificial intelligence company that amplifies human intelligence by building "hive minds" modeled after biological swarms. Learn how swarm intelligence can combine the brainpower of humans and computers to solve humanity's biggest problems.
Additional resources discussed:
Follow Curiosity Daily on your favorite podcast app to get smarter with Cody Gough and Ashley Hamer — for free! Still curious? Get exclusive science shows, nature documentaries, and more real-life entertainment on discovery+! Go to https://discoveryplus.com/curiosity to start your 7-day free trial. discovery+ is currently only available for US subscribers.
Full episode transcript here: https://curiosity-daily-4e53644e.simplecast.com/episodes/can-swarm-intelligence-solve-humanitys-biggest-problems
CODY GOUGH: I'm curious, how can swarm intelligence change the world?
LOUIS ROSENBERG: For groups of people who are trying to solve problems, whether it's a business team or a group of people trying to plan a fantasy football league, whatever the issue is, from something casual to something very rigorous like sales forecasting, nature tells us that swarm intelligence is a really good way of doing it.
[THEME MUSIC]
CODY GOUGH: Hi, I'm Cody Gough with the problem-solving curiosity.com.
ASHLEY HAMER: And I'm Ashley Hamer. Today, we're going to talk about swarm intelligence and what humans can accomplish by tapping into it.
CODY GOUGH: Every week, we explore what we don't know because curiosity makes you smarter.
ASHLEY HAMER: This is the curiosity podcast.
[THEME MUSIC]
CODY GOUGH: Imagine a global hive mind that could tap into the knowledge, wisdom, insights, and intuitions of millions of people and produce a superintelligence that is way smarter than any individual person.
ASHLEY HAMER: It's not science fiction, it's science fact. Today, our guest is Dr. Louis Rosenberg, founder and CEO of Unanimous AI, an artificial intelligence company.
CODY GOUGH: He's working on technology that amplifies human intelligence by building hive minds modeled after swarms you might find in nature like honeybees.
ASHLEY HAMER: And it's already been able to make some amazing predictions. Stick around to hear what the buzz is all about.
CODY GOUGH: AI is hot these days, artificial intelligence. You focus more on swarm intelligence, but let's level that before we dive into the differences between those two and just to find what in this context is intelligence.
LOUIS ROSENBERG: Sure. That's actually a great question. So the way I like to define intelligence is as a system that will take in noisy information about the world, and then use that information to make decisions or have opinions or solve problems. And ideally, it does it creatively and it does it by learning over time.
And so if that's an intelligence, then an artificial intelligence is a system that we create that can do those things, that can take in information about its world and make decisions, have opinions, solve problems. And again, ideally, do it creatively and, ideally, learn over time.
CODY GOUGH: But then you take that to the next step. So artificial intelligence mimics the human brain or another organism's brain. You've kind of got a neural network that's connecting different actions or different thoughts in a way. And then you've got swarm intelligence which is a step beyond that. Where did this whole idea of swarm intelligence come from? And why do you think it's so valuable in contrast to artificial intelligence?
LOUIS ROSENBERG: Yeah, so when we look at the evolution of intelligence in the natural world, typically organisms, especially social organisms, go through a process where, first, they will evolve their internal intelligence. Basically, their brain. And as we know, a brain is a system of very simple processing units, neurons. When you take very, very large quantities of those neurons and you connect them together in systems, an intelligence emerges. Biology has proven that and in a lot of AI research has shown that we can do that artificially.
If you look at social organisms that evolve that way, we often see that many species then take another step where once they have individual brains that are connections of neurons, they then form a larger system, a system of organisms, a brain of brains. And biologists call this swarm intelligence, and it is when you have a large population of individuals that are working together in a system, and that system is so tightly connected with feedback loops that a higher level emergent intelligence is created.
And ultimately, this is why birds form flocks, and fish form schools, and bees form swarms. They're creating this higher level intelligence, and they're able to be smarter together than they would be on their own. They're able to make better decisions, they're better able to survive by solving problems together as a system, as a brain of brains.
