- 6 hours ago
Aaru co-founder & CEO Cameron Fink talks with Maurice Levy about simulating humanity, reaching the unreachable & rethinking the science of prediction.
Category
🤖
TechTranscript
00:28What is interesting today is that you will have the two extremes of the demography in VivaTech.
00:36The oldest one and the youngest one.
00:40Or the other way around.
00:42And in total, 58 years old.
00:47Unfortunately, in two months' time, Cam, you will be able to buy drinks.
00:55That's true.
00:56And that you will be able to do things that you are forbidden to do.
01:02But you can create a company.
01:04That is very true.
01:06And you have done it with two of your friends who are co-founders.
01:13So maybe you start by telling a little bit of the story.
01:17I will tell you how we came here today.
01:21I read a story in the Wall Street Journal which was staggering about the success of our room.
01:34And I thought, okay, this is the guy I want absolutely to be in VivaTech and to share his story.
01:44I called Cam.
01:46I discovered that he's a French man, but he doesn't speak French.
01:52And that he is probably one of the most clever persons I have met.
01:58And we had a wonderful conversation.
02:01And now, Cam, maybe you tell a little bit of the story.
02:06I know that it is a bit boring that everyone say, oh, you are so young and so bright.
02:13But maybe you can tell.
02:15Not at all.
02:16Not at all.
02:16First off, thank you guys so much for having me.
02:19The story of Aru begins 813 days ago.
02:23Or actually, maybe it begins four years before that.
02:26I, you know...
02:27Four years you were...
02:29Four years ago.
02:30Four years before that, I would have been 14.
02:33And from 14 to 18, I was a researcher.
02:37I was doing quantitative political science and then artificial intelligence and machine learning.
02:42I actually say AIML because I was doing computer vision, which at the time was, like, the fun, interesting part
02:48of ML.
02:49And it was not NLP or language models, which is, like, the oft-forgotten area.
02:54And so I was a researcher in that area for some time doing some really interesting stuff.
02:58I actually started a company doing skin cancer screening.
03:02You know, I'm very pale.
03:04Skin cancer runs in my family.
03:05And so me and one of my now co-founders started that business.
03:09And I always was interested in entrepreneurship and starting a business.
03:14We ran that business together for actually longer than Aru.
03:17That was about two and a half years.
03:20And we were about to sell the business.
03:22And that's when I met John, my now co-founder, via LinkedIn cold DM.
03:28He sent me a message on LinkedIn.
03:30And we scheduled a 30-minute Zoom call.
03:33We ended up talking for two and a half hours.
03:35And that was the moment I said, I'm dropping everything.
03:38We're starting Aru together.
03:40And our mission was always to predict human behavior.
03:42We always wanted to train models to understand how and why people make the decisions they do.
03:49Easy.
03:51Isn't it?
03:52Tom, maybe you can tell us a little bit what Aru does exactly.
04:00I'm not sure that we understand very clearly.
04:05And probably that the venture capitalists and the funds we have invested have not understood anything also.
04:14But maybe you can tell us a little bit more.
04:17Yeah, so what we do is we train models to predict human behavior, right?
04:22So we look at the world of traditional research just to explain why we do what we do.
04:27We know that polls, focus group surveys have sampling bias, incentive bias, survey exhaustion.
04:34They have all these issues that make it very difficult.
04:36I mean, when you look at the consumer products world, only 30% of products that launch are still on
04:41shelves two years later in the CPG industry, right?
04:44So we know that market research is broken.
04:46If market research works, then 100% of products would be on shelf.
04:50And so what we did instead is we said rather than ever training on top of what people say they
04:55do or who they say they are,
04:57we train on top of real behavioral outcomes directly.
05:01So we train models on top of things like credit card purchase history.
05:04We train models on top of foot traffic data, music streams, podcast listens,
05:09anything where we can actually measure a decision that a human made and by measuring that decision be far more
05:15accurate at predicting the real world outcome.
05:17And so our models are able to segment out and predict the composition of any population on the globe
05:23and then understand what decisions they make, what elections they're going to vote in,
05:28who they're going to vote for, what products they're going to purchase,
05:31what marketing campaigns they're going to click on and will impact them.
