From Our Neurons to Yours

NeuroForecasting: how brain activity can predict stock prices or viral videos | Brian Knutson

• Wu Tsai Neurosciences Institute at Stanford University, Nicholas Weiler, Brian Knutson • Season 8 • Episode 7

Share episode

Neuroscientists have spent the past few decades tracing the network of brain systems—some deep and emotional, and others more analytical and deliberate— that work together as we make tough choices like where to invest our money as well as more everyday decisions like which videos to watch online—or, for that matter, which podcast to listen to.

You can imagine that the ability to listen in on the brain systems that guide our choices might start to let scientists predict our decisions. But today's guest has taken this a step further, showing that measuring brain activity in just a few individuals can actually forecast widespread social behaviors, like which stock prices are likely to go up or down on the market, or which videos are likely to go viral. 

Join us as we talk with Brian Knutson, a professor of psychology in Stanford's School of Humanities and Sciences, about the frontiers of neuroeconomics, bridging psychology, economics, and neuroscience. 

Learn More

Send us a text!

Thanks for listening! If you're enjoying our show, please take a moment to give us a review on your podcast app of choice and share this episode with your friends. That's how we grow as a show and bring the stories of the frontiers of neuroscience to a wider audience.

We want to hear from your neurons! Email us at at neuronspodcast@stanford.edu

Learn more about the Wu Tsai Neurosciences Institute at Stanford and follow us on Twitter, Facebook, and LinkedIn.

Nicholas Weiler (00:07):

This is From our Neurons to Yours, a podcast from the Wu Tsai Neurosciences Institute at Stanford University, bringing you to the frontiers of brain science. 

When we make decisions, our brain draws on a whole network of systems, some deep and emotional and others more analytical and deliberate. It's a complex process, but obviously a fascinating one. And neuroscientists have spent the past few decades tracing how these different brain circuits work together as we make tough choices like where to invest our money or more everyday decisions like which videos to watch online or for that matter which podcast to listen to. 

Now if you have been listening to this show, you can imagine that the ability to listen in on the brain systems that guide our choices can let scientists start to predict our behavior. Today's guest has taken this a step further though. He's showing that measuring brain activity in just a few individuals can actually forecast widespread social behaviors, like which stock prices are likely to go up or down on the market, or which videos are likely to go viral. 

Brian Knutson is at the forefront of the field of neuroeconomics, which bridges psychology, economics, and neuroscience. In other words, he studies the science of decision-making. So let's just get right into our conversation. 

Brian Knutson, welcome to From Our Neurons to Yours. It's great to have you.

Brian Knutson (01:40):

Thank you. I'm very excited to be here. I'm a fan of the show and I hope I don't mess it up too badly.

Nicholas Weiler (01:46):

No, we're going to have a great conversation. The thing I want to talk about today is one aspect of your lab's work in an area called neuroforecasting. This basically involves a series of studies that suggest that you could use brain imaging to predict large-scale human behavior, like stock market decisions, whether a video is going to be popular, whether people are going to buy certain things, and so on. But I think before we get into that, why don't we start by talking about what we know about individual human decision-making, what we are coming to learn about the committee in our heads. I like to think of it like the movie Inside Out with different characters in our heads helping us make decisions. Could you take us on a brief tour of the parts of the brain that are involved when we make decisions? I was thinking maybe we could start with a toy example. My producer Michael recently gave into his children's demands for a puppy and now he has this cute little fluff ball that's destroying his house, and he wanted to know how his brain got him into this mess. And so I thought maybe we could do a little tour of what are the key systems involved in making this decision?

Brian Knutson (02:58):

Yes. How do people make decisions? How is the brain involved? And it really is the story of a field called neuroeconomics. It's evolved over the last 20 years. And it really came about, I think because we got these new techniques for looking at brain activity and we could see not just what was happening in the cortex at the surface, but activity oxygenation, neurons being active, we're below the surface even deep in the brain and the brainstem and the subcortical areas that I'll talk about.

Nicholas Weiler (03:28):

And this is with the rise of functional magnetic resonance imaging in the late '80s, early '90s.

Brian Knutson (03:34):

Yeah, exactly. Which I'll now say fMRI after having said that. And that arose in the early '90s and people started using it to look at phenomena that they understood better like vision, looking at the back of the head, motor output, moving your hands or something that's going to be up here at the top of your head. Those are well-mapped cortical areas. But then people started to ask more sophisticated questions about, well, what about choice? What about ... Well, maybe not buying a puppy, but maybe do you want to take this risky gamble or not? This is an old behavioral task that has been used since the time of Descartes. People, and people still want to know who's going to gamble, who's not going to gamble, when are they going to gamble? What's going to cause them to gamble? This is the stuff of actuarial science. This is the stuff of the casinos of Las Vegas. So we can take that as a behavioral fruit fly, if you will of understanding choice. And we can ask the question, what happens before people take a risky choice? What has to happen? Can we see that in the brain? Can we use it to predict what they're going to do?

