How AI Can Make Groups Smarter Than Any Expert | Louis Rosenberg

Vit Lyoshin (00:00.93)
Welcome Luis. Thank you for your time today.

Louis Rosaenberg, PhD (00:33.283)
Thanks for having me.

Vit Lyoshin (00:35.628)
Yeah, of course. so you spent many years working with technology, with augmented reality now with AI, and you're working around this area of collective intelligence. So what kind of convince you that now AI technology is kind of like the best way of connecting people?

Louis Rosaenberg, PhD (00:58.721)
Yeah, so so the field of collective intelligence is really about you know how can can large groups combine their knowledge and wisdom and insight and amplify their intelligence above the average person in that group. And it it actually the the field itself of collective intelligence goes back over a hundred years to nineteen seven to Sir Francis Galton, when in England he he

Interviewed a whole bunch of farmers and asked them all to estimate the weight of an ox. And he discovered for the first time that when he asked, I think it was 800 farmers to each estimate the weight of an ox, if he took the statistical mean, it was actually very accurate and significantly more accurate than the average person in the group, and that the field of collective intelligence was born. And about 12 years ago is really when when I got very interested in collective intelligence, and and I realized that.

The technology really hadn't changed for a for a century. when when people were were basically looking at how do we make groups of people smarter, we were doing the same thing that Sir Francis Galton was doing in 1907, doing polls and surveys and and aggregating. And and it works. You get group groups get a little bit smarter. And but I I really asked the question of, well,

It you know, is there a significantly better way to connect groups of people together? And and if so, why aren't we doing it? And like a lot of researchers, what I did was look look to Mother Nature to say, well, how does nature do this? nature doesn't take votes or polls or surveys. nature actually forms systems. And and it turns out that that evolution has produced

Really interesting solutions to this problem multiple times. Flocks of birds, schools of fish, swarms of bees, they all have developed methods to make significantly smarter decisions that are not just smarter than the average individual, but smarter than every individual in the group. And so biologists usually refer to those groups as a superintelligence.

Louis Rosaenberg, PhD (03:23.239)
that that I asked and and the reason I founded my company Unanimous AI a decade ago was to say well if birds and bees and fish can can connect themselves into these really really efficient systems and form a collective superintelligence, why can't people do it? And and it turns out that we can. And so and that's really been the the focus of my research for the last decade of how can we connect large network groups of people together of

you know, of any size, you know, 20 people, 200 people, 200,000 people, and enable them to to work so effectively together that they can outperform every member of the group and ultimately form a collective a collective superintelligence. And and it turns out to do that we we need to use not just the principles of swarm intelligence from nature

Vit Lyoshin (04:08.013)
Mm-hmm.

Louis Rosaenberg, PhD (04:21.155)
but also the power of artificial intelligence to help us to to help us connect groups, because humans didn't evolve the same way as as b birds and bees and fish, but AI can can correct for that and allow us to to form these systems where human groups get really, really smart.

Vit Lyoshin (04:41.91)
Okay, I see. And so this ca collection collective superintelligence system, I guess, does it find actually a better answer or is it looking through like people that people agree on the same answer basically?

Louis Rosaenberg, PhD (05:01.369)
So when we when we connect groups of people together in a in an optimal way, they will find they will find answers that are that that are better than any individual would have found on their own. and and sometimes answers that are significantly outperform what most of the people would have got. So it you know, had they just taken a vote, had they all voted on this is, you know.

Vit Lyoshin (05:17.538)
Mm.

Louis Rosaenberg, PhD (05:31.085)
This is what we all think is the best answer in a traditional way. When they're connected into a you know a super intelligent system, they will converge on an answer that s that will s in as statistically significant rates outperform what you know a traditional, you know, poll, survey, vote type of of scenario is. And and the the

Vit Lyoshin (05:52.95)
I see.

Louis Rosaenberg, PhD (05:59.012)
The way that it works, and the reason that it's different than a vote or a poll or a survey is that we turn the human group into a system rather than treat them as a source of data to aggregate. And so again, the traditional way is to say, okay, I want to I want to harness the input from 200 people. Let me give them all a survey, they'll all give me their individual perspectives, I can aggregate that data, and it will get a little bit smarter. But

I but really what I did was I took a whole bunch of people in isolation. They didn't really exist as a system, they just existed as data in a spreadsheet somewhere, and then statistically they got better. Now the the magic that nature has is that you they form real-time systems. And so in nature, whether it's a flock of birds or a school of fish or a swarm of bees, biologists refer to this as swarm intelligence. And so swarm intelligence is a system.

Vit Lyoshin (06:40.003)
Mm-hmm.

Louis Rosaenberg, PhD (06:58.711)
Where the the group doesn't take a vote, it basically forms a dynamic system where the groups can push and pull on each other. They're they're basically negotiating in real time with different levels of sentiment, and they're f they discovering the solution that optimizes their collective confidence or their collective conviction. And when they converge that way, they can find solutions that maybe was a minority opinion in the group, but as the deliberation happened.

the group converged on on this other answer. And polls and surveys and votes never do that. They just a poll and a survey and a vote will tell you the most popular answer, but the group can't deliberate. They can't, but when a group can deliberate, they they can basically push and pull and and and find this better answer. Now in a fish school they you know fish

Vit Lyoshin (07:32.706)
Mm-hmm.

Louis Rosaenberg, PhD (07:57.36)
Imagine a fish school has, you know, 2,000, 3,000 members, right? And and imagine that they have a problem to solve. And the thing about a fish school, it forms a swarm intelligence. There's no leader in a fish school, right? there's nobody in charge. They make all of the their important life decisions as a swarm intelligence. They they navigate the ocean, they evade predators, they find food.

Vit Lyoshin (08:02.765)
Mm-hmm.

