Skip to main content

Tom Griffiths - Five Way Interview

Tom Griffiths introduces himself: I’m a cognitive scientist — a professor of psychology and computer science at Princeton University — so I think about minds and mathematics every day. But we are in an interesting moment where people who aren’t professional cognitive scientists are grappling with the same questions: Can machines think? Is it possible to describe minds using mathematics? What are the limits of different approaches to building a mind? Will we be able to create super-human artificial intelligence? These are questions that have come into focus in the last few years with the creation of chatbots that can hold conversations and solve challenging problems, but answering the questions we have about modern AI requires going further back into the past. In writing the book, I hoped to give readers the context for this moment and some of the language for talking about it, as well as highlighting the stories of discovery that brought us to this point and that suggest possible paths forward. 

Tom's new book is The Laws of Thought.

Why does cognition interest you?

I got interested in understanding human cognition because it remains one of the genuine mysteries left to science. I was excited to discover as an undergraduate that people had used math to understand the mind, and I’ve been on that path since then. 

Why this book particularly?

I say in the acknowledgments section in the book that it is my love letter to cognitive science, and that’s absolutely true. I have enjoyed learning all the things in the book and wanted to share them with others. The way I learn things is through stories, which give more depth and texture to ideas, and so that’s the way that I put each idea in context. It helps to understand an idea when you know something about the problem that somebody was trying to solve, why solving that problem was difficult, and what other things they tried along the way. 

Back in 2016 you said you were interested in human strategies for solving decision-making problems. Does AI help or hinder this?

AI helps us understand the strategies people use for making decisions, and also suggests some ways to better support human decision-makers. Even though people are often derided as bad decision-makers, in my lab we explore the hypothesis that people actually do pretty well when you take into account the cognitive constraints that they operate under. Expressing and solving that problem is actually much harder to do than using traditional ideas about rationality, and we use sophisticated ideas from the AI literature to do so. 

Thinking about human errors in terms of resource constraints also suggests a way to help human decision-makers by providing them with extra resources, and we have explored ways to use AI to add computation into the environments where people make decisions to improve the outcomes. For readers who want to dive into the technical details, I just had another book come out with my co-authors that explores these ideas called The Rational Use of Cognitive Resources. 

What’s next?

I’m still thinking about possible directions for my next book, but want to dig deeper into the ways in which human minds and AI are different from one another. In the meantime I’m putting out the interviews I conducted for The Laws of Thought — plus lots more — as a podcast on the history of cognitive science called The Cognition Project. 

What’s exciting you at the moment?

The creation of intelligent machines is a pretty exciting moment for cognitive scientists, as for a long time we’ve only had one kind of system that could use language and answer complicated questions. This creates all sorts of opportunities to use the tools we have developed for studying human cognition to study AI systems, as well as a chance to better understand people by comparison. I’m also excited about the ways in which AI can expand the scope of our science — letting us develop theories in domains that previously resisted formalization — and accelerate discovery by helping us automate the creation of experiments and theories.

Photo © Sameer Khan

These articles will always be free - but if you'd like to support my online work, consider buying a virtual coffee or taking out a membership:


Labels:

Comments

Post a Comment

Popular posts from this blog

Meet the Neighbors - Steven