And if you look at evolution, it is a natural step in the pathway of most social species. And so for me, the thing that made it interesting was to say, well, if birds and bees and fish can get smarter together by forming these real-time systems, why can't people do it? And that inspired us to start researching down this path of saying, well, hey, can humans form swarms and create basically a superintelligence, a hive mind where we are smarter together than we would be on our own?
ASHLEY HAMER: You've probably heard of the wisdom of crowds, how estimates by many people can arrive at the right answer. If a bunch of people each guess how many jellybeans are in a jar or how much an ox weighs, you can take an average of those guesses and usually arrive at something very close to the actual number. That's great, but it's not swarm intelligence.
Take fish, for example. Fish swarm together or school by following simple rules. Stay a certain distance from your neighbor and swim the same direction they're swimming. If one fish sees a tasty morsel a ways away that would make it break those rules, it's free to swim over to eat it but then it loses the protection of the school and might have a hard time catching back up.
If it looks tasty enough, though, it might go ahead and swim over. In which case, its neighbors probably will, too. If enough fish follow, the school has made a decision on where to go and every fish gets to chow down on the new food source.
In the case of human swarm intelligence, everyone has an incentive to cooperate with everyone else and decide on one answer. But if one person feels very strongly that their answer is the right one, they can urge the group to go their way and others can follow suit. It's as if you got to guess the ox's weight in real time with every other person at the fair. You can't do that in person, but you can with swarm intelligence.
CODY GOUGH: And that's interesting because some of the examples that you gave from nature, a bee, for example, their brains are very, very small and very limited compared to the human brain. So part of swarm intelligence with bees and certain parts of the animal kingdom seem to be a bit of a necessity because otherwise, their brains wouldn't be capable of doing some of the things that human beings have been able to do.
So what's the onus for looking at human beings who have these complex brains and the ability to have complex thoughts and to look at that and say that swarm intelligence is something that we could really benefit from? Because it's not really a one-to-one comparison, right?
LOUIS ROSENBERG: So you're absolutely right that birds and bees and fish have simpler brains than humans, and bees have particularly simple brains. A honey bee, which is the most studied organism that forms a swarm intelligence, a honeybee has less than a million neurons in its brain. And that actually sounds like a lot, but a human has about 85 billion neurons.
CODY GOUGH: Wow.
LOUIS ROSENBERG: So however smart you are, divide that by 85,000, and that's a honeybee. So honeybee has very, very simple brain. And you're also correct that honeybees evolved the ability to form a swarm intelligence out of necessity, but that's really true of every evolutionary change. Every species creates changes and amplifies their abilities because of some need.
And so for honeybees, they needed to solve complex problems that they couldn't solve as individuals. Now, what's remarkable is if you look at honeybees and how they do this, they can actually solve problems that a human brain could not solve or could not easily solve. And so the level of amplification that they get by forming a swarm intelligence is actually remarkable.
And I can give you an example just so you can appreciate how amazing honeybees are when they form this swarm intelligence. The most significant problem that honeybees have to solve each year is they outgrow their current home and they have to find a new home. And that new home could be a hole in a log or the hole in the side of a building, or for me, they found a crawlspace in my garage. And so they need to find a new home. And for honeybees, again, this might sound like a simple problem but it's actually a life or death decision. It could impact the survival of their colony for generations.
And so to solve this problem, what bees do is they send out scout bees, hundreds of scout bees that will go and they'll search 30 square miles of area and find dozens of candidate sites, potential places they could move into. And this is really data collection. They're out there collecting data about the world, all these different potential home sites, and then they bring that data back to the colony. And that's the easy part.
Now, the hard part is that they need to pick the best possible site to move into out of all these different options. And again, it sounds simple but it turns out that honeybees are very, very discriminating in how they pick their home because their new home, it has to be large enough to store the honey they need for the winter, it has to be ventilated well enough to stay cool in the summer, it has to be insulated well enough to stay warm on cold nights, it has to be protected from the rain but it also has to be near a good source of water. And it needs a hole, an entry that's small enough to block predators but big enough to allow lots of bee traffic to come in and out. And of course, it needs to be well located near good sources of pollen.