(04:40):

Is it just visual cortex? Is it motor cortex or is there's something in between the input and output that is determining which choice they're about to make when they haven't made it yet, but they're about to do it? And so let's return now to your example of buying the puppy. So when we think about risky choice, we think about a few ingredients. One ingredient is potential gain. You wouldn't be interested if there wasn't something in it for you. You're anticipating that something good might happen. But when most people think of risk, they think of the downside of risk. So they think of the scary part or the potential loss. So yes, you could potentially gain something, but you could also potentially lose something and you don't know what's going to happen. Maybe you have some information like this is a pretty good bed, or that's not a good bet.

Nicholas Weiler (05:24):

Your neighbor has a puppy. You had a puppy when you were a kid.

Brian Knutson (05:27):

Yeah. Yeah. It's so cute right now, but you're not thinking about the pooping. But then somebody mentions that. So there are all kinds of considerations that go into a choice like this. Some of those considerations you can write down on a ledger sheet and you can say, "Okay. The puppy costs this much. That's a cost, but it also is very fluffy and I love to touch it, and so I'm going to put a benefit on that and balance those out." But then maybe something happens at the last minute, like your kid breaks down crying because you can't have a puppy and that tips the balance.

(05:58):

So all kinds of ingredients go into this risky choice. It's amazing that we can make these choices, and yet of course we have to make these choices because otherwise you wouldn't be here. If a big bear runs into the room, I better run out of there. I better stop talking to you and in a few seconds I have to make that decision. And just because we're very sophisticated podcast people talking about abstract topics doesn't mean those parts of our brain went away and that they're not monitoring the situation and not guiding our actions. And I think that's one of the lessons that we're learning from neuroeconomics actually, is that even in sophisticated kinds of choices, we're using a lot of our brain. We're not just using the parts of the brain that represent numbers and the parts of brain that represent probabilities. So neuroeconomics is the study of that.

Nicholas Weiler (06:45):

And one of the amazing things with the rise of fMRI ... People had have been doing experiments in animal models for some time trying to understand choice. But with fMRI being able to really control what is the choice that the person is trying to make? What are their values, what are the risks and the rewards that they're evaluating? With fMRI can really see as someone is making a decision, what are the pieces of the brain that are responding? And so I'd love to get an introduction, dramatist, personae for this conversation. What are some of the key brain systems that we're going to be talking about as we talk about how we make these decisions at an individual level?

Brian Knutson (07:23):

So now that we've introduced fMRI and we can see a millimeter resolution activity, if you will, at a second to second scale, what that means is that it's not that fast, but it's pretty fast. It means at the speed of choice, you can see what's going on in there and what happens before the choice. Now you have a behavioral problem. How do you set up an experiment with humans where you can elicit this kind of choice and separate out the components and time and then see what is active during that choice? Let's just use the risky gamble example because we've used that a lot. So you show people a couple of gambles, so we can ask you the question, okay, and I can ask you this right now. Would you rather have a gamble where you can have a 50% chance of a gaining $2 and a 50% chance of losing $2 or a gamble where there's no chance of anything happening? What would you prefer?

Nicholas Weiler (08:12):

Yeah. I guess I'd prefer the one where there's a chance of gaining $2.

Brian Knutson (08:15):

And losing $2. Okay.

Nicholas Weiler (08:17):

Yeah.

Brian Knutson (08:17):

That means that you're risk seeking. Because most people ... I'd say two thirds won't take that gamble. And the reason is the expected value of that gamble, if you add that up, is zero. So you're comparing zero to zero. Most people are risk averse, but I can do things to that gamble to make you risk seeking. But most people are, so you're a risk seeker. Congratulations. I can evaluate the gamble for you too. I have money here. I'm good for it.

Nicholas Weiler (08:42):

Change the value and you'll find my threshold.

Brian Knutson (08:45):

That's right. If I raise the loss, you're going to say no. So anyways, we can change all this stuff and present it to people in the magnet, have them make this choice. So we can say, here are the gambles now make a choice for real money. And then they see the outcome so they know we're not joking and they see how much they're making. It's like a video game. People love this actually. They love to do this in the magnet.