05:34So you create synthetic personnel.
05:41Yeah.
05:42Simulations of people.
05:43Simulation people.
05:44Clones.
05:46Yes.
05:47In many ways.
05:47Virtual clones.
05:48Yes.
05:49How do you feed them in order to have a representation of the population?
05:57Because if you want to have a good result in terms, at least of what the state of the situation
06:07is,
06:08you need to have a fair representation of the population.
06:12It's all focused on ground truth data, right?
06:15So we're not looking at who self-represents themselves as, say, you know, a lover of a given brand
06:21or a person who self-represents themselves as, like, a tennis player or a golf player.
06:26Instead, we look at the raw consumption data.
06:29We look at census data.
06:31We look at academic research.
06:32We look at real population data.
06:35We look at social media engagement.
06:36We look at foot traffic data.
06:38Anything that actually helps us figure out who exists in a population.
06:42And then what you can do is you can give me the description of any audience on the globe.
06:45Genuinely, almost any audience you can name, whether that be, you know, the simplest, highest level,
06:50U.S. household purchase decision makers.
06:52And then the most niche, complex category of software users for third-party risk management software, right?
06:59Or for consultants who work in big four consultancy firms.
07:04You know, any niche group of individuals you would want.
07:07And we can segment out that group, predict who exists inside of it, and understand their behavior as well.
07:13So how do you get the access to the data?
07:16Yeah, it's tough.
07:18I mean, the data access is one of the most interesting problems we have to solve.
07:22So we always start with the three different groups of data sources that we have, right?
07:28You can think about us as leveraging a lot of publicly accessible data, things like government data sources, academic research.
07:35Then there's a lot of purchase data, right?
07:37So we buy data from a bunch of different vendors.
07:39We're buying credit card data.
07:41We're buying audience data.
07:43We're buying measurement data.
07:45And then the third group for us is partnerships.
07:48So we actually work with a bunch of NGOs, nonprofits, even small businesses who want access to the impact that
07:56we can drive.
07:57They want simulations, but they don't fit the shape of a traditional client of ours.
08:00Typically, they're too low in volume.
08:02And so we actually exchange simulations with them to benefit their cause and license to train on their data.
08:08And that's a very powerful motion for us as well.
08:11So you mean that Nielsen and Kantar and Ipsos will be out of job soon?
08:19There's always going to be a place for traditional market research.
08:23You know, I think, like, an agent can't feel how much something weighs.
08:28An agent can't tell you how a product tastes.
08:31But certainly, when you look at what we're doing, we work with a lot of the big CPG companies.
08:36You know, when you send a product to human research, you have eight product concepts, right?
08:39Or six product concepts.
08:40And you send it to human research and you say, how do these six different beverages taste?
08:45Well, with us, if you use us before you use that human research, you can cut down on the amount
08:51of human research you have to do, which is good.
08:52Because it is costly.
08:54It is time costly as well.
08:57And it is inaccurate in many ways.
08:59So we can instead test 200 concepts.
09:03And then the six concepts you already go to human research with, we know have already been validated by the
09:09market through simulation.
09:10And so it's a lot more efficient, even if there is always still going to be a place for traditional
09:14market research.
09:16You are not a so old company.
09:19So how do you know that it is really working?
09:23We measure our accuracy rigorously all the time.
09:26And we measure it in a lot of different ways.
09:28The best type of test you can imagine is a live test.
09:33So we will come and we will give a, you know, election prediction before the election occurs.
09:38Or we will help predict the performance of product concepts and how well four different SKUs are going to rank
09:44before those products actually hit the shelf.
09:47And that's the best test.
09:49And can you avoid, for example, when we are speaking about the votes and the election, can you avoid that
09:56people are voting for the wrong people?
10:00That's not our job, unfortunately.
10:02No, it's not your job.
10:03But we can predict if they're going to vote for the right people or the wrong people.
10:08It depends who you think the wrong people are.
10:10So you have not yet the right to vote and you can measure it?
10:15I'm 20, so I can vote in the States.
10:17And you can vote.
10:17But my co-founder...