And so it turns out this is a complex multivariable problem. And to pick the optimal solution would require multivariable optimization. A single bee with a tiny brain could not possibly solve that problem. In fact, if you were a human looking at all the data, very, very difficult to find the perfect solution. Or if you were a human trying to solve a similar problem, like a CEO trying to find the perfect location for a new factory with all kinds of competing constraints, very, very hard to optimize.
And yet biologists have shown that honeybees pick the optimal solution 80% of the time. And when they don't pick the optimal solution, they almost always pick the next best solution. So they, by forming a swarm intelligence, they converge on the best possible combination of the knowledge and wisdom and insight and intuition of their members of this colony that have gone out and search the world and come back. They combined their insights in an optimal way, and they converge on the best possible solution.
And so again, they can solve problems that would be hard for a human brain. A human brain that's 85,000 times bigger which begs the question of, well, if bees can see this massive amplification of intelligence by thinking together in a system, by forming this real-time system with feedback loops that can converge together on solutions, why can't we humans do it? And if humans do it, it seems like that should allow us to amplify our intelligence.
And that was really the question that got me excited a number of years ago when I started working in this field saying, hey, if bees can have this massive amplification of intelligence being so much smarter together, if humans could do the same thing, could we be smarter together? And all the initial studies that have been done over the last few years show that we humans can. We can be smarter together if we think together in systems modeled after natural swarms.
CODY GOUGH: But it's very different than polling, right? Because you mentioned, this isn't just sending out a survey over email or having people vote on something. It's kind of a real-time communication. So what does it look like when an organization or a group of people is utilizing this swarm application? How is it different?
LOUIS ROSENBERG: Yes, so the first thing is you're absolutely right which is that this is a real time process. The thing about a poll or a survey is that you might collect input from hundreds of people or thousands of people but they're all working in isolation. They're filling out their survey and there is no group. The only group exists in somebody's spreadsheet when their data is statistically combined.
In a swarm, all of the people have to interact and behave and log in at the exact same time. And they're literally forming this system where they're pushing and pulling on each other. And again, it's modeled after how natural swarms do it. And so it's worth mentioning, first, how honeybees do it and then I'll tell you how we turn that into how people do it.
So for honeybees, you have the same problem. How can they combine the knowledge and wisdom of hundreds and hundreds of bees in a system in real-time? And the way they do it is actually amazing. They do it by vibrating their bodies. And biologists call this a waggle dance because to us humans it looks like the bees are dancing. But really, they're generating signals that represent their preference for the different home sites that they've gone out and visited, and different levels of vibration is different levels of preference.
And what happens is by combining these signals in real-time, they engage in this multidirectional tug of war where they're pushing and pulling on the different options until they can finally converge on the one option that they can best agree upon.
ASHLEY HAMER: When bees do this waggle dance, they vibrate their abdomens while walking in a figure eight pattern. It seems hard to communicate anything that way much less where a hive is or how to find a great new flower patch they discovered. Luckily, scientists have been studying this special dance for decades and they're finally starting to crack the code.
For a 2014 study, researchers at the University of Sussex spied on more than 5,000 waggle dances over several years to see how they've changed with the seasons since the dance would indicate different foraging locations depending on the time of year. They figured out that every second the dance lasts equates to a distance of 750 meters or about half a mile. The angle of the dance in relation to the sun tells other bees which direction to fly.
Of course, the swarm doesn't just take one bee's word for it. When they're looking for a new spot to set up a hive, a whole bunch of scout bees check out different sites then come back and do a waggle dance to let the swarm know where to find the best one. Other bees fly off to verify then do their own waggle dance to announce what they think is the best site. This happens again and again until all the bees are eventually doing the same dance and a decision is made. Like Dr. Rosenberg said, this process results in amazingly accurate decision-making. Humans could definitely take a lesson from honeybees.