Nicholas Weiler (09:05):

Online gambling. Big deal.

Brian Knutson (09:07):

In the life, right? Look at the lottery. So anyways, so that's the setup. Now we say what happened in the brain when you saw those gambles? Can we track where the risky gamble is as in the top or the bottom? And lo and behold, we can look in the visual cortex and we can see that mapped in the visual cortex.

Nicholas Weiler (09:22):

You can see what they're actually seeing, the screen that's presenting.

Brian Knutson (09:25):

Yeah. And people tend to dwell on the risky gamble processing it. Then we can also ask, we can counterbalance this button means you choose the risky gamble, or this one does on this trial. We can counterbalance left or right choice. We can say, can we see that in the brain? And we can. It's beautiful. We see you press the right button, you see the left motor cortex, left button, right motor cortex and now the question is what's in between? And we can use fMRI to look at that too. So we can say, "Okay. Now fMRI, tell me if people are going to choose a risky gamble or not." Lo and behold, the brain regions that tell us that are not the visual input or the motor output, they're in between and they're lower. And they're in these regions, these subcortical regions, some of them or old cortical regions that we think receive neurotransmitters that are modulatory. So one set of neurotransmitters, dopamine, I'm sure people on your podcast have talked about dopamine. That projects from the midbrain into some cortical areas, subcortical areas, including the striatum, which is in the middle. It's got a weird shape. And at the bottom. And the middle of that is the ventral striatum, which has an area called the nucleus accumbens. Now we're getting into jargon. Nucleus accumbens just means that the clump of neurons that lean against the third ventricle.

Nicholas Weiler (10:36):

All of these things were described first and then we figured out later what they did.

Brian Knutson (10:40):

Exactly. Yeah. These were described with obscure Latin terms, who often fought with each other over the descriptions.

Nicholas Weiler (10:46):

My favorite is the substantia innominata, which is the unnamed substance. We don't know what it is.

Brian Knutson (10:51):

We don't know what it is. Well, that's what we'll call it.

Nicholas Weiler (10:54):

Okay. So nucleus accumbens, it's one of these structures deep in the brain. I like that you mentioned that this is an older structure, and I think what you mean there is that this is something we've inherited from our mammalian ancestors deep in time. It's not this fancy neocortex that makes us so clever, but it's something that is really core to mammalian decision making.

Brian Knutson (11:16):

This goes way back in the mammalian and even before that lineage. So if you look at the neuroanatomy of dogs and rats and mice you can see it. There are homologues of cortical areas too, but there's certain parts of the cortex that are unique to humans, like the very frontal pole of the cortex.

Nicholas Weiler (11:33):

The nucleus accumbens you were saying is getting dopamine signals. And so what role do we know the nucleus accumbens to play in that decision-making process? Let's say Michael is seeing a picture of this cute dog, he's thinking about buying.

Brian Knutson (11:46):

Yes. So one thing we've seen over and over with these risky gambles is if you want to predict if somebody is going to choose a risky gamble, look in the nucleus accumbens. If that activity is going up, then they're more likely to choose that risky gamble. And if it's not, they're more likely to take the safe gamble. This is consistent with dopamine being released there. We're not measuring dopamine, but we can in animals. And so that's one character that's very important for us. And there's another character, I guess, if you will, and it's an old part of the cortex called the insula, which means island deep inside and you can't see it from the outside. So this area seems to be more active when people are not going to take the risky gamble. And so why is that? We don't know, but we know that other neurotransmitters also project there like norepinephrine. So it seems to be having a countervailing role.

Nicholas Weiler (12:33):

And when you're looking at the activities, you can see nucleus accumbens is getting active and that predicts that someone might be interested in the risky option. And the insula getting active predicts that someone is going to avoid that risk. Are these things that people seem to be consciously aware of? Does this correspond to an experience or is this happening before or under the radar of our conscious evaluation of the decision?

Brian Knutson (13:00):

Now, the funny thing about our research is even though we've been doing this for over 20 years, we don't know exactly how conscious people are of this stuff because in order to figure that out, you've got to probe them in real time and measure the brain activity in real time, which we're doing. But the consciousness layer of this is an additional complex layer that we haven't worked out. We think sometimes people can be conscious of it, sometimes they might even be able to change it, but certainly not all the time because this activity is happening fast. And so almost by definition, we know how long it takes for people to reflect and report what their feelings are. It's like a couple of seconds. But we're seeing this activity happen quite rapidly. So it may promote conscious awareness of some feelings.