Louis Rosaenberg, PhD (08:25.313)
And they're doing this as this system, and they're making decisions smarter than any individual could. So imagine a fish school has you know thousands of members, and imagine that there are predators approaching from multiple directions, right? So now you have predators, and now the thing about that is that some you know fish that are on the right side of this school will see a predator coming, they'll have that piece of information. Other fish won't even know about that predator, they can't see it from their location.

Vit Lyoshin (08:38.638)
Mm-hmm.

Louis Rosaenberg, PhD (08:53.471)
fish on the left side might see a different predator coming. You know, how do they aggregate this information through this school and almost instantly evade both predators? with and the way it works is that each little group of fish well, so each fish has a special organ on their side of their body. It's called a lateral line. And this organ detects.

Pressure and vibration changes in the water around them. And so they basically can detect the intentions of their neighbor the neighboring fish. So if each fish is monitoring the intentions of their neighbors, and when something happens, they all have an you know, they all have a sense of well, what should we do based on our little local amount of information? And they push and pull on each other through this, through their detecting the pressure and vibrations in the water.

But the magic happens is that each little group of fish overlaps the next group of fish, overlaps the next group of fish, and so information propagates through the school really quickly. And so in that scenario, you had a group of fish on the right saw a predator, they had a little deliberation, information propagated. A group on the left saw a different predator, they had a little deliberation, information propagated, and instantly it it all the information spread through the network, and the fish made this decision and evaded.

Vit Lyoshin (10:02.956)
Mm-hmm.

Louis Rosaenberg, PhD (10:20.707)
the the predator. And so the the the the question is, well, how can we do that in humans, right? So now humans, we deliberate conversationally, right? The way that we optimally br if you want to bring a group of people together and have them optimally combine their knowledge and wisdom and insight and intuition, you can bring them into a room and they can have a conversation and they'll argue about things and they'll

Vit Lyoshin (10:22.99)
Well.

Vit Lyoshin (10:27.715)
Yeah.

Louis Rosaenberg, PhD (10:45.583)
You know, somebody will bring up a point and somebody else will bring up a counterpoint and they'll they'll share evidence and they'll make a good decision. And and research shows that you can do that pretty well in a group of about five or six people, but you can't do that in a group of a hundred people or a thousand people. And so this means, and that's the reason why collective superintelligence was never possible, because we humans can deliberate really, really well in small groups.

And so the so the the first question that that you know the research team at Unanimous AI looked at was saying, well, the magic of a fish school, which also deliberates in small little groups, is that every little group overlaps, right? So and so information can propagate through. And so we said, well, maybe we can take a human group and we can break them up in a way where where instead of having you know a hundred people in one big group, let's take a hundred people, break them up.

into twenty little groups of five, have all the groups overlap, and information can propagate through in in every direction and they can and they can have one big conversation. Well, it turns out that people can't do that. People can't because people aren't fish, right? And and and by that I mean you know a fish can be in multiple conversations at once because its conversation is really simple. It's just detecting vibrations. A human

Vit Lyoshin (11:45.346)
Mm-hmm.

Vit Lyoshin (11:58.356)
Mm-hmm.

Louis Rosaenberg, PhD (12:11.213)
Can't be in multiple conversations at once. It can't be in these two overlapping conversations. And there's actually a name for it. researchers call it the cocktail party problem. Because if you're at a cocktail party and you're talking to a small group of people and there's another group near you talking about something else, if your attention shifts to that other group, you lose your focus on the group that you're in. Like you just can't do it. And so again, this is why

Vit Lyoshin (12:36.478)
Right.

Louis Rosaenberg, PhD (12:41.111)
Humans have never been able to form a collective superintelligence where we can have a conversation with a hundred people or a thousand people. And so if you can't allow groups to overlap, how do you make group how do make information propagate? And so the way that we solved it was with AI, and specifically was was with conversational AI agents. We said, and and the basic idea is that a person can't be in two conversations at once.

Vit Lyoshin (13:03.992)
Mm-hmm.

Louis Rosaenberg, PhD (13:09.465)
But an AI agent can be in two conversations at once. And so what we did is we said, okay, well, what if we take a hundred people, we break them up into 20 little groups of five, and and five is a good number because it's like the optimal size for a thoughtful conversation. And then we add to those 20 little groups of five an AI agent. It goes into every one of those groups. And that that AI agent, its job is to participate in that little conversation of five people.

Observe that local conversation, like watch watch exactly how that little conversation is happening, what are the different opinions, what are the strengths of each person's opinion? It's it's really understanding that conversation, it's extracting insights from that conversation, and it's sharing those insights with the agents in other little rooms, little groups, right? And then when an agent in one of the other little conversations gets that insight.

It participates in the conversation. It expresses those insights into the conversation. And so if I'm in a little group of five people, there's an AI agent with us, we're having a conversation together, and the AI agent, every once in a while, will bring up an insight. It didn't make up that insight. It didn't grab it off the internet. It's not, you know, it didn't generate it itself. It got that insight from another human group.

And it but it expresses it into our conversation. And that and that's happening simultaneously in every one of these little groups. And that weaves all of these little groups into one big conversation. That could be a hundred people, it could be five hundred people, it could be five thousand people in one big conversation. And when we do this, when we connect groups this way, this group can like just like a giant fish swarm, fish school, this group of you know a thousand people.

Can brainstorm together, and you know, they're building off of each other's ideas. Ideas are passing all the way all around through this thousand-person conversation. They can forecast together, they can estimate together, they can prioritize together, they can, they can discuss and debate issues. And and so they again, it it's taking this concept of swarm intelligence from nature. It's it's modeling this large human group like birds and bees and fish.