LOUIS ROSENBERG: When we built systems for humans, we started with this problem that, well, we humans can't waggle dance but we can interact with computer interfaces. And so we built an interface, which allows people to see what looks like a swarm on the screen of all the participants who are all logged in at the same time. And they're each essentially controlling a little magnet that is allowing them to pull on the swarm. And so people are pulling on the swarm in different directions, pushing and pulling, and their behaviors are being watched by AI algorithms that are figuring out their varying levels of conviction.
And what happens is this system will start moving in a direction based on all of their combined input but everybody's reacting in real time. So as the swarm starts moving in a direction, everybody reacts and they might start changing their pull, and so the system is basically this multidirectional tug of war that allows the swarm to converge on the solution that the group can best agree upon by combining not just their opinions but their varying levels of confidence and conviction to find the solution that is really the ideal combination of all of their competing insights.
ASHLEY HAMER: I signed up for UNU, Unanimous AI's swarm platform, and I tried it myself. You start out looking at this hexagon shape with a different possible answer at each of the six points. In the middle is this big round magnifying glass called a puck. It kind of reminded me of the planchet Ouija board, that thing you move around the board to spell words.
When you get near the puck, your cursor turns into a horseshoe magnet, and little green lines show up to let you know that your magnet is pulling on the puck. The closer your magnet is to the puck, the greater its pull. When someone asks a question, a timer starts ticking down from 60 seconds as every user in the room pulls the puck toward their preferred answer.
The idea is that you can change your answer in the moment. If you really think Nikola Tesla is the world's greatest scientist but the group is heading toward Einstein, it's OK to be like, yeah, Einstein was pretty great, too, and at least nobody's saying Edison. I'll go with the crowd on this one.
Once everyone has pushed and pulled their way toward a final answer, the platform calculates what's called a brainpower score. The faster the group converged on the final answer, the higher the score. It basically lets how well you work together as a group.
There's incentive to work together, too, since you earn credits for each question. If the group can't converge on an answer or if you ask a question the group thinks is nonsensical, you lose credits. It's not a bad way to spend an evening. You could try it at go.unu.ai.
CODY GOUGH: It's very impressive what has been produced out of your swarm intelligence applications thus far. I know that last year, I think, was it the swarm intelligence that predicted the Super Bowl score and predicted Time's person of the year?
LOUIS ROSENBERG: Yeah, so we've taken on a lot of challenges from journalists just as a way to demonstrate the value of swarm intelligence since we've predicted things like the Super Bowl, like Time Magazine's person of the year. The prediction that was maybe the most impressive was last year, we were challenged by CBS Interactive to predict the Kentucky Derby. And they didn't just want us to predict the winner of the Kentucky Derby, they said let's see you predict the first four horses in order. And in horse racing, that's called the superfecta. And last year, it went off at 540 to 1 odds.
And so what we did was, and again, at Unanimous AI, we don't know anything about horse racing. We're not experts on horse racing, we'd never done that before. But what we do know is we can amplify the intelligence of human groups. And so what we did was we built a swarm intelligence of horse racing enthusiasts. And we had 20 horse racing enthusiasts connect to our system online. And what happens is they all log in and they're all connected at the exact same time and we're capturing their knowledge and wisdom and insight and intuition, and we're moderating it by AI algorithms that allow them to converge together on the best possible prediction.
And so we built this swarm intelligence of horse racing enthusiasts, and we had them predict the winner, the second place, third place, fourth place. And in this example, we gave the predictions to the reporter at CBS Interactive. And she actually went to the Kentucky Derby and wrote a story and she actually placed a bet and tweeted out her ticket. She bet on the superfecta and those four horses came in perfectly in order.
And so anybody who had placed a bet on those horses that was generated by this artificial swarm intelligence, if they had placed a $20 bet, they would have won $11,000. And fortunately, I placed a $20 bet so I won $11,000. The reporter placed a bet, a bunch of her readers placed bets. And in fact, one of her readers reported winning $50,000 which is an amazing result. And for things that are sports-related, there's always some luck involved.