(13:43):

Now I think what you're getting at here though is another phenomenological layer of what's going on, which is what does this mean in terms of experience? Doesn't mean anything. And we do think it means something, and this is the part I haven't told you yet, but it's really the core of what we do. Is we think it's related to certain feelings that people have before they make a choice or before something happens. So nothing has happened yet, but people still imagine it might happen and they have feelings about it. Those feelings motivate them to go forward or to move back. And we call it anticipatory affect. These are feelings akin to say excitement and anxiety. And so we think of the nucleus accumbens activity being more aligned with excitement, the anterior insula activity being more aligned with anxiety. Both of those things are going to happen under a risky choice. But it's the relative balance of those that will determine whether you move forward or move away. So that's how we think of this anticipatory affect.

(14:33):

And in fact, when I talk to my colleagues about this, they're puzzled because they're used to thinking about feelings as being a consequence of what happened. So you took the gamble, you lost money, you feel bad. That certainly happens, but this has to happen before to get you to move forward or move away. And that's critical adaptively. And so when I talk to stockbrokers about this, they're like, "Oh yeah, of course. This is greed and fear." They intuitively have a sense of what this is about. Now, are people conscious of this? Are they aware of it? Are they monitoring it? Are they regulating it? How often are they doing that? That's what we don't know. And I suspect it's a lot less than people think.

Nicholas Weiler (15:11):

Right. Our decisions may be more guided by our impulses than we like to think. There are occasions when we approach decisions deliberatively. Hopefully Michael did this with his new dog. But there are many times when we just do what feels right and see what the consequences are. And you mentioned just to make sure that we've brought this up, areas in the front of the brain that are particularly developed in humans, this prefrontal cortex that's behind your forehead, above your eyes, or behind your eyes. And is that where we think that some of the more evaluative processing goes on where you're saying, "Well, I'm a little anxious about this, but I think that the kids will be really happy and there might be some downsides, but I had a puppy when I was a kid." Bringing in your personal experience and your values and the things that make your decisions different from someone else's decisions.

Brian Knutson (16:02):

Yeah. So there's a neuroanatomical aspect and a phenomenological aspect of this. So neuroanatomically we have lots of really great data from animal studies showing that these prefrontal regions, especially the middle of the prefrontal cortex, are connected directly to the striatum, the ventral striatum, the nucleus accumbens, and probably partially to the anterior insula as well. And so they can talk back. They have direct top-down, as we would call them, projections. Now the nucleus accumbens in the anterior insula are also connected to those areas, but in an indirect manner. So they loop back through the thalamus and then forward. It's not a symmetrical information highway, if you will. So we have reason to believe that input from those areas will matter in terms of what's going on in these deeper regions. So that's the anatomical layer. The phenomenological layer that I haven't talked about yet is time.

(16:58):

Sometimes we just consider what's going to happen next, and that's important. But we don't always just consider what's going to happen next. Sometimes we think about the future, sometimes we think about the past. Sometimes we think about how will my kid respond to this dog every morning? Sometimes we think about, I had a dog and I love that dog and it was a big pain in the neck, but boy, it was one of the best things that ever happened to me. We can step outside of the here and now and then integrate additional experience or imagined experience into our choices. We think the prefrontal cortex, especially the midline of the prefrontal cortex may be important for integrating that kind of information in.

(17:34):

And this leads to like us looking at these three areas basically, before people make a choice. We actually have a little framework, what we call the AIM framework or the affect integration motivation framework. It's a hierarchical idea that when you first encounter a risky, you have this anticipatory affect, but then you have to integrate those responses with other kinds of value that are maybe more long-term. And then you have to move. You to be repetitively or reversibly motivated towards or away from that choice. And by the way, that implies that you don't have to do it right now, you could do it later. So you might be able to do the affect and the integration and then decide, yeah, I'm going to buy that item later. So that's why we use that acronym, but it also helps us to know where to look when we're trying to predict if you're going to choose the gamble or buy a dog.

Nicholas Weiler (18:22):

And that's why a salesperson or a marketer would prefer for you to make the decision right now, because then you're going to be making the decision, presumably based on those initial emotional responses. Someone who's trying to sell you something doesn't always want you to have time to reflect on it more cognitively.

Brian Knutson (18:37):

That's exactly. Look how cute this puppy is. Don't think about the poop.

Nicholas Weiler (18:41):

Okay. So I feel like we've got a good cast of characters here. I'm sure there are many more we could talk about, but primarily we're talking about nucleus accumbens excitement for something, even if it might be risky. Insula, a little anxious about risks, not sure if it's a good idea, prefrontal cortex, looking at the evidence, integrating past experience, imagining the future, all that stuff.