Louis Rosaenberg, PhD (15:32.836)
But it's using AI agents to do the thing that we didn't evolve to do. Fish evolved to do it, birds evolved to do it, we didn't evolve to do it, but AI agents get us l allow us to jump over a hundred million years of evolution, which is how long it took for fish to learn to do that. And and it turns out when we connect groups this way, they become significantly smarter. They they and and it is

Vit Lyoshin (15:48.429)
Yeah.

Louis Rosaenberg, PhD (16:00.815)
Much, much, much, much smarter than if you just said, okay, I've got a thousand people, let me just give them a survey and and aggregate the data. The group will get, you know, 5% smarter, 10% smarter. When we bring a group together as a collective swarm intelligence, they be they essentially can become a superintelligence where they become smarter, not just that not just smarter than the average person in the group.

Vit Lyoshin (16:08.439)
Yeah.

Louis Rosaenberg, PhD (16:28.003)
they become they can outperform every person in the group. And and that's you know that you know that's the reason why you know collective superintelligence is so interesting and so valuable. And it's also, you know, the reason why it's never been done before is that it really required two things in combination. One, something you know, inspired by nature, which is you know

Vit Lyoshin (16:31.606)
Mm.

Louis Rosaenberg, PhD (16:57.389)
don't aggregate data, connect connect groups into systems. And two, you know, humans can't form these large systems conversationally unless we had the power of of you know conversational AI agents. And once we have that power, now we can we could really you know allow groups to communicate in a new way and and groups get smarter and they get faster. And the the most interesting thing

Vit Lyoshin (17:13.902)
Yeah.

Louis Rosaenberg, PhD (17:26.763)
is that people really like doing it. And and that's one of the things where we just didn't know. Like we like we started from the premise of, okay, nature says that this should work and groups should get smarter. But we also, you know, it could have been the case that people, you know, people thought it was, you know, they didn't want to do it, right? But it turns out when it it turns out when a group is connected this way and we interview them after, they say they feel more heard than they've ever felt before in a large group.

Vit Lyoshin (17:45.582)
Sure.

Louis Rosaenberg, PhD (17:57.028)
Because normally if you're like in a big company with you know thousands of people and they want to get everyone's input, they'll give everybody a survey, right? And then people will feel like, yeah, nobody cared what I thought, you know. But instead, if you connect them into this large conversational swarm and they know they're, you know, what they're saying is getting spread around the network and and they can they can see how the whole group is converging together on a solution, they feel like they were really part of, they were part of a group.

Vit Lyoshin (18:05.632)
Yeah.

Louis Rosaenberg, PhD (18:26.507)
intelligence. They weren't just a piece of data that went into somebody's spreadsheet somewhere. And so and so it turns out you know groups get smarter, groups get faster, and then groups actually feel more collaborative and and individuals feel more heard.

Vit Lyoshin (18:42.23)
Yeah. Yeah, that's interesting how this technology actually enabled us to have really like the next sense, if you will, right? And it's also part of this kind of negotiation because you provide your inputs, your knowledge, your expertise, and somebody's arguing and doing counter arguments with you and then you come up with a solution. That's what

Typically happens in the smaller groups in a meeting, but scaling it to a large number of people never was possible. That that's very like a good practical use case for a for AI because lately I've been hearing all the you know weird use cases rather than something interesting like this. So I I yeah.

Louis Rosaenberg, PhD (19:26.159)
Well well the I mean so you so you're you you you're exactly right that you know groups of people are really good at you know having you know raising points and counterpoints and evidence and and in fact the one thing the one place that we're able to do better than fish schools and and bee swarms and bird flocks is that in in a in a fish school

Vit Lyoshin (19:41.634)
Mm-hmm.

Louis Rosaenberg, PhD (19:54.852)
The way that information propagates across the school is by proximity, right? So you know, a fish see a predator, they they have a signal, that signal propagates through. Now, when we're connecting a group of people over the internet, and we're having AI moderate the system, we don't have to allow information to pass through the system based on proximity. We could have information pass.

From any little subgroup to any other little subgroup. And so what the way information gets shared around this network is the AI is actually looking and saying, okay, this little subgroup is talking about this solution to the problem. There's no point in bringing in information from another subgroup that has the same solution. It's much more efficient to bring in information from another subgroup that either has a counterpoint that

That argues against their solution or has a very different solution. And so we can we can connect groups in a way where information is passing to maximally challenge every little deliberative group with ideas that that are different than their own. And so, in a sense, the the way we describe it is that instead of creating a swarm, the way a biological swarm works, we're basically creating a hyperswarm where

Vit Lyoshin (20:57.347)
Yeah.

Louis Rosaenberg, PhD (21:22.315)
Information can pass between any two nodes in the network at any moment based on what's the most efficient connection. And that's one of the reasons why when a group deliberates as a swarm, they get much faster because they first of all they're all working in parallel. Like if I had one big group of 20 people, it's very slow. They're taking turns, each person's talking one at a time.

If I take a group and I break it up into these small groups, they can they can they're all talking in parallel. They all have diff they're all going down different threads. Information is passing around this network to optimally challenge every group. So they're all being challenged at the same time. And so they converge on solutions very, very quickly. in in so they that's you know part of why they get they get faster, and also why they can find they can find solutions that

Vit Lyoshin (22:03.67)
Mm-hmm.

Louis Rosaenberg, PhD (22:15.815)
maybe are a minority opinion at the start. And and that's like the biggest downside of a traditional collective intelligence system is that if the conventional wisdom in a group is wrong, you'll always get stuck on that. Like if most people just if most people have a misperception or have like bad, you know, bad data, it will, you know, you're gonna take a vote and it's just always gonna land on

Vit Lyoshin (22:33.869)
Right.