10:18Certain states.
10:20My co-founder, who is 17, cannot vote.
10:25But he is very good at predicting elections.
10:27Our technology is the most accurate predictor of elections on the globe.
10:29What are the skills of your 17 years old co-founder?
10:37And he is, if I understand well, he is the CTO.
10:41He is the CTO.
10:42He's one of the smartest, if not the smartest person I've ever met in my life.
10:47I think...
10:47You have not met many people.
10:48Well, you know, you do have a good point.
10:51You do have a good point.
10:52It's fair to say.
10:53It is very fair.
10:54But I moved 17 times.
10:56So now I have, inverse, met more people than the average 20-year-old, I would say.
11:02His super skill is the capability to take very complex problems
11:06and reduce them into their simplest components
11:08so people like me can understand it.
11:12You can understand.
11:13I can understand the problems when he breaks them down for me.
11:17Great.
11:17So I understand also that you are moving to a new level and new approaches.
11:28Is it because the synthetic personnel is a dead end
11:35or is it because you have found something more interesting?
11:39Some things are dead ends, right?
11:41So there's a lot of noise in this market right now.
11:43There are a lot of people who are looking at it.
11:45You know, this went from, like, off-forgotten research area
11:48to all of a sudden being truly important because of its economic impact almost overnight.
11:53What is a dead end is this idea of taking a language model
11:57and training that language model on top of survey responses.
12:00We know that's a dead end because we tried it.
12:03And it doesn't work.
12:04The issue is once you fall out of domain of what survey responses you've trained on,
12:08it's no longer going to be accurate.
12:10And you have to ask yourself, are you trying to replicate surveys
12:13or are you trying to replicate the real world?
12:16Are you trying to replicate behavior?
12:17Because I think the far more powerful thing to do is try and replicate behavior.
12:21Surveys try to measure behavior.
12:23So why don't we try and replicate that behavior directly?
12:26And that's why we ended up taking the approach we did.
12:29We built all of our own models in-house.
12:30We train on our own data.
12:32We're the only people who train on this idea of real-world objective data,
12:37actually measuring the decisions that people make.
12:40How many people do you have in our room?
12:43As of today, 33.
12:4533.
12:47We're growing, though.
12:48We'll be over 40 by the end of the month.
12:49Yeah, you're growing fast.
12:51Incredibly quickly.
12:52We just moved into a new office,
12:53and we're already looking for the new office after it.
12:56Okay, great.
12:58Who are your customers?
13:01The politicians, the parties, the marketers, the advertising agencies?
13:09We don't really work in politics anymore.
13:11Our biggest customers tend to coalesce in one of three sectors, I would say.
13:16The first is consumer businesses or consumer brands.
13:18So CPG, retail, technology, anything that touches a consumer at some point in its lifetime.
13:25Then we do a lot of work in financial services.
13:27You know, we work with wealth and asset management providers,
13:30helping them sharpen their own value propositions when they go to clients
13:33and helping them test the features and functionality that attract new people to their service.
13:39And then I would say the third category for us is just broader kind of media marketing,
13:44the big agencies we work with as well.
13:49When you think about the future, because the future is yours,
13:56where do you see Aru moving?
14:00Do you think that you will be able to really predict the future
14:04and what the future will be for human beings, particularly in the world of AI?
14:09I think the question no longer becomes just how do we predict the future,
14:13but it becomes how do we shape it, right?
14:15If you believe that human behavior is, you know, the most critical component of our planet, right?
14:21If you think humans are the preeminent species on the globe,
14:24which I think you and I probably agree on,
14:26then predicting behavior or predicting human behavior is the same thing as predicting the future.
14:32And if you can predict the future, you can shape it, right?
14:35This isn't just the power for us to go and say, you know,
14:39here's which product of these four we think will succeed.
14:42This is the best marketing campaign among the ones you gave us.
14:45This is now the power to actually shape the products that will perform in market
14:50and the power to go create new marketing campaigns out of total greenfield
14:55and actually figure out the best possible customer to sell to.
14:59It's the power to shape so much more about the world.
15:02I think that is surely the direction we're headed in moving forward.