(19:22):

Now let's get back to what we teased at the top, which is your work on neuroforecasting. I find this really fascinating. It's a little bit science fiction to me. The idea that you can use people's brain activity and often a small group of people's brain activity to predict the behavior of a much larger group, how people are going to respond to stock prices, how people are going to respond to a video online. And you know what it reminds me most of actually? You know the Foundation series by Isaac Asimov.

Brian Knutson (19:51):

Oh yeah.

Nicholas Weiler (19:52):

So this reminds me of Hari Seldon a little bit with his psychohistory where he basically can predict hundreds or thousands of years of human history based on the principles of human psychology. So I assume we're not at that point, but it does have a flavor of that. And so I guess maybe you can tell me what is your hypothesis about how you can get insight into large scale behavior by just studying the brain activity of a few people?

Brian Knutson (20:20):

Yes. So when we first started trying to do this work, it seemed crazy. And one of my grad students who's now faculty at Rotterdam, Alex Genevsky said, "Hey, I'm really interested in charitable giving." And he had done some work actually on what causes people to give to orphans. And it turns out if you see their face that causes them to give 33% more. It turns out the nucleus accumbens is involved in that. So coming back to the puppy example, if you see the puppy, you be more likely to buy it. But at any rate, he said, "Hey, I know there's this platform online, kiva.org, and it's a micro lending platform and people can go to this platform and contribute to give people a loan." And we can get the data from them. It turns out like some Stanford students in symbolic systems started that platform. But we also happened to know that some of our colleagues had been looking at music and had been able to use brain activity to forecast which of the music would be very popular on MySpace at the time.

(21:16):

So at any rate, to make a long story short, we took some of the appeals off of the Kiva website and had people make incentive compatible decisions. Do you want a loan to this person and that person and so forth. And he was able to use brain activity actually to forecast which of those appeals would eventually receive the loan on the platform. So not only would a person make the loan or not, but also what we call forecasting, which is ... By forecasting we mean out of sample. And that's why we say prediction within a person and forecasting out of sample.

Nicholas Weiler (21:47):

So you can predict what a person will do, and you can forecast from a small group of people what a larger group might do.

Brian Knutson (21:53):

Yes. Maybe that doesn't seem very surprising. But what was surprising is that the behavior of our sample was related to what happened on the platform. It correlated but not that strong. But what was really correlated with what happened on the platform was their Nucleus accumbens activity when they saw that appeal.

Nicholas Weiler (22:10):

So you're saying that the actual decisions that the research subjects made was not as predictive as their brain activity?

Brian Knutson (22:17):

That's right. Yeah. Which doesn't really make sense if you're a behaviorist or a revealed preference theorist in economics because you think the best indicator of a group's choice would be an individual's choice, not some piece of that choice. But it started to plant the seed in our minds that well, maybe by breaking down the choice into components, some of those components are going to generalize more than the others, and we can just see those with brain imaging. That led to us doing other experiments in other kinds of markets, including crowdfunding. And we saw a similar thing with crowdfunding, but even more pronounced, which is that the Nucleus accumbens activation in our group forecast which of the crowdfunding appeals would succeed on a Kickstarter. But the choices of our group did not. So in this case, the brain activity was what was actually allowing us to forecast whether these Kickstarter appeals would succeed or not. So the more that we did this, the more that we started to find these kinds of anomalies where the brain activity is either doing better or the only information that we can use to forecast what's going on online. So another example of this is not just giving money to some cause or buying into something, but it's spending your time on something. So when you think about what we do on the internet-

Nicholas Weiler (23:34):

It's the attention economy.

Brian Knutson (23:35):

It's the attention economy, and it's scary. You asked about insight. You asked about do people know what's going on? And this is one of the reasons I think we don't, because if I look at my iPhone and it tells me how much time I'm spending a day on the iPhone, when this first happened, it provided feedback. I was shocked because I was spending four hours a day on my iPhone. And that's like half of my working hours.

Nicholas Weiler (24:00):

That's activating your insula right there. You're like, oh boy.

Brian Knutson (24:03):

Yes. At least momentarily, right? Yeah. I was like, "Holy smoke. I got to get this down." Anyway, the point being that we spend a lot of time online and I'm not sure that people are aware of how much time they're spending online, and a lot of the platforms seem designed to get us to spend more time online even if we don't want to.