Louis Rosaenberg, PhD (22:44.675)
the wrong answer. But when it's deliberative and the groups are connected this way, then information propagates not based on popularity, but based on its its deliberative merit, based on how you know its arguments, how strongly somebody can argue for why one idea is better than another. And so we did a a really interesting experiment with some researchers at Carnegie Mellon to test this, where we gave groups of people

IQ tests. And so these were groups of about 35 people, and we had lots of them. And we said, okay, we're going to bring groups of 35 people together, we're going to give them an IQ test and have them solve the IQ test together. Either by traditional collective intelligence, which is just take a vote, or a know a collective superintelligence based on a swarm structure. And and what we found was that when we we brought together all these groups,

The the first of all, across all the participants, the average IQ was 100. Because that's and that's how IQ IQ tests are designed. If you the if you re if you take a random sampling of a population, it should average out to an IQ of 100. So that so that worked. Then we said, okay, if we take groups of these random samples of a third groups of 35 people and have them just take a vote,

They will amplify their IQ. They they will amplify their IQ up to about an IQ of about 112, which is, I think, a close to like the 80th percentile. So they get smarter. So traditional collective intelligence works. Again, that goes back a century. but they can never get they can never get s smarter than that because again, like it's it's a popularity contest. And so, you know, okay, maybe.

Vit Lyoshin (24:27.939)
Mm-hmm.

Louis Rosaenberg, PhD (24:43.183)
You know, an IQ test is maybe like six or s six choices for every question. And so you know, maybe you know a a slight plurality would see the right answer and and it gets this group up to a IQ 112, but that was kind of the limit. When we took this you know same groups of 35 people or random random selected groups of 35 people and we hyperconnected them in this swarm structure.

It boosted their IQ all the way up to 128, which was the ninety-seventh percentile. and the most interesting thing was that if we looked at every individual group of 35 people, the swarm always outperformed every member of the group. And so whereas in the traditional collective intelligence, you know, the group would outperform the median person, but almost never outperform every person.

But in when we c when we connect them in this in a swarm structure, it outperformed every person. And so you know, that was really interesting test in of really showing and and it it gets at this point of the only way that a swarm structure could really enable this in an IQ test, because when you get to you know an IQ of like 128, which is gifted level IQ, like almost genius level, but gifted level.

Vit Lyoshin (25:42.765)
Yeah.

Louis Rosaenberg, PhD (26:09.889)
It means that if you take a random group of 35 people, you might only have one or two people in that group who can s who can can consistently see the answer. And so how does that one or two people sway you know the other 33 people? If it was a vote, it would never happen, right? But when it's a deliberative system and somebody has a you know has a good argument, that argument can move around the network and and the the

It will, it will get the group to basically converge on the right answer. And so we've done, you know, so IQ test is interesting, but we've done a lot of different experiments. We did a big study with Stanford University with groups of medical doctors, radiologists. And and there we were looking at small groups of just five or six doctors, and we asked had them diagnose chest x-rays. And they either diagnose a chest x ray.

just by taking a vote, or they diagnosed the chest x-ray as you know as a swarm intelligence. and when they did it when they did it as a swarm, they reduced their diagnostic errors by 33% compared to the the the ability of the small group or or an individual member. And and so we see it you know in we've we've seen it in you know medical diagnosis, we've seen it in financial forecasting.

Vit Lyoshin (27:14.998)
Mm-hmm.

Louis Rosaenberg, PhD (27:37.55)
we've seen it in in just general problem solving you know when when groups of people are connected this way they they can you know can per can perform at these upper limits and the other thing that that's interesting that we see is that as groups get bigger they also get smarter as well and so you know it works at groups of five we works really well groups of 35 but we can do you know 100 people 200 people

Vit Lyoshin (27:52.322)
Mm-hmm.

Louis Rosaenberg, PhD (28:07.359)
you know, 500 people. we've we've never gone larger than that yet, but there's nothing, there's no technical barrier to doing a group of a thousand people or ten thousand people or a million people because it's a very efficient way to connect people and and a really interesting way for them to solve problems or just brainstorm together and and and build off the ideas of each other or and so we you know we look at

enabling groups to forecast more accurately, brainstorm more creatively, you know, do diagnosis or problem solving. And yeah, it's it is a it's a really valuable general purpose way of making people smarter. And the reason that you know the reason I've been excited about it, you know, really for the last decade is that it it is an alternative to

A purely artificial superintelligence, right? So you know purely artificial superintelligence is really smart, but it's not human. And and it can pretend to be human, but it doesn't mean that it inherently has human values or or sensibilities or interests. But when we build a collective superintelligence of human, which uses AI to connect the people together, we're amplifying intelligence, but we are definitely keeping

Vit Lyoshin (29:09.197)
Mm-hmm.

Louis Rosaenberg, PhD (29:34.039)
you know, human values, morals, sensibilities, interests inherent in the system. And and so it becomes a you know it becomes a way to amplify intelligence without without leaving the human part out of the system.

Vit Lyoshin (29:51.895)
Yeah. I I guess the interesting part you just said is that and I talked with a bunch of scientists also on the topic of AI and how it works. And the conclusion is basically it's an average intelligence, right? And like you said in the beginning, it's like a collective intelligence because it doesn't argue with each other and it doesn't contradict each other, it just takes all the facts, compares them like a survey, and picks the mean or or average.

And says this is the average, the best answer out of all this information I learned from the internet. But if if they tweak it somehow so that it not only does that, but kind of argues with each other and compares all the facts and figures out which one is best, which one is worst, and so forth, it will be the next level of intelligence. Then it will be truly artificial intelligence, right?

Just like with swarm intelligence, when each person provides their best input and it gets selected to move to the next argumental level, if you will. And at the end you get better and better and better solution or outcome or decision or whatever it is that the group is trying to decide. That's very interesting.

