When we say ‘smart’ what do we mean?

Leslie Valiant, the T. Jefferson Coolidge Professor of Computer Science and Applied Mathematics at the John H. Paulson School of Engineering and Applied Sciences, has spent decades studying human cognition. His books include “Circuits of the Mind” and, most recently, “The Importance of Being Educable.”

Notions like smartness and intelligence are almost like nonsense. We think we know what they mean but we can’t define them with precision. Even psychologists can’t agree on one definition. And intelligence tests don’t tell one very much — they are usually justified in terms of correlations with other things.

How do you recognize whether someone is intelligent? There is not one answer; there are many, and they can be inconsistent. Some, like Howard Gardner, have emphasized that there are many kinds of intelligence. I think we’ve reached the expiration date for the usefulness of the term “intelligence” both for humans and machines. We should be able to do better.

I’m a computer scientist and I take a computational approach to understanding what the mind does. In computer science, the main questions, theoretically speaking, have been: What is easy to compute? and What is hard to compute? Some decades ago, I decided that the secrets of human cognition must also be hidden in this problem — that some things are hard to compute for the brain and some things are easier. The main advantage that computer science offers is that one can express capabilities that are more complicated than ones reasonably implied by conventional phrases, and, at the same time, evaluate their feasibility for the brain. 

I started working on a computational viewpoint on cognition 40 years ago. The fundamental challenge I set myself was to find a useful definition of learning. In “The Importance of Being Educable,” I define the concept of “educability.” My view is that it wasn’t “intelligence” that allowed humans to create civilizations, but educability, which involves three aspects.

The first is learning from experience. The second is being able to chain together the things you’ve learned; it’s a kind of low-level reasoning capability that even the simplest animals have because it’s so essential to life. The third is being able to incorporate knowledge acquired from instruction. This last one is very important for humans because this is how culture spreads and science progresses.

“Computers will not take over the world just because they want to. This will happen only if we allow it to happen.”

Educability incorporates both the ability to generate new knowledge by learning from experience and also the ability to transfer that knowledge directly to others. There isn’t the time or the need for everyone to gain the same experiences, such as repeating difficult scientific experiments.

I’d say that machines can be also made educable, and ultimately, we won’t be able to claim that we’re fundamentally different from machines. Current AI systems are not designed to be educable in the sense I define, but machines will likely become more and more capable in that direction. I don’t see AI as an existential threat; it’s just another powerful technology. Obviously, in bad hands, it can be misused, just like chemistry or nuclear physics. Computers will not take over the world just because they want to. This will happen only if we allow it to happen.

There is a downside of being educable. Educability gives us very powerful ways of acquiring new information — we can soak it all up. But we don’t have comparable abilities to check whether the information we get is true or not. We are not well-equipped for evaluating knowledge, theories, or facts. If someone tells us something, if we believe it, we will incorporate it into our knowledge. This can be dangerous. The only cure is to educate people about what propaganda has done over the centuries and make them aware of this human weakness. To inoculate ourselves against disinformation we need to acknowledge our basic weakness. 

— As told to Liz Mineo, Harvard Staff Writer