(24:21):

So another student, Lester Tong, who's at UBC, he's a postdoc, said, "Hey, I wonder if we could look at video virality on YouTube because we can get those metrics off of YouTube. What would forecast that? Would the brain activity forecast with people's choices what they wanted to watch?" And again, what we saw was that our group's nucleus accumbens activity to the beginning of the videos would forecast how popular they were on YouTube, that is how many downloads per day. Whereas what our group told us they wanted to watch was not related to that. So there was again, some information in the brain activity that allowed us to forecast popularity of these YouTube videos even when our subjects behavior wasn't necessarily telling us very much. Again, this is a paradox because it suggests that some components of choice that are more generalizable than choice behavior itself.

Nicholas Weiler (25:06):

So you've had all these experiments where you've seen that the brain activity can forecast what larger groups of people are going to do, whether it's with spending time watching videos online or making financial decisions based on market trends and so on. But my understanding is that for a while it was unclear why this was happening. Why is it that the brain activity is even better than looking at people's self reports of what they were thinking? So you had a paper this summer I think in PNAS Nexus that tried to understand a little bit more what is happening here? Why is the brain data so effective at forecasting what large groups of people are going to do? It comes back to this model you were talking about where we have these very fast emotional responses, whether they're excitement or anxiety, followed by more individualistic evaluation in areas like the medial prefrontal cortex, where we're bringing in our past history, our values, our imagination of the future and so on.

(26:09):

And that maybe it's because we share these emotional responses from our deep evolutionary history. Those are going to be more predictive of what lots of people are going to do. And the more individual evaluation of the choice is going to be more variable. I'm going to make the decision, I'm going to evaluate the decision differently than other people. Was this the hypothesis you were coming in with? Because I should say that latter thing, the slow decision-making to use Kahneman's thinking fast and slow model, that's what you're seeing with or people's self report or people's behavior, that's the output of their behavior. But maybe it's less predictive of what lots of people are going to do because it's so individualistic.

Brian Knutson (26:53):

Yes, that is definitely the spirit of what we were trying to look at. Part of it came from pushback really, which is great from my marketing colleagues who said, "I don't understand how this can work. Because if your sample is representative of the population, then your behavior should be like the population. And if it's not, how can the brain activity then forecast if your sample isn't representative? It doesn't make sense."

Nicholas Weiler (27:16):

If you're looking at a group of Stanford undergraduates, how can it translate to the population at large who isn't much like Stanford undergraduates, for example.

Brian Knutson (27:24):

And it let us start to think about generalizability. It's like, wait a second. Representatives? What does that even mean? You're thinking of that as an individual, this an individuals like those individuals, but what if we think of representativeness in terms of choice components? This part of your choice is more like everyone else, and that part of your choice is not. So that kind of flips everything so that suddenly it's like, oh, well, if we break down the choice along these components, we can actually assess which of them is more similar and which is not.

Nicholas Weiler (27:55):

And they're parts of our brain.

Brian Knutson (27:58):

Here's your sample, here's the market. And so again, this is Alex Genevsky in Rotterdam who really conceptualized this, but he said, "Look, let's just measure the demographics of our sample. We can measure things marketers would measure. How old are you? What is your gender? What is your race?" Very basic kinds of demographic attributes they would measure. And then we can actually create on a market. So this is what he did. Created a market on the internet for these. Because we had the stimuli that we had used in the magnet for watching videos or for funding crowdfunding requests. So create these markets of two or 3000 people on the internet, measure their demographics too. And so then what we could do is split up the markets in terms of who is similar to our imaging sample and who's not. So then the question is pretty straightforward. It's like, okay, let's look at behavior first. Are the marketers right? Is it true that if we look at our samples behavior and the market sector's behavior both representative and not does it match? And what we found was what they would predict, which is the behavior of your sample is related, is associated with the behavior of the representative part of the market, but not the unrepresentative part of the market.

Nicholas Weiler (29:03):

So you're studying some people in the magnet, and then you're looking to see if the behavior of those people doing these videos or crowdfunding matches the large group of people on the internet. And if the people on the internet who are similar to your sample in the magnet, their behavior matched pretty well.

Brian Knutson (29:20):

That's right. That's exactly right. I wouldn't say well, but it matched at least a little bit. Now let's look at the brain activity. Again, we have brain activity from this group, and we're looking at is that forecasting the behavior of these different market sectors, which are representative or not? And what we see there is that there's no difference in every sector the brain activity is forecasting the behavior.

Nicholas Weiler (29:42):

So it's forecasting the behavior for people who are totally different from the people you are actually studying.

Brian Knutson (29:46):

Yeah. Yeah. People who look demographically different. We have the vector of eight attributes.