Louis Rosaenberg, PhD (31:07.125)
And the it's a great kind of great segue to the the the technologies that we're working on now, which is you know, AI systems have gotten smarter and smarter and smarter and and they're very impressive. humans are also really s really smart and we don't we don't give ourselves enough credit in terms of our our knowledge and wisdom and and and intuition and and experience.

Vit Lyoshin (31:20.397)
Yeah.

Vit Lyoshin (31:27.041)
Yeah.

Louis Rosaenberg, PhD (31:37.036)
And so we've been looking at well, what happens when we create a you know a swarm intelligence of humans and AI agents? so in our you know in the systems that we were building a few years ago, we were using the AI just to connect the humans, to allow us, allow them to form a swarm intelligence. Now we've been saying, okay, we're gonna form this swarm intelligence, but we're also gonna put AI agents into the these little groups.

Vit Lyoshin (31:45.43)
Mm.

Vit Lyoshin (31:57.101)
Mm-hmm.

Louis Rosaenberg, PhD (32:06.499)
that can participate with the humans. And and so they become collaborators. And and so we call them coagents because they're they're basically collaboration agents. And and and so we just did a study we just finished it a few weeks ago where we said let's get groups of people together and let's have them deliberate and predict something that we can score.

very you know very easily. And so we had them predict the outcome of baseball games. And so we had them week after week predict the outcome of baseball games and we had them do it in small groups of just four people. And then and that was you know half of the trials were groups of four people deliberating together and conversationally, they're just arguing about the baseball game. And then in the other half of the trials we had four people arguing about the baseball games and a an AI agent.

that was in there also participating with them. And and that AI agent, it didn't it it didn't give an opinion about which team was going to win. What it did was it listened to the human conversation and it would identify information that would be useful to the humans. It would say, okay, these guys are talking about these two teams and it would realize that they need some information about the picture of that team. And so what it would it would just

express that in and it would and you nobody would have to ask the AI agent a question. It would just it was just participating in the group and watching the conversation and and it would say you know the you know the that pitcher won its last five out of six games and it would bring it into the conversation. And we found was that that when then we created this structure and we gave and put an AI agent into every one of these deliberating human groups the

Vit Lyoshin (33:35.715)
Mm.

Louis Rosaenberg, PhD (33:58.062)
The forecasts got much better. The groups got between eighteen and twenty-five percent more accurate in predicting the outcome of of baseball games than if they were just deliberating amongst themselves without the value of an AI agent. And again, the AI wasn't it wasn't offering an opinion, it wasn't making a forecast, it was just providing information, which is what, you which is the the thing that the AIs are you know unquestionably good at.

Vit Lyoshin (34:24.066)
Mm-hmm.

Louis Rosaenberg, PhD (34:27.715)
Right? They have access to all the world's information. These agents could go out and search for information in real time during the conversation. And so what really what the AI agents were doing was was saying, okay, there's this swarm of humans that are that are forecasting. We're going to make sure that the people are, you know, are very well informed as they make their decision. And yeah, and that was you know, and so that's what we would call you know a a hybrid.

Collective superintelligence, where it's you know it's a collective of humans and AI agents deliberating together. And it's really interesting, and and we're we're doing lots of experiments in that area, but that to us is one of the most exciting directions because there's no question that that AI has capabilities that humans don't have. I believe that humans also have capabilities that AI doesn't have.

Vit Lyoshin (35:22.488)
Mm-hmm.

Louis Rosaenberg, PhD (35:26.291)
And and when they can deliberate together at potentially any scale, then we could, you know, we could really have a a collective superintelligence that becomes much smarter than any individual human, but it's still is it's still human, right? It's still rooted at its core in human values, human morals, human sensibilities, human interests, and and what that's really important, especially in large

Vit Lyoshin (35:42.934)
Right.

Louis Rosaenberg, PhD (35:54.872)
you know, large organizations making important decisions. If it's making an important decision, you know, it's my belief that we want to keep humans in the loop of that decision. And not just have a human that is just, you know, approving or rejecting something the AI spits out, but have the humans part of the deliberation. Have the humans and and AI deliberate, each offering, you know, you know, offering information from their strength.

And and and so I you know I really do believe that the combination of humans and AI deliberating together at scale is you know is ultimately gonna be smarter than humans or AI by itself. And and we've seen hints of that in in some research studies we've done, but it's you know it's still an open open research topic to really prove that humans and AI

Vit Lyoshin (36:39.565)
Mm.

Louis Rosaenberg, PhD (36:51.929)
deliberating together at scale can you know can significantly outperform both both populations by themselves.

Vit Lyoshin (36:57.174)
Yeah.

Yeah, that's an actually interesting concept. It's like having a coach basically, on and helping you look into areas that you may not know or may not like think about. It's like you know, when they say in in like an organization, when you do some brainstorming and planning and investigation, it's good to know what you don't know, right? You don't know what you don't know. AI may be not

like completely on hundred percent expert in the area, but at least it can question and maybe sometimes even ask a dumb question and say, Hey group, did you think about this one or that one? And people start thinking about it and and they may, you know, the light bulb goes on.

And you come up with a good idea. It's actually a great idea. I think AI is like augmented intelligence. There is another term that I recently heard and I like it a lot. So average augmented intelligence for a human. I kind of like like this idea lately. And I was actually about to ask you about are people have to be an expert on the group? Seems like no. Seems like it's anybody can be there because we have all this collaboration going on.

Louis Rosaenberg, PhD (38:02.285)
Yeah.

Louis Rosaenberg, PhD (38:13.559)
Yeah, so it's it's a really interesting question and w there's a few different things that we see. One, when we bring a group together in in this, you know, c swarm intelligence to form a you know collective superintelligence, but what we'll find is if we take in a a group of people who are just novices or or you know average people, they will they will get smarter than

Vit Lyoshin (38:37.389)
Mm-hmm.

