Elliot Ludvig is an associate professor in the psychology department of the University of Warwick, who coordinates the MSc Behavioural and Economic Science, and teaches the most amazing course there: Neuroeconomics. In this part of the interview, he’ll explain how he came into touch with neuroeconomics and whether he would call himself a neuroeconomist. Disclaimer: everything mentioned in this interview is, of course, an opinion.
Q: As a quick introduction, would you tell us some more about your own background, and whether you would call yourself a neuroeconomist or not? A: It’s a funny thing. The thing that got me into research is the paper by Schultz, Dayan, and Montague. I read that as an undergraduate. In April 1998, I went into my thesis supervisor’s office and he gave me that paper. It immediately struck me—the combination of cognitive science, neuroscience and AI. It was just such a fantastic opportunity and led me to pick up some of the older papers as well.
The lab I was in then might have been the original neuroeconomics lab, or at least one of the very first. My supervisor was doing neuroeconomics in the 1980s, before neuroeconomics was an actual thing. He was trying to find a neural basis for utility through brain stimulation in rats. This was a long time before there was a movement toward neuroeconomics as a field.
At the time, I was doing animal work: brain stimulation, lesion work and even a little bit of recording. I was trying to do experiments looking at aspects of timing and how it related to decision-making and learning. I did that for a few years. I was always on that edge of the field and I ended up doing my PhD in animal behaviour with regards to timing, learning and decision-making. It was later, in my post-doc with Rich Sutton and beyond that I became computational. Rich Sutton is effectively one of the founders of reinforcement learning within computer science. Funnily enough, I have the newest edition of his book, often called the bible of reinforcement learning, right here. Opening up a random page, you can see these figures from the very same Schultz, Dayan, and Montague paper. They do look familiar. And what even pleased me more is that some of my work and notes made it into the book. I was especially pleased that some of my notation made it into the book, as he was more of a mathematician and I was the “resident psychologist,” studying in an AI, computer science lab. There, we were thinking about AI and how it could apply to neuroscience. I wasn’t in a neuroeconomics lab per se; I was in the computer science lab, from which a lot of ideas in neuroeconomics were drawn.
Later, when I was at Princeton, I was in an actual neuroeconomics lab, although I did more behavioural work. There was fMRI, and lots of collaboration. My neuroeconomics, however, was still computational. Most of the time I was still modelling or predicting behaviour, although that was heavily inspired by the neuroeconomic perspective. Funny thing is, in all that time, I didn’t speak to a single economist.
Laughs. Was that an issue you think? It wasn’t on purpose! In some ways there’s two fields, there is neuroeconomics and decision neuroscience. There are reasons why people affiliate with one or the other. They do try to answer the same questions, but from different perspectives, So I was doing neuroeconomics, but it wasn’t mainstream neuroeconomics. For example, I have never even been to the (Society for) Neuroeconomics conference. I don’t go because I don’t do hardcore neuroscience: I don’t record from brains, I don’t do optogenetics, I don’t do neuroimaging. I work a little with people who do, but most of my work is behavioural and computational, which is just not the same.
When I came here (Warwick), I adopted a very interesting position, as someone with a background in neuroeconomics, teaching one of the very few neuroscience modules here on campus altogether (the MSc Behavioural and Economic Science, Neuroeconomics module). It made me very attuned to what is going on in the field. And I continued contributing on my side of it, as the computational work very much feeds into the neuroscience. There are absolutely amazing behavioural science, computational modelling and cognitive science communities here, which have pulled me into their directions.
The thing that also happened here, is that I started talking to economists for the first time. Instead of getting a second-hand take, I finally got their direct take on neuroeconomics, whilst moving away from it. I still read neuroeconomic studies though. It’s most of what I read, and my own papers still fall in that tradition. For example, my latest study focusses on the creation of habits, suggesting that the way we think about habits is misdirected. The field tends to think that habits are reward-based, based on outcomes. What we argue in that paper is that habits don’t come from reward, habits come from repetition. But habits tend to be actions that get repeated, because they get rewarded. As such it is very difficult to disentangle them. They sometimes are literally the same. We put together a paper, we studied animal behaviour, we studied lesions. An earlier version even had a whole bunch of stuff on human fMRI, which has been reduced and moved to the discussion section. So, it is still a contribution to the field of neuroeconomics, but we did test this behaviourally, and did not use neuroimaging studies. I might end up collaborating with someone who does use neuroscience tools, as often happens in the field, but really that is just not what I do anymore.