But clearly, when this swarm intelligence was able to converge on a really good prediction, what's most fascinating, however, is if we go back and we look at those 20 individuals who had formed the swarm, as individuals we look at their own predictions, not a single one of them on their own would have gotten four horses-- all four horses correct. In fact, had they had they just taken a vote, they would have only gotten one horse right.
But by combining their knowledge and wisdom and insights as a real-time swarm, as a system that can converge together, they were perfect. And that's really the power of swarm intelligence, is it finds the best way to combine the information that's in their heads to converge on optimal predictions and decisions and forecasts.
CODY GOUGH: How do I get in on that next betting prediction?
LOUIS ROSENBERG: [LAUGHS] So we as a company at Unanimous AI, we're not sports experts but we do a lot of sports forecasting because it's a great testbed for continuing to improve our system. Because with sports, we can make predictions and get an answer the very next day. And so all through the year on our blog at unanimous.ai, we will put out weekly predictions for-- we're currently predicting English Premier League Soccer, the NBA, the NHL.
We just finished predicting the whole football season where the swarm actually did better than most ESPN experts. And the swarm we had for NFL were just football fans. We took groups of football fans, had them think together as a swarm, and they became essentially an artificial expert that made more accurate predictions together than professional experts at ESPN. And that's really the amazing thing about a swarm, is that it recognizes the fact that people are actually really, really smart. Out there in the population, people have amazing amounts of knowledge and intuition. And if you can harness that, if you can combine that together in an efficient way, you can create these artificial experts that give really, really deep insights.
CODY GOUGH: Is there ability for input of specific ideas as well? Because I'm just thinking, I mean, you mentioned that people are smarter than they might appear in a large group. But there might be certain situations where one person is maybe more educated in a particular decision than others. Is that person able to provide input? And do you think this is the death of expertise? Do you think it's always going to be beneficial for everyone's view to be kind of weighed equally?
LOUIS ROSENBERG: So when people engage in a swarm, their views aren't necessarily weighed equally because they have different levels of confidence, different levels of conviction, different levels of knowledge about an issue. What's really amazing about the process of swarming is that it takes all of those things into account.
And so an interesting example is last year, we were asked to predict the Oscars for Newsweek. And the Oscars, it's hard to predict the Oscars. There's a lot of movies, there's a lot of categories. And so what we did was we built a swarm of 50 movie fans. And those 50 fans came together online and we asked them each of the different categories to predict.
And as individuals, when they predicted the outcome, they were just on their own. Like first, just on a survey, they were 40% accurate predicting on their own. But when they work together as a swarm, so combining all of their different perspectives, they jumped all the way up to 76% accurate. So they almost double the accuracy of the individual members, which is remarkable.
Now, we then asked the participants how many of you have actually seen all of the movies, not a single one of them had seen all the movies. In fact, people most people had seen less than half of the movies. But that's the amazing thing, is that they're filling in the gaps in each other's knowledge. When participants engage in a swarm, they know what they know and they know what they don't know. And so they will be providing input related to the actors or actresses that they have strong feelings about, they will provide less input related to the topics they have less knowledge about or less confidence about.
And so because they're not just putting their opinions down on a survey, but they're interacting in real time and they all have different levels of confidence, the fact that they all have partial knowledge is not a problem. It's actually a benefit because we can then combine their insights to fill in the gaps in each other's knowledge and converge on answers that I said was 76% accurate.
Now, what's also interesting is that there's also statistics on how the average professional movie critic did. So while these regular movie fans, when thinking together a swarm were 76% accurate, the average professional movie critic was only 66% accurate. And so, again, we created this artificial expert that did better than the New York Times and The LA Times and Variety Magazine. And so it does go to your other point, which is that really we can create artificial experts that are more insightful than traditional experts.