Louis Rosaenberg, PhD (38:42.253)
The vast majority of the individuals in in the group. So they'll outperform almost everybody in the group. They'll they'll you know go towards collective superintelligence. So you know, when we do these sports studies, like I mentioned, baseball, those weren't experts, those were sports fans, right? So we brought together sports fans and they produced a super fan, right? They did better than now. We could also bring together a group of sports experts and they'll form a super expert. And what we find is that

Vit Lyoshin (39:01.709)
Hmm.

Vit Lyoshin (39:07.693)
Yeah.

Louis Rosaenberg, PhD (39:12.225)
And we've done a lot of experiments in with you know we've we've had financial we've had financial analysts come together and predict, you know, market trends and and outperform you know a professional analyst. We've had groups of sports fans come together and and outperform the average fan. We sometimes can have just groups of just completely novice people come together and and outperform.

Or we've had we've had medical professionals, like we like when we brought together groups of radiologists, you know, these radiologists went to school, you know, medical school for four years, then then they became a specialist for another four years, and they then practiced as radiologists. So these people had you know 20 years of experience as radiologists. We you know we thought like we can't really make them any smarter, but we could. When we connect them together, they actually get smarter.

Vit Lyoshin (39:59.247)
Yeah.

Vit Lyoshin (40:05.58)
Yeah.

Louis Rosaenberg, PhD (40:08.673)
And part of it is that when you bring together a group of experts, they they know what they don't know. And you brought that up. Like it's it's important to know what you know, and it's important to know what you don't know. And so when you bring together a group of experts, they know what they don't know. And so they're more likely to listen to other insights spreading around the network because they realize, like, somebody saw something I didn't see. When you know, sometimes when you get down to the novice groups,

Vit Lyoshin (40:22.264)
Mm-hmm.

Louis Rosaenberg, PhD (40:38.905)
People are overconfident because they just don't even know what they don't know. And so it's it it in some sense it works better as groups get smarter or or not even smarter, as groups get more professional or more experienced, then they they they can amplify their collective intelligence more than groups of people where some people might not even realize how hard the question actually is. so that's like so that's a really interesting.

Vit Lyoshin (40:43.17)
Yeah.

Louis Rosaenberg, PhD (41:08.579)
factor. the other thing is that when we bring to when we bring in an AI agent into the mix, you know, the the AI agent is it's gonna listen to their conversation and it's gonna go out and find information that that fills in knowledge gaps in their conversation. And so if they're a group of novices, the AI agent might say, you know, you really should consider the you should consider the pitcher that's pitching tonight.

when you're when you're considering who's gonna win. and because they because if that group of people doesn't realize don't realize that's the most important factor, the AI agent will will will bring that up. If you're bringing together a group of pretty experienced sports forecasters, they know that. But what you'll see is the AI agent will come out and say something that will always shock me. You could say, you know, the, you know, it's gonna be very humid in that ballpark tonight, and humidity has this effect on, you know, reduces the number of home runs.

Vit Lyoshin (41:54.434)
Mm-hmm.

Louis Rosaenberg, PhD (42:08.001)
Or and like things that are just that you know, that I wouldn't have known. but and so there's you know, there are things that the AI is good at, and the the the AI is good at having massive amounts of information. The thing that humans are good at is almost the opposite, which is it and and we've seen we've seen studies that show this, that you know, the the AI can make a really good decision when it has optimal information, and a human makes remarkably good decisions when it doesn't have enough information.

Vit Lyoshin (42:09.247)
wow. Yeah.

Louis Rosaenberg, PhD (42:37.719)
Right. Like hum like humans are really good at having at making good decisions in in when there's big gaps in the knowledge. And and we call that you know intuition, or or sometimes we call it wisdom. but it's you know it is a genuine skill, and and the AI doesn't have that at the level that a human does. AI, you know, it is really good at having depth of knowledge.

Vit Lyoshin (42:51.606)
Uh-huh.

Louis Rosaenberg, PhD (43:07.585)
And finding the most statistically likely solution. And that's just a different skill. And and and when you combine the two together, the human skill and the AI skill, you really do see that they complement each other.

Vit Lyoshin (43:15.435)
Yeah, interesting.

Vit Lyoshin (43:23.498)
Mm-hmm. Yeah, yeah. I haven't thought about this like this, but I guess yes, people learn. Once you learn something that like fire burns your hand, you're not gonna try it anymore. You just know that that's it. That's the decision. Like, don't touch it, right? I also wanted to ask you about how practical it is to apply to like a global scale, like global issues or countrywide issues. we have

sometimes you know pretty controversial things going on in societies and how an application of this system may have helped it since you mentioned that there is no like limitation on the number of participants or maybe we can sample size it to a few thousand people. can you say a few things like what you're thinking about that?

Louis Rosaenberg, PhD (44:09.891)
Yeah, yeah. So it's that's that's an area that we're very interested in, which is bringing large groups together, especially groups of people across countries and across regions, because they have different perspectives and and usually the most interesting deliberations are when people have very different very different views. And so and we do this often. The the one thing that's exciting is we just so so our our core product is called Thingscape, which is

Vit Lyoshin (44:21.133)
Mm-hmm.

Louis Rosaenberg, PhD (44:37.913)
Where we bring groups of people together to deliberate, and we just enabled a new a new technology inside of Thinkscape that we call hyperlingal, where we allow groups to deliberate across many languages at once. And so we can now support 10 different languages. So we can bring 250 people together across 10 different languages, and they can have us have a deliberation. And it's it

Vit Lyoshin (44:52.099)
Wow.