15:05As this technology gets more powerful,
15:07it's only going to be more and more important that we use this as a source of impact
15:13and as a source of good in the world.
15:15It's funny because it looks like a fairy tale in an area which is quite sophisticated
15:24and which requires a lot of knowledge, a lot of AI and a level of imagination,
15:33which is normally the case for more mature people.
15:38What have been the obstacles or the difficulties that you had to overcome?
15:46Yeah.
15:47I tell you what, it's a very, very tough problem to solve.
15:50Like on a technological level, the fact that we're this accurate at predicting behavior,
15:55you can think of us broadly, like to put a number out there, 97, 98% plus able to predict
16:02behavior,
16:02at least in recreating the general social survey.
16:05The fact that we're that accurate is incredible.
16:08And so one of the problems we've had to solve is figuring out exactly how to do that, right?
16:14Which status sources do we leverage, training our own models.
16:17It's all a long journey.
16:18And I actually think for us, the biggest question, you know, hands down,
16:23the biggest obstacle is building the right team and recruiting the right people.
16:27It's really, really, really tough.
16:30It's the toughest job you ever have to do as an early stage startup founder
16:32is bringing on the best possible people you know.
16:36Because that is make or break for your company.
16:39You can have the best idea, the best technology.
16:41But if you don't have the best people, it's very tough to make it work.
16:45And this is something we know well in more mature companies,
16:49that if you have not the right talent and if you have not access to the right people,
16:54and creating a team, which is not only a series of individuals,
16:58because creating a team is something which is very difficult,
17:02you go nowhere.
17:06I was playing chess, and less so now.
17:11I have no time for that.
17:13And one of the things that I was always doing is playing differently
17:18according to the people I have in front.
17:21And I could, if not predict what he will do,
17:25but according to his behavior, the kind of personality he is or she is,
17:31I knew that some moves will create some issues to him or to her,
17:39and I may win, even if I was at a smaller level,
17:44or not as good as he or she was.
17:49And this is something which is difficult to take into account,
17:54because this is the human factor, the emotional factor.
17:58How do you do it?
18:00The emotion is one of the most important components behind what we do, right?
18:05I think part of our technology,
18:07we actually never aim to replicate objectivity.
18:11We want to replicate subjectivity.
18:13There's some biases we want to set away.
18:16We don't want sampling bias.
18:17We don't want incentive bias.
18:18But there's some biases we want to keep.
18:21Humans are naturally biased individuals in how they make their decisions.
18:24And so we go to great lengths.
18:26You have not did that.
18:28It doesn't take much effort, even in my 20 years.
18:32And so we go to great lengths to make sure that our models
18:35are able to reflect the reality of emotional decisions that humans make.
18:39You have to.
18:40So we have seen, with the rise of internet and the development of mobile phone, etc.,
18:51the rise of what we call the digital natives.
18:56And this is already an old idea.
19:02So are we seeing, and do you see, the rise of AI natives?
19:08And if that is happening, how different will they be from their predecessors?
19:15I 100% think we are seeing the rise of AI natives.
19:19There's no question.
19:20I have two younger siblings.
19:22My younger sister likes to call ChatGPT chat just by itself, which is now becoming increasingly popular.
19:29I think most people in school right now would call it chat, and using chat.
19:36And 100%, I think, like, when I...
19:38And they are not using Claude?
19:40They do use Claude.
19:41They use ChatGPT and they use Claude, as the big two, I would say.
19:45And some people use Gemini.
19:47But I'll say, like, ChatGPT released when I was in my senior year of high school.
19:52At that point in time, like, I was insane for trying to use ChatGPT to, like, go write an essay
19:58or help write an essay or write an outline for something.
20:01Nowadays, I think that's the standard practice.
20:04Like, I would tell my siblings they actually need to use AI because it can provide a leg up.
20:10And I think if you're not willing to use AI in the academic world or in the professional world, you
20:16probably will not be as competitive of a job applicant, a college applicant, whatever it might be.
20:22That said, I think a human touch is still important to everything.
20:26So, if we move back to your younger sister, how old is she?
20:33She is 17.
20:3617, and she has not yet created her startup?