Of course you can take that the next level up, which is could also form a swarm of experts and create a super expert. And we've looked at that as well, which is that if we take a population of people who all have very, very high expertise as a swarm, they will still do better than they would have done as experts. So what's fascinating about a swarm is that it almost always performs better than the individuals who make it up. And the smarter a group that you have, the better the group will do when they're working together as a swarm.
CODY GOUGH: I think I understand what you're saying. If I understand correctly, is people kind of self-police almost their own level of expertise. So if I'm asking a question about architecture and one person's an architect, he's perhaps or she is going to say that this really matters, I'm going to put a lot of pull on this magnet. Whereas somebody else who just thinks that maybe a landscape is pretty or not pretty, they just kind of inch along a little bit and say, OK, well, here's what I think but I'm not really going to say, oh I'm a big time expert on this.
LOUIS ROSENBERG: That's exactly right with one extra subtlety, which is if you just ask people to tell you their level of confidence, they're not very good at doing it. They might not even know how confident they are. But in a swarm, we're really just asking them to behave. We're asking them to pull on this system. And so the AI algorithms that are watching their behaviors are actually figuring out their various levels of confidence even if they themselves don't know how confident they are.
And part of the reason is that if you give somebody a survey and you say, how confident are you in this answer and somebody puts down 7 out of 10, well, how do you know that my 7 out of 10 is the same as your 7 out of 10 in confidence? And how do you know that I even really know how confident I am? But in a swarm, it's the system where people are all interacting together.
Again, it's like this multidirectional tug of war. And because we're watching them behave, they're really revealing their true levels of confidence and conviction in a deep way that allows us to converge on answers that really are very accurate representations of the collective wisdom of the whole group.
If you go back 100 years, it was very convenient to take a poll, very convenient to take a survey. You couldn't even have thought about connecting groups of people all around the country in real time. But now, we can do that. When we form a swarm to answer a question, we could have people on every continent and they're all working together as a system enabled by this amazing technology that we humans have made to allow this real time connection. And so this type of human swarm intelligence, it wouldn't even been possible 20 years ago just because the infrastructure wasn't there for real-time high-speed connection basically to everybody. But now, it is.
CODY GOUGH: What benefits can swarm intelligence provide to people in their everyday lives or to society that other forms of intelligence can't? And how is this going to change people's lives in the future?
LOUIS ROSENBERG: We've done a lot of projects for large companies. In fact, we have a service called Swarm Insight where we generate intelligence for large companies where we build a swarm intelligence for them and that swarm can be of their internal people. We've built swarm intelligence of a sales team and allow them to work together to make a sales forecast that's more accurate together than they could have made alone.
We've built a swarm intelligence of customers. We've built a swarm of a company's customers and allowed those customers to predict what the best new features are of a new product. And in many cases, the more expertise that exists in a group, the even better the swarm process works.
We did a project late last year actually for the XPRIZE Foundation where they had their group of experts coming together at a meeting to figure out what the future XPRIZE category should be. And we actually formed a swarm intelligence of these amazing experts, people who had deep insights and deep knowledge about all kinds of different fields. And that's actually the most interesting thing to do when you're forming a swarm intelligence because when you have a group of people who have different perspectives and different insights and different expertise, a swarm can enable them to combine all of those different competing perspectives and converge on the best combination of their collective wisdom.
Now, when we started working in this field we were inspired by this idea that we could make groups of people more intelligent. But we've also been discovering is not only are groups of people more intelligent, but they can also converge on decisions that are just better for the group as a whole. And so there's this idea of, is a swarm intelligence moral?
And in the world of science fiction, the phrase "hive mind" often has negative connotations, as if all the individuals are these mindless automatons that are just working together. And so we wanted to know, well, if really a swarm intelligence was less moral, why would nature evolve so many species to take advantage of it?
And so we've done a variety of research studies where we've been looking at, when a group of people form a swarm, will they converge on answers that are better for the group or worse for the group? And what we found is that swarms actually can reach decisions that are more socially moral than if you took a vote or a poll or a survey.