So, that is a very longwinded answer to your opening question, as to how I got where I am, and where I am standing right now.
Q: Do you think, situational factors taken into account, that you might go back into hardcore neuroeconomics? Or do you prefer the angle and methods you are currently approaching the field with? I like being a step removed. Partly because neuroeconomics is hard. Neuroscience is incredibly technically difficult, it's labour intensive and there are a lot of things to get right. From the animal care to the surgeries. I am up to date with most, but not all techniques. And to set up a lab, is an endeavor, definitely not something you do on the side. Even from the neuroimaging side, there are a lot of technical skills, statistics and neuroanatomical things you need to get right. Not an easy feat.
So, I’m kind of content to not be so directly into neuroscience now. In some ways I feel I would like to, because it allows us to directly test the theories we try to develop, but I am more of a computational person. I am quite happy to work with the ideas and predict as to what should happen, and as best as I can test those behaviourally. Being one step removed also allows me to not be invested in a particular technique or set of findings, or even the success of a particular approach. Imagine that it turns out neuroscience doesn’t help us understand behaviour all that much and that we are better off doing other things. Like doing better surveys or eye-tracking, or any other external metric, maybe that is the case. It gives me the distance to be more skeptical, because the field does have issues. With behavioural science we have become good in dealing with some of these issues. We have pre-registration to get rid of p-hacking and other ways to stop bias from seeping into the research. Neuroscience has not done this at all. There are very few, maybe even no pre-registered fMRI studies.
Q: Would it be possible for such a study to be pre-registered, given the complexity of fMRI? Why not? If you are skilled and you know what you’re doing, you should be able to lay out your pipeline in advance. You should be able to say: “This is how we are going to do the motion-correction and the spatial-smoothing. These are the metrics we are going to use and how we are going to assess that.” That’s not how it’s done, mostly researchers spend months looking at the data from different angles. There is good reason for that. It is an incredibly valuable but expensive resource, which opens up the possibility for torturing the data to make sure you get a result. I’m not saying people in the field aren’t aware of these issues and are not trying to handle them as well, but I like not being the person having to do all of this. I like standing back, and say: maybe not. I don’t have this stake in neuroimaging being the right tool. If it works, great, if it doesn’t, ok. I have other things I am invested in, such as computational modelling. If it turns out that all we need is a good regression or statistical model, rather than a computational model, then I am out of business.
It’s also a question of impact. One thing you see a lot in neuroeconomics and the way the field has been developing is a movement towards computational psychiatry. Which is applying reinforcement learning, decision-making and other types of learning in the brain to psychiatric disorders. Especially for brain disorders, you need to look into the brain. However, if it’s all about behaviour, it becomes a different question. I’m pretty confident that studying the brain matters, it teaches us about how behaviour works, rather than studying behaviour alone. I’m not 100% sold on that this will always be true. That is an opinion I hold with openness to revision, such are all my opinions. But this one I hold with even less certainty. I see a lot of things that have been dubbed as neurobollocks. They are very popular articles, yes. Yes and Molly Crockett had a Ted Talk on it as well. There is a lot of good stuff out there, and some not so good stuff. Some of my students send me articles for their blogposts (In the module writing blogposts is a form of assessment) and then sometimes I’m like: “really?!” Only sometimes though! And then I have to invent a way to politely discouraging them from using those. Unless… They are aiming to discredit it? Or critically analyze the study done? Sure, but that is harder to do. And some of the students might feel they don’t have the experience to do so and as such are not as willing to do that. But we do have to teach critical thinking especially when it comes to science! Absolutely. I try to do so in a formative and gentle way, suggesting that there might be stronger pieces of work out there. I do encounter a few things that seem less than useful. I am hopeful that good science will grow and continue, and we do have good science. There are really fantastic things out there, such as dopamine learning and classical conditioning. There are things we have learned from merging fields into neuroeconomics. There are a lot of cool papers coming out about what we call model-based learning. These are slightly more sophisticated. They require more planning, they are more thoughtful. They don’t involve a common currency such as reward but focus more on predicting individual behaviours.
The story of dopamine itself is becoming enriched in that way. How we understand what neurotransmitters do in the brain is becoming more enriched. This feeds back into our understanding of how people learn from experience and how they learn to make decisions in the real world.
This is the end of the first part of the interview. In the next part we will discuss where Elliot thinks neuroeconomics is going and what to be excited for next! The next part will also carry the references for articles and books mentioned here.