20:40So, what's happening in the family?
20:42What's happening in the family?
20:44Oh, she'll feel terrible if she watches this.
20:47I have a younger brother, too.
20:49He's 14.
20:50And he has created his startup.
20:53I think he...
20:55We're going to get him to get to work on it, is what I tell him.
20:59Okay.
20:59That's the right combination because you are Australian, French, American.
21:05I grew up mostly in the States, but I moved around a lot, yeah.
21:09My parents...
21:10My parents...
21:11And how are you French?
21:13My father's French.
21:15Oh, yes.
21:16Because there is something right in your...
21:18That's where the entrepreneurial spirit comes from.
21:21In your DNA.
21:22Yeah.
21:23Feel lucky.
21:24Okay.
21:26Cam, college dropout is something which is quite common in the tech world.
21:32We have seen a lot of those very brilliant minds who dropped out.
21:40And you have done exactly this.
21:44And I understand that you have been in the college for one day.
21:49One night.
21:50That's all it was.
21:51One night, yeah.
21:53At Dartmouth.
21:54So, can you explain that you don't believe in higher education?
21:59You don't believe in what knowledge can bring to you?
22:05Or is it that suddenly the grace from the sky came and...
22:11I actually...
22:12I fully believe in higher education.
22:14And I fully...
22:15Even above higher education and the institution of higher education, I believe in even one
22:19greater principle above that.
22:21And that is education.
22:23Just the power of education.
22:24The power of learning.
22:25The power of knowing things.
22:26Of getting to meet new people.
22:28When I decided to drop out of college, it's not because I didn't believe in college.
22:31I think had I gone to college, I probably would have been happy.
22:34I think I would have succeeded.
22:36I think I would have enjoyed it.
22:37And there's a lot to learn.
22:39Like, trust me, I was very, very excited to take a bunch of classes at school.
22:42But the thing about Aru and what I was doing, there are two things.
22:47The first is, if you truly believe in the principle of education, go out and do what
22:53teaches you the most.
22:55And for me, that was 100% going to be going out, starting a company, and entering the real
23:00world.
23:01And learning and meeting so many people in this business has provided me more opportunities
23:06to meet interesting people and to do cool things than ever before.
23:10I mean, I don't think you would have invited, you know, college student Cameron Fink, you
23:15know, junior at Dartmouth up to speak at the very least.
23:18And the second thing is, in starting a business, it's like an irresistible draw.
23:23Is it the wrong decision when we start the company?
23:26100%.
23:26There's so many people who drop out of college, and for every story of Mark Zuckerberg, there
23:30are 100 people who did not succeed as much.
23:35But you feel this irresistible draw to want to go prove to yourself, at the very least,
23:40I need to try.
23:41And I feel very, very, very lucky that I followed that path, because otherwise, I would not have
23:46gotten to start Aru if I hadn't said to myself, you know what, I'm at the very least going
23:51to give it my best effort.
23:53So you are at the age of my granddaughter.
23:58My youngest granddaughter is 21.
24:02And she's now moving to a U.S. university.
24:09And what will I tell her?
24:12I will tell her, I had a nice conversation with Cam Fink, and he dropped out, and maybe
24:21you should think about stopping your education, or should I advise her to continue?
24:28I think for most people, you should be in higher education.
24:34I think if you're in higher education, it's largely a good thing if two conditions are
24:38true.
24:39If you don't know what you want to do, university and college and higher education is the best
24:45institution in the world to explore things and find out what you want to do.
24:49And the second thing is, if you need a degree to get there, if your granddaughter is studying
24:53to be a doctor, if she's studying to be a mechanical engineer, if she's studying to
24:59be a lawyer, then she should stay in college.
25:02It would be very tough to be a doctor without a degree.
25:04I don't know if anyone would show up.
25:05You can be.
25:06There are some.
25:08But legally?
25:09No.
25:11I would like to come to something slightly different.
25:20When you are meeting a CMO and meeting a boardroom executive, how the relation is made?
25:36What are the difficulties or the problem when that CMO sees that he has his son in front
25:44of him, will be teaching him what to do for the best approach to sell his product?