Louis Rosaenberg, PhD (45:02.743)
It feels like science fiction and we just this is new. We just really started doing these big studies. And so one of the we just just last week we did a study where so in the US one of the big issues that people are talking about is data centers. And there's like all of this anger about data centers in the US, where you know over 80% of the population is against data centers. And and so we were asked by a research partner to say, well,

Vit Lyoshin (45:20.262)
yeah.

Louis Rosaenberg, PhD (45:31.3)
How does that compare to the the rest of the world? How does the rest of the world think about data centers? And we we said, we can do that. And so we brought together over it was like over 200 people from 27 different countries and 10 different languages, and we had them discuss data centers to you know in in like imagine doing a focus group, but instead of having eight people in a focus group, you had 208 people in a focus group.

across 27 countries, 10 different languages, and they converge on solutions. And so we did that and and it was you know the languages were you know really diverse. It was you know there was English, German, French, Spanish, Portuguese, Hindi, Mandarin, Russian, Japanese, Korean, so really, really diverse languages all deliberating together. And

Vit Lyoshin (46:25.336)
Mm-hmm.

Louis Rosaenberg, PhD (46:27.937)
what we found and was very and and we were able to get statistically significant results really quickly when you bring together a group this big is that you know the global views or date of data centers are really different than Amer views of Americans. Americans really, really negative on data centers. the rest of the world much more I mean there's concern, but

Vit Lyoshin (46:44.535)
Ha ha.

Louis Rosaenberg, PhD (46:55.779)
What we saw was you know, 80% of Americans against data centers were globally like 63 to 65% of the population is like positive on data centers. And so a complete flip. And we also saw that you know the Americans were really mistrustful of the AI companies, rest of the world less mistrustful, and it was different in different places. You know, we saw that people in India.

much more positive on data centers, people in Spain much more positive on data centers. People in the UK more negative. They were the the closest to the US was the UK, which was interesting. and so but but you what you know these interesting insights come up when you can do a global conversation. And again that was you know that was a little over two hundred people, but you know we're on a path where we could, you know, ideally bring together 10,000 people and have, you know,

Vit Lyoshin (47:36.8)
Interesting.

Vit Lyoshin (47:46.798)
Yeah.

Louis Rosaenberg, PhD (47:56.184)
a really large scale conversation and get really, really deep insights really quickly. and so yeah, it's it connecting the connecting the world together in this way is is definitely something that we that we have a lot of excitement around and you know it has you know civic applications in terms of just enabling communication and it on on

Vit Lyoshin (48:05.474)
Yeah, that's very cool.

Louis Rosaenberg, PhD (48:25.827)
you know, across across borders. It it also has, you know, business applications in terms of you know, there's lots of multinational companies that have, you know, salespeople all around the world and they can't necessarily have a coherent conversation together. And so we do work with big companies where they wanna connect they want to connect people across countries, across time zones, across languages, and and and enable them to forecast together or

Vit Lyoshin (48:38.126)
Show

Louis Rosaenberg, PhD (48:55.105)
enable, you know, we have a a use case where there are customer service representatives all around the world and they're all hearing different complaints from customers in different languages all over in all over the place. We can bring them together and they can they can have a very fast discussion about you know what's the you know what's the most important issue that we need to fix in our product to satisfy customers all around the world and you can have you know

Vit Lyoshin (49:20.493)
Mm-hmm.

Louis Rosaenberg, PhD (49:24.023)
Two hundred and fifty people have that debate and get a really good answer in you know, in forty minutes. It's so it is it enables things that you just would never have thought was even possible.

Vit Lyoshin (49:37.879)
Yeah. Yeah. I mean when we're talking about different languages, cultures, places people live and day to day issues they have to deal with in different countries and regions, that's taking this solutions to the next level because for many years companies operated in regions and for one region it's one product, for another region it's a different product, right? So or flavors of like you know, coke or whatever.

So that was interesting because I came from Central Asia and we had different flavors, which I still remember in my childhood. I mean when I came to America, it's slightly different. I'm like, wow, okay, new experience. So go figure. Yeah.

Louis Rosaenberg, PhD (50:23.425)
Yeah, and and the other thing that's interesting was when you bring a group together to have a deliberation, you really understand the reasons why they feel the things they do, because they're they're discussing it and they're arguing it. Whereas when you give people a survey, you might know, okay, most people think this, but don't really know why. And what we f like in looking at the data that that we collect when we have these large groups deliberate, you know, we saw, you know, we were looking at like who are the different regions that are that are against data centers.

And we saw you know the US and the UK were against data centers. We also saw that there was in Russian-speaking population there was a a lot of negativity about data centers, but it was for a completely different reason. It was because their crypto mining data centers were were creating problems in in those regions, the where the US were upset about you know AI, AI megascaler data centers.

And so they had the same they they had the same negative feeling, but they but when you look at the underlying reasoning because they're discussing it, you see that okay, they you know they're aligned for different reasons. and then you can look at like, well, why do pe you know why does you know people in in India support data centers that you know they're looking at it as, you know, this is a path of economic growth, this is you know, this is an engine for generating revenue.

Vit Lyoshin (51:34.11)
Yeah.

Vit Lyoshin (51:48.269)
Mm-hmm.

Louis Rosaenberg, PhD (51:48.772)
You look at Europe, why they're supportive of data centers. It's you know, they don't want to fall behind the US and China and other places. And so they all have different reasons, but again, if you gave a survey, you wouldn't really know those reasons. You would just know the you know the high level the the high level facts.

Vit Lyoshin (52:00.663)
Interesting, yeah.

Vit Lyoshin (52:11.316)
Mm-hmm. Okay, interesting. we're almost coming to an end here and I just wanted to ask lastly if you have any sort of final thoughts for people to leave them with in terms of how to think about this collective superintelligence, maybe some practical things they can try or you know, learn more about your tool, or whatnot.