Nicholas Weiler (29:51):

And when you say brain activity, is it all the brain activity or was it ... We've been talking about-

Brian Knutson (29:56):

No. And that's what's also interesting. Yeah. So it's mostly nucleus accumbens activity in these markets. These are markets are for crowdfunding, for video viewing, and it's mostly nucleus accumbens that is forecasting. If you look at medial prefrontal cortex activity, which we know in these subjects is predicting their choice, it's not forecasting the market as well. Even within the brain, there are some circuits that seem to generalize more than others. So this kind of changes the whole idea of what is representativeness. And it's consistent with this idea that you mentioned at the beginning that maybe some choice processes are more generalizable than others. Now that might depend on the market, but at least in these markets, it looks like these initial affective, as we would call them, responses seem to generalize more than the later integrative responses or measured responses or whatever.

Nicholas Weiler (30:45):

Well, the thing that I love about this is that basically what this is saying is, okay, two things in my mind. One is these emotional, let's say brain regions, the nucleus accumbens and perhaps the insula, we haven't talked about that as much, are more deeply conserved. And so you're going to see that signal in the large group. And the more cognitive response, the evaluation, that's going to be much more idiosyncratic. And so over a large group of people, that's going to average out. Some people are going to go one way, other people are going to go another way. That signal is not going to come out in the large aggregate data. And so what you're left with is that overall the immediate nucleus accumbens response is going to be a better predictor of the responses of the large group because everyone shares that nucleus accumbens response. And then what they do with it might vary a little bit, but you've got the same starting place.

Brian Knutson (31:35):

Yeah. That's our working model right now. We call it partial scaling, but again, we're looking at different markets. One thing we haven't talked about is time scales too. So we're making these point forecasts of the markets because we have these nice markets where either it's funded or not, or this video is downloaded a lot this day or not, but the market's changed so that's another whole caveat.

Nicholas Weiler (31:57):

Your work has many different applications. I mean, we've talked before about how some of this decision making work is relevant to addiction, for example, why people make certain choices despite really negative consequences. And that's something I'd love to have you back on to talk more about. That could be a whole episode on its own of course.

Brian Knutson (32:15):

Absolutely.

Nicholas Weiler (32:16):

And we've talked to your colleague, Keith Humphreys from the NeuroChoice Initiative and Rob Malenka as well. But I'd love to have that conversation more in depth. The thing I wanted to ask you about now is what are some of the implications of this? You're talking with colleagues in economics, and clearly there's probably a lot of interest in people who make viral videos and knowing what's going to make them viral. There's interest by companies in influencing people or knowing what's going to make people buy stuff. So what do you see as the implications of this? Does this become psychohistory where we can predict what large populations of people are going to do? Where do you see this going?

Brian Knutson (32:54):

I feel like we're one step closer to Psychohistory, but not like Hari Seldon. It's like maybe we can make a point estimate in a couple of months. That's pretty good. That's better than what we had before. And when you think about markets like the stock market, what's your reference standard that you're comparing to? If you can do 54% prediction, stock market's doing 50%. So it's like even a slight advantage might really matter. And I like to think about those kind of markets now. Where are the markets where we're doing terrible at forecasting? So advertising might be one for instance. And there are other ones too. Right now, for instance ... And I'll just tease you with this, but we have data and we're writing it up. Can you forecast first month box office returns on movies based on people watching trailers? The answer is yes, you can.

Nicholas Weiler (33:40):

Interesting.

Brian Knutson (33:40):

So that's an interesting market, right? Because it's hard to evaluate, it's dynamic. There are all these factors. So I think there's a big potential in these very uncertain markets, particularly for novel types of products. Are marketers using this? No, not to my knowledge. It takes a lot of overhead. It takes a lot of expertise. The equipment is large and in basements, and so it hasn't really penetrated into market applications. There have been waves of neural marketing applications where the people have thought, well, if I take this other more peripheral measure, maybe I'll get the information I need that remains to be seen. I don't know if they're going to be able to do that or not. I've witnessed about two or three waves of this. For us, we're really interested in human choice, and now we're interested not just in individual choice, but how does that scale to the aggregate?

(34:27):

I thought coming into this, there'd be well-established theories of how this happens in economics and or psychology, and I cannot find great theories about this. So I think we actually need a theory of how this happens and how it dynamically unfolds and what kinds of generalities will spin out over time. So we have a lot of work that we need to do in that domain as well. But also I think there's nothing that speaks like application. So if you could show that you can get a better forecast than the industry standard, to me that's very compelling and it means it's a vindication of whatever minor theory that we have so far.

