 So, the internet has opened new markets through which billions of dollars now flow. And at the same time, neuroscientists have devised new methods for predicting individual choice. The question I'd like to explore today is, can we use those new methods to say something to forecast aggregate choice, even on the internet, even in markets? For a long time, neuroscientists haven't even contemplated the ability to predict choice in humans, because we've known that circuits deep in the brain in animals will support self-stimulation, highly motivated behavior, and even avoidance of stimulation. But we couldn't visualize those circuits with human brain imaging methods. We just lacked the resolution. But in 1992, a method was developed called Functional Magnetic Resonance Imaging, or fMRI. And the advantage of fMRI is it allowed us to look in behaving human brains at a scale of millimeters and seconds, which means that we can observe what's happening in the brain as people choose, and even before they choose. And so, people started to look not only at sensory input, motor output, but also responses to rewards and punishments. And using simple games in our lab, we've actually investigated the responses of these deep circuits and indeed seen that they activate in response to rewards and punishments. But we've learned a couple of things. One is they don't just respond to simple rewards like getting tasty juice, but also to abstract rewards like money. Another thing that we've learned is that these circuits respond as powerfully, if not more powerfully, to anticipating rewards and punishments as to actually getting these rewards and punishments. And so we've been able to identify targets, if you will, that respond to the anticipation of reward and the anticipation of punishment, which suggested maybe we could take activation out of those targets and actually predict choice. And so we've done this in many scenarios. This is a shopping scenario that we did in my lab, where we presented people with various products at various prices, and then we asked them to choose whether they wanted to buy those products or not. This was for real, for real money. And we looked at what happened in the brain, and we saw that reward anticipation circuits responded to attractive products and predicted purchasing. But punishment anticipation circuits responded to excessive prices and predicted not purchasing. Now that's individual choice, and by now many labs have been able to do this with fMRI Predict Individual Choice. But what about markets? Individual markets like the stock trading floor you see are very chaotic and hard to measure in terms of dynamic choice activity. But now we have internet markets, which allows to measure choice on a massive scale. So we've started to try to extend from individual choice to aggregate choice, which might seem simple, but it's not actually. There are a number of things that could happen. No scaling account, I will call it, is the idea that individual choice is so variable that it just sort of averages out in the noise and you can't scale at all. Another account, which I'll call total scaling, is the idea that individual choice is shared by everybody and it scales very neatly up to an aggregate market choice. A third account that we're exploring, and I'm going to call partial scaling, is the idea that if we can deconstruct decision making, we can find the components that are similar across people and improve our forecasts not only of individual choice but also of aggregate choice. And we've started to do this in internet markets. I'm going to describe a new experiment that we've just done on crowdfunding. So crowdfunding is an internet market for entrepreneurship in which entrepreneurs ask for investments and if they get enough investments by a certain deadline, they get funding and if they don't, the investments are returned to investors. And so what we did is we put a number of subjects in the magnet and had them evaluate a number of appeals like this and actually invest in these appeals or not. We measured three things about them. The first thing is what was happening in their brain as they were viewing those appeals. The second was what were their actual choices? And the third was after the experiment is over, we said to them, okay, tell us how much you liked this appeal, what was that appeal, and so forth, so ratings. The question is not only which of these will predict the individual choice, but weeks later which will predict what happens on the internet, which of these appeals are funded and what we found as it turns out is something very interesting. Brain activation in these deep circuits we've been talking about, averaged over the group, predicted which appeals would be funded and which would not. That was not the case actually for the behavior of the people in our experiment, the average choices that they made, nor was it the case for their explicit ratings of these appeals. Now you might think, well they got lucky this time. What about other markets? So we've looked at another market, microlending, which is a similarly sort of crowd sourced market in the sense that people will view appeals and decide whether or not they want to make a loan, and if the loan gets enough funding, then it will be funded otherwise the money is returned. And we did a similar experiment and we observed a similar result. Activity in those deep brain regions actually predicts that a loan appeal will get funded weeks later, better than behavior itself. So it's a sort of miraculous to us, it's new stuff, and it's a new technology. But the big idea here is that by deconstructing individual choices, deconstructing the neural components of choices, we can do a better job of forecasting not only individual choice but also group choice. The fact that affective circuits are involved suggests to us that it might be the choices that involve less reflection, which are the most amenable to forecasting. So these would be choices like, I'm going to take a drug now, or I'm not going to save money for the future. These are the choices that we're most interested in understanding and improving. So my question for you is, if this works in its early days, how can we use this kind of a technology to make the world a better place?