There's a very famous moral dilemma called the "Tragedy of the Commons." So we did this experiment on the Tragedy of the Commons to see how two groups converge on solutions to very difficult moral problems. And it's kind of a fun problem. So a Tragedy of the Commons, a classic example that professors sometimes give to students in their class is they'll say, OK, here's an extra credit question. For extra credit, tell me if you want 5 points or 15 points for extra credit. Whatever you write down I'm going to give you with one rule, which is if more than a third of the class asks for 15 points, nobody gets anything.
And so the best solution for the group is for everyone to just put down 5 points. If everybody puts down 5 points, everyone will get 5 points. But what happens when you run this type of experiment is almost all the time, more than a third of the people ask for 15 points and everybody fails. People just can't-- they can't help themselves. They feel like, well, if I put down 5 points, somebody else is going to put down 15 points, and so it's hard for people to let go of their own self-interests over what's really the best solution for the group as a whole.
And so we did an experiment where we had swarms of people solve that same problem. And while the groups would fail the Tragedy of the Commons dilemma as a vote, they would actually solve it. They would succeed. They would optimize their answers as a swarm, which is, again, preliminary research but it was very exciting to us that it shows the potential that swarms don't just make a group of people smarter. It actually could make a group of people wiser.
And so we see this potential social benefit of groups thinking together in swarms as being able to solve difficult problems where groups have complex competing interests but they can actually converge together on solutions that are really best for the whole group.
CODY GOUGH: Yeah, can we get this hooked up with our government?
LOUIS ROSENBERG: [LAUGHS] Yeah. Well, it's-- I mean, we see all these challenges with polling. So much of politics these days is driven by polls. And the fact is that polls are polarizing. They actually drive people to extreme positions and they force people to entrench. They actually reinforce our differences. When you take a poll, you're not trying to find where groups agree. You're trying to find where groups disagree, and then you're highlighting those disagreements and causing those disagreements to entrench.
Swarms really do the opposite. Swarms highlight where groups agree. They actually encourage a group to find their commonality and converge on the solutions that really optimize the group's collective satisfaction. And so, again, nature has found this way of taking diverse populations and having them find solutions that are best for the group. And it does it in a way where the groups don't entrench. They don't have gridlock. A swarm of bees does not have gridlock like human political systems.
And the reason for that is that if a swarm of bees entrench the way we humans do when they're trying to pick a new home, that swarm of bees would have died out millions of years ago. So we could definitely learn a lot from how other species have evolved methods of finding good solutions from diverse groups with lots of competing opinions, but also lots of diverse knowledge and wisdom to combine.
CODY GOUGH: I just want to wrap up with our final segment called the Curiosity Challenge. And I'll give you a break for just a second, while I ask you a question that is from an article on curiosity.com the people can read all about. Researchers at the Berkeley Artificial Intelligence Research Lab are working on teaching artificial intelligence to be curious. So they've programmed some surprise-driven exploration into their AI. It involves having it predict an outcome, generate intrinsic rewards, and they are using a video game, an 8-minute video game to teach an AI how to be curious.
And it kind of learns just the way that most people learn how to play a game. It learns where the right place to go is, where the wrong place to go is. So my super random trivia question for you is, what video game are researchers at the Berkeley Artificial Intelligence Research Lab using to teach AI to be curious?
LOUIS ROSENBERG: [LAUGHS] That's a good one. I don't know the answer but I'm going to guess. I'm going to guess Ms. Pac-Man.
CODY GOUGH: That would be a really great one. Ms. Pac-Man is a much more complex game than the original Pac-Man, which is actually part of the reason why it was so popular. That's not on curiosity.com, that's just the nerdy gamer in me coming out.
LOUIS ROSENBERG: [LAUGHS]
CODY GOUGH: But the game that they actually used is the original Super Mario Brothers.
LOUIS ROSENBERG: Oh. [LAUGHS]
CODY GOUGH: Yeah, it tries out all the buttons, quickly learned that while down doesn't do anything, pressing right takes Mario to unpredictable places. And those unpredictable places can yield some big rewards. But as of the time we wrote about this in late 2017, the AI had yet to beat level 1. So it may be a lower bar than Ms. Pac-Man. I wonder how much better or worse it would perform on that game.