Nicholas Weiler (35:06):

I guess I just want to represent a discomfort that I think people sometimes have around the idea of using brain data for commercial ends. And I'm curious what you think about that. Is that something that you've grappled with in thinking about how some of the work that you're doing might be used or abused in the future?

Brian Knutson (35:24):

Yeah. I'm not rich.

Nicholas Weiler (35:26):

You're not using this to play the stock market, Brian?

Brian Knutson (35:31):

Maybe I will. Maybe I will, but not yet. What that implies is that I'm a little bit hesitant too about trying to capitalize off of the applications. It's not that I don't think that we should. I think when a new technology arises, it can be used for good or evil. And so in the forecasting sense, we're averaging the data across a group. So there's less of this individual identifiability thing going on, although one can imagine that maybe I want to predict what you are going to do and not somebody else. And we do have that issue with looking at addiction, for instance, and relapse. But ethically, yeah, if you can forecast on certain markets, which ones should you be dabbling in? Which ones shouldn't you be looking at? I don't have a clear line that I know to draw there. Perhaps ethicists can help us with this.

Nicholas Weiler (36:19):

Yeah. It feels like there's a concern that these social media companies are mining our attentional foibles for profit. And so this is where ... Wow. You see why that works. You see why it's effective to really delve into some of these core human impulses and reactions. And the brain data is reminding us that we can be manipulated. These are deep-seated emotional systems to guide our behavior from millions of years ago, and it's good to stay mindful of that. And for me, I'm coming away from this just remembering that those feelings of excitement or anxiety should be a first reaction and maybe not a final reaction.

Brian Knutson (37:01):

I agree. I guess two points I'd like to add. One is that if I discover something and I write it down and I publish it, then that's available to you, Nick to read, and you have access to that. If a company discovers it may go into their basement. You may never see it again. They may use it to extract profit from you and worse. Your attention. I don't need to go much further because this is happening on the internet right now. So I think it's very important as a scientist to actually study this and understand it and try to convey to others what is going on. Because a lot of it is probably happening below the radar of our consciousness, our conscious reflectiveness. That's one reason I'd rather have me discover it than a company that may not have the user's interests at stake. The other reason is that maybe there could be technology that counteracts that other technology that's trying to seduce us in and make us spend four hours a day on our iPhone and be angry at other people and basically be the kinds of people that we don't necessarily want to be.

Nicholas Weiler (38:04):

Right. How can we use this decision science to become the people we do want to be?

Brian Knutson (38:08):

Exactly.

Nicholas Weiler (38:09):

Yeah. That's beautiful. And I love a thing that you said earlier about maybe this can also be helpful for just for improving, getting things to people that they actually want and like. Companies are making big guesses. You talked about movie companies. Maybe this reveals that some small time movie producer is making something that people are really going to respond to, and that gives the studios evidence enough to take their own risky gamble on something. So it's nice to think about some of the potential positive outcomes as well as the things we should have our eyes out for.

Brian Knutson (38:45):

Absolutely. And it is my responsibility to think about that so thanks for helping me to do that.

Nicholas Weiler (38:53):

Well, this has been such a fascinating conversation. Always happy to talk to you about the future that's coming out of your lab and others.

Brian Knutson (38:59):

Likewise.

Nicholas Weiler (39:00):

So thank you so much for coming on the show. This has been a pleasure.

Brian Knutson (39:03):

Thanks so much, Nick.

Michael Osborne (39:05):

Fantastic. I still don't know why we got this damn dog. But when my wife asks-

Nicholas Weiler (39:12):

You will.

Michael Osborne (39:12):

Well, when my wife asks, "Why did we get this damn dog," I can say, "I have just the podcast for you."

Nicholas Weiler (39:19):

Hopefully. Just blame your nucleus accumbens.

Michael Osborne (39:24):

Thanks again so much to our guest, Brian Knutson. He's a professor of psychology and Stanford School of Humanities and Sciences. Check out his work in the show notes. We've got links to publications. We've got links to past news stories about this fascinating work. So if you want to learn more, check out his work in the show notes. And if you've been enjoying the show, please subscribe and share with your friends. That's how we grow our audience and bring more people to the frontiers of brain science. We'd also love to hear from you about whatever is working for you on the show or things you'd like to hear more about. If you've got thoughts, leave us a comment on your favorite podcast platform or send us an email at neuronspodcast@stanford.edu From our Neurons to Yours is produced by Michael Osborne at 14th Street Studios with Sound Design by Mark Bell. I'm Nicholas Weiler, until next time.