Hello, hello. And the topic today is goats, research, and AI. When I was a student in psychology my advisor, Tom Walston, had this story about goats. And the story was the following: Imagine you have goats and you want to take them to eat grass. There are two ways to do it. There’s a tall hill that takes a while to get there. But at the top there’s fantastic grass. You can take the trodden path up the hill and when you eventually get to the top of the hill, the goats can eat really good grass. And if you take the trodden path, there’s no grass along the path because all the other goats have gone the same way. You just have to get to the top. The second approach is to meander around, take the untrodden path and get to the top. It will take longer but by the time you get there, the goats would have eaten some. Because as you walk the less trodden path, they’ll find all kinds of grass and some of it will be fantastic along the way.
And, he said, that this is a metaphor for how to do research. You can take the trodden path, read all the literature and kind of define your path based on what everybody else has done before. And then you can say, “Person A observed this, Person B observed this, but maybe these are the conditions under which person A is correct.” And then you can work on a paper and show something critical in the literature, in a new light. On the other hand, if you take the untrodden path, you will find lots of wonderful things in the middle. What is the analogy in terms of research? The analogy is that if you read everything in advance, you will force yourself onto the trodden path. You will have created constraints of how you understand the problem. Every paper you read reduces your creativity. Every paper you read says, “Oh, this is how people have defined the problem. This is how people have thought about it.” And your mind will set onto that path. The other way is to not start with reading all the literature, but to start exploring yourself, thinking, and wondering. Start even by doing experiments before you find out everything in the literature. Sometimes it will be costly because you’ll do an experiment and find that somebody else has already done it before or something very similar. But, he explained, that many times you will have more creativity and will start something that is unconstrained by what other people have done. And his recommendation to us was to take the untrodden path.
And that’s basically what I’ve done in my career. When I had something that I wanted to examine, I didn’t start by looking at all the literature and then thinking about what I really wanted to do. I would start with what I really want to do, sometimes even doing the first or second experiment see where the was going, then reading the literature, and redefining the problem, and so on. Sometimes I wasted my time. But most of the time it was a good approach.
Now where’s the AI connection? I think the AI connection is exactly what Tom warned us about. Imagine that every idea we have now, we will go to AI first to help us define, refine, and rethink the idea. What will happen? Our idea will be constrained by the literature. AI, by definition, takes all existing current knowledge and helps us integrate it. So, what will happen to Tom’s approach? Imagine all companies now consult AI in thinking about how to innovate within dating apps or how to innovate within applications for compliance in healthcare. All of this could be very good, but it will not involve exploration. It’s not random. It assumes that there’s a right answer that’s knowable from somewhere in the literature and will do it.
I think that’s what AI is likely to do to us. It’s likely to make us more constrained. It’s likely to reduce human creativity. In the short-term, I think, it will save lots of energy and time and so on. But I think long-term, we might get solutions that are too homogenous. Imagine that everyone wrote essays with AI. Everybody wrote poetry with AI. Everyone invented the next new car with AI. I think it would be a shame because it would be way too similar. Productive and efficient, yes, but way too similar.
So, let’s worry about becoming too homogenous and letting AI do it to us.
