I spoke with Blake Richards, one of the co-authors of the paper, who applies artificial intelligence theories to his study of how the brain learns. He says that in the AI world, there’s something called over-fitting — a phenomenon in which a machine stores too much information, hindering its ability to behave intelligently. He hopes that greater understanding of how our brains decide what to keep and what to forget will lead to better AI systems that are able to interact with the world and make decisions in the way that we do.
We hear a lot about the study of memory. Is the study of forgetting a relatively new thing?
Within psychology, there’s a long history of work examining forgetting. So that’s not a new field of study. But the neuroscientists — those of us who work with the biology of how the brain works — have not really examined forgetting much in the past. Generally, the focus for the last few decades in neuroscience has been the question of how do the cells in our brains change themselves in order to store information and remember things. It’s only been in the last few years that there’s been an upswing in scientific studies looking at what’s happening inside our brains at the cellular level that might actually produce forgetting.
In fact, I would argue they’re not just irrelevant, but they can be detrimental to living our daily lives.
An interesting case comes from this old study by a Russian clinical neuropsychologist [A.R.] Luria, who had this patient, Patient S., who he reported was able to memorize basically everything from his life. But Patient S. was actually at a disadvantage, because he had trouble identifying the commonalities, the patterns in the world. And those commonalities, those patterns are what actually allow us to make intelligent decisions.
If you see a dog off leash, barking at you, and previously you’ve been bitten by a dog barking at you off leash, you probably want to avoid that dog, even if it doesn’t look exactly like a previous dog you’ve seen. It doesn’t matter that it doesn’t have brown spots or a white tail. What matters is that it has those features that indicate it’s one of those broad categories of aggressive dogs.
Our memories ultimately are there to help us make decisions, to act in the world in an intelligent manner. Even though we all get a big kick out of people who can smash a trivia game, evolution doesn’t care if you can remember who hit the home run in the 1968 World Series. Evolution cares about whether or not you are an individual who’s making appropriate decisions in the environment to maximize your chances of survival. And reproduction, of course.
Actually, a lot of people become more forgetful after having a kid. How does this work into your hypotheses?
I actually love the example of after you’ve had a kid. This is another fascinating piece of forgetting that people have speculated on. And let me just clarify that this is me speculating — I don’t know if there’s scientific evidence for this or not — but if everyone remembered very accurately what childbirth was like, and we all remembered very accurately what having an infant or toddler is like — and I say this as someone who has a toddler — most of us would probably not have more than one kid. So it’s actually probably beneficial for us, in terms of making the decisions evolution wants us to make, to forget those parts of our lives.
Would you go so far as to say forgetting is a function of memory?
Yes, that’s exactly what I’d say, and what we’re trying to argue in this paper. When the goal of memory is to win Jeopardy!, then you should try to remember everything you can. But when the goal of memory is to help you make intelligent decisions in a complex, changing world, then the best memory system will be a memory system that forgets some stuff. So a healthy, properly functioning memory system is one that does engage in some degree of forgetting.