25:53It's all about the work.
25:55You know, it's not about who I am.
25:57It's not about how young we are.
25:59It's not about even the business or being a billion-dollar company.
26:02Like, all of that is helpful.
26:04It's something that people look at and they say, that's validating.
26:07It's impressive.
26:08But if it weren't for our product and what we've been able to do and the outcomes we're
26:14able to drive, there would be no business, right?
26:16And so for whatever CMOs or people who would look at us and say, wow, you're young.
26:22You're my son's age.
26:23Ultimately, it just comes down to the fact that the proof is in the pudding and the product
26:27proves itself.
26:27And we win because we're accurate and because we drive real outcomes for real businesses.
26:35You know, including when we had that first conversation, I was thinking that in some
26:43religions, there is the reincarnation.
26:47And I have the feeling that there has been a lot of generation which have been reincarnated
26:53in your body.
26:55I feel very lucky.
26:57You know, I get told sometimes, as much as I am 20 years old, I do read, I was telling
27:01you yesterday, the Financial Times and the Wall Street Journal every morning in paper.
27:07You know, I think it's very important to actually feel and read the news.
27:10Since when?
27:12Since, I think I started doing that when I was nine.
27:19I was in middle school.
27:21Maybe I was 10.
27:22I hope that you had a childhood and that you had some games and the pleasure of sharing some
27:32moments with kids and not reading the Financial Times at nine.
27:37I think it's good to read the news.
27:39It teaches you a lot about the world.
27:41I feel very, very, very lucky for having moved a lot growing up.
27:45I feel very lucky for, you know, my parents and having seen everything that I've seen
27:50at the age of 20.
27:52I think I'm lucky.
27:53But I did have a childhood.
27:55I played a couple video games.
27:58I was always really bad at them, too, is the funny thing.
28:00Like, there's this...
28:04There's this perception that a lot of startup founders are really, like, excellent League
28:09of Legends gamers or something.
28:11You know, like, it wasn't me.
28:13I was not that good at video games.
28:15You remind me of another grandchildren that I have, a grandson, who, at age nine, has organized
28:26a voting bureau.
28:30He was reading Le Monde, and he decided that he had to organize the vote of the kids in
28:38order to get the white president elected.
28:40That's incredible.
28:42That's incredible.
28:43I applaud any people who start in the world of politics or take interest in politics from
28:50a young age.
28:50The first, you know, the first business I ever started, and this is when I learned...
28:54Yes, because he has had already many businesses.
28:58This is when I learned I was not going to do physical goods.
29:01We imported 20,000 boxes of crayons from China, and they were called political crayons, and
29:08the idea was to encourage people at a young age to be involved in politics and get out
29:11to vote.
29:12But I spent my bar mitzvah money on this.
29:14I was 13.
29:16This was...
29:17Now, bar mitzvah is always 13.
29:19Yeah.
29:19Always 13.
29:20At least you know that.
29:22And I bought these 20,000 boxes, and they had names like Bernie Blue, Trump Tangerine, Liberal
29:28Lime, Conservative Crimson, and I went out to sell them.
29:32And of all 20,000 boxes, I think I sold 500 of them.
29:35I think I worked at least 500 hours to make this work, and I turned a profit of $250.
29:43You know, for anyone who's read up on America, that is approximately 5% of minimum wage.
29:49A little above.
29:51Maybe 6.5% of the minimum wage.
29:53But I learned a lot building a business, you know, at 13, or at least selling crayons at
29:58the farmer's market at 13.
30:00You know, I think if it weren't for that, I wouldn't have been able to do what I do now.
30:05So, Cam, thank you very much for your time.
30:08Thank you for coming to Paris.
30:09You have French roots, and it's the first time that you were in Paris.
30:14Yes, it is.
30:15Thank you for inviting me.
30:16And you are fantastic.
30:18We wish you all a great, great future, and to make ARO as famous as Meta, Google, maybe
30:30NVIDIA.
30:31You're setting big expectations.
30:33I'll get to work on it.
30:34No.
30:34Okay.
30:35Predict the future.
30:36A rose of applause.
30:39Bravo.
Comments