Louis Rosaenberg, PhD (52:34.987)
So I mean my you know my high level advice, especially for people working in the AI space or or leveraging AI, is to really remember that people are smart. And and I say that because there's so much of a push to just replace people with AI. And and it's again, we we just don't appreciate how smart people are and how creative people are, and again, how good people are at solving problems with very limited amounts of information. And so

I think this general path of having humans and AI work together is is a really good path whether or not you're doing it at large scale like we do, or even just you know at small scale. So I you know I'm a big proponent of keeping humans in the loop and and definitely not replacing humans from critical decisions, which is which is happening. in terms of our you know, our product, we have we have a platform called Thinkscape.

that it that enables large groups of up to 250 people to have you know conversations in real time and that the upper limit keeps going up and it's really easy to use and people could find it on our our website which is just unanimous dotai and and now you can bring groups together across language or they can all be in the same language they can get together and they can have a large scale conversation by by text or by voice or by video and

Vit Lyoshin (53:35.33)
Mm-hmm.

Louis Rosaenberg, PhD (54:05.449)
And we you know what we find is it's it's really useful for everything from you know doing really accurate group forecasts or or prioritizing or decision making to just open-ended brainstorming. And we see groups get really, really creative when they can all brainstorm in these large groups. So it is it's a easy to use and it's a again, it's the it's a thing where large companies don't

Vit Lyoshin (54:21.89)
Mm-hmm.

Louis Rosaenberg, PhD (54:34.125)
don't do it because they don't realize it's even possible to have, you know, a division of 250 people, you know, to just prioritize their their objectives together in in real time. And once they realize they can do it, they realize, you know, yeah, like large groups of people are actually smarter than s than small groups of people. And and so they get, you know, part of it is about getting, you know, harnessing all of the human intelligence and part of it is just that

Vit Lyoshin (54:55.502)
Ha ha ha.

Louis Rosaenberg, PhD (55:04.719)
You know, we you know large human groups can very quickly converge on really smart answers.

Vit Lyoshin (55:12.918)
Yeah, yeah. No, that's great. Yeah. so before you go, Luis, I have three my typical innovation questions I ask every guest. So hopefully this will be quick. so can you first question is can you please define innovation in a few words?

Louis Rosaenberg, PhD (55:19.959)
August.

Louis Rosaenberg, PhD (55:32.238)
Yeah, so to me innovation is you know making a leap to s to solve a problem in a new way that's never been solved before. And and it can be it doesn't matter how sophisticated the solution is. It's that it's just as never sometimes the most magical innovations are really simple things that solve a problem in a way that nobody's done before. And sometimes innovation is a really complicated thing, but it

Vit Lyoshin (55:58.115)
Mm-hmm.

Louis Rosaenberg, PhD (56:01.296)
You know, it it to me it is a new solution that just is not n nobody's seen before and it makes it you know, it makes things either better or faster or or or just run smoother and that's innovation.

Vit Lyoshin (56:17.002)
Okay. K second question is which one innovation in the history you think changed the world the most and why?

Louis Rosaenberg, PhD (56:26.605)
Yeah. I I mean I'd have to say it's written language just because without written language in terms of that that leap is really what enabled almost every other leap past because it allows it it it allows the advances to spread across time. I mean there you know before written language advances did spread across time, but it was it spread slower.

with written language it spreads across time and also spreads across you know a across populations. So I I feel like almost everything you know hinged on that innovation.

Vit Lyoshin (57:08.576)
Okay. Got it. And the last question is something that we use today, like a technology or software or thing, that we will be laughing at ten years from now.

Louis Rosaenberg, PhD (57:21.303)
Yeah. So, you know, so right now one of the you know, obviously one of the the biggest technologies is AI and we're you know we we all sit down the computer and we can ask, you know, get all kinds of input from AI. And I think, you know, five to ten years from now we'll we'll look back at the fact that we had to sit in front of a screen to engage the AI as as silly. Because the the AI will be, you know, a voice in our ears that we can just talk to.

And and we won't think about it as I gotta go to a computer to engage it. And so these you know these AI systems will be embedded in wearable devices that go with us wherever we go. we can talk to them, they they can give us answers and responses and they they're aware of the context in our world. And it will and it will it will really change what we think of an AI as being.

It also raises all kinds of new risks and and dangers when we have, you know, AI that is going with us wherever we go and seeing whatever we see. but I I think there's there's no question that that will be a very rapid transformation. It's a rapid transformation because it is the direction that every major major technology company is pushing. whether it's you know and it's really the direction that the whole mobile phone space is going.

Vit Lyoshin (58:40.385)
Uh-huh.

Louis Rosaenberg, PhD (58:45.879)
And we will get to a place where you know, a mobile phone will also seem very old fashioned that you have to look down at a screen and instead, you know, it'll be a wearable device and it will just feel like it's in eventually just feel like it's invisible part of part of our life. And and we're close to that. And again, it's you know, meta it it meta and Google and Apple and

Vit Lyoshin (59:02.038)
Yeah.

Louis Rosaenberg, PhD (59:15.155)
you know, every large company's pushing, you know, pushing very hard in that direction. And it's I think it'll be the the next the biggest change in just what we think of a what we think of a computer as is is coming over the next over those next five five really five years.

Vit Lyoshin (59:36.426)
Okay, well looking forward to this. would be curious how it looks and feels and works. So okay. All right, Luis, thank you very much for your time and sharing with us all these ideas and information. That's been great.

Louis Rosaenberg, PhD (59:52.151)
Yeah, yeah, thanks. It was a fun conversation.

Vit Lyoshin (59:56.236)
Yeah. All right, we'll stay in touch. Thank you. Bye bye.

Louis Rosaenberg, PhD (59:59.214)
Thanks, bye.