LOUIS ROSENBERG: Yep.
CODY GOUGH: I believe you brought a question prepared for me as well.
LOUIS ROSENBERG: Yeah, so I've talked about honeybees a bunch during our conversation so I thought I'd ask you a honeybee trivia question, which is, so honeybees are pretty good flyers. They can fly very well even though they don't look aerodynamic at all. They're kind of big round fuzzy balls flying around. And so the trivia question is, for honeybees to fly, they need to flap their wings faster than, say, birds. How many times per second does a honeybee flap its wings?
CODY GOUGH: Oh, wow. I am going to say 30?
LOUIS ROSENBERG: 230.
CODY GOUGH: Are you serious?
LOUIS ROSENBERG: Yes, it's remarkable.
CODY GOUGH: Wow. I mean, I was going back to, again, the nerd at me coming out and knowing that a lot of games and movies can be at 60 frames per second on a television screen. So that would be four times faster than some of the most high resolution images that you can see in a television screen. Wow.
LOUIS ROSENBERG: Yep. So honeybees are amazing from a lot of perspectives. [LAUGHS]
CODY GOUGH: [LAUGHS] They certainly are. Thank you again, Dr. Louis Rosenberg, founder and CEO of Unanimous AI. Thank you again so much for the swarm intelligence 101.
LOUIS ROSENBERG: Yeah, thank you.
[THEME MUSIC]
ASHLEY HAMER: And now for something completely different, here is your extra credit question. You can email your question to podcast@curiosity.com. This week's question comes from Rhonda Neher who asks, why do books develop a certain smell as they age? Your answer after this.
[THEME MUSIC]
CODY GOUGH: If you've channel-surfed recently and you heard a familiar voice, then believe it or not, that might have been me. Top 30 is a television program that delivers 30 bite-sized stories in 30 fast-paced minutes and gives you all the stories that you need to know for the day. Top 30 does air on television and you can see and hear me talk about some of the latest and greatest stories from the Curiosity editing room on Top 30.
Check your local listings to see where and when you can catch Top 30 or visit top-30.com. You can also follow them on Facebook and Twitter @Top30TV where, again, you'll be seeing some cool segments from curiosity.com. Tune in.
ASHLEY HAMER: I'm back to give you the extra credit answer. Rhoda Neher asks why old books smell the way they do. And it all comes down to the chemistry of paper. Paper is made from trees and trees are obviously plants. Plant cell walls are made of a polymer called lignin and it's the breakdown of lignin that gives old books their characteristic scent.
Ever noticed that they smell a little bit like vanilla? Well, lignin is closely related to vanillin-- the primary component of vanilla extract. People actually use the lignin produced in the papermaking process to produce synthetic vanillin. Of course, the smell is more complex than that. A 2009 study determined that there were hundreds of different compounds in every whiff of old book smell. But the next time you smell an old book, see if you can pick out the vanilla.
CODY GOUGH: That's all for this week. Next time, we are going to welcome mathematician Eugenia Cheng.
ASHLEY HAMER: I'm very excited about that one.
CODY GOUGH: Why are you so excited about that one?
ASHLEY HAMER: Because I saw her talk and she had my boyfriend juggle, and it's just-- she's an amazing person and presenter, and it's really exciting.
CODY GOUGH: Yeah, we're going to talk about mathematical thinking, logical thinking. And I am not a math person, I don't know if you are.
ASHLEY HAMER: I'm a math fangirl, I think.
CODY GOUGH: OK, well, I'm not a math fangirl, but I still found her super fascinating and it's going to be a really exciting podcast. So stay tuned next time on the Curiosity podcast. That's all for this week. I'm Cody Gough.
ASHLEY HAMER: I'm Ashley Hamer.
CODY GOUGH: Thanks for listening.
[THEME MUSIC]
SPEAKER: On the Westwood One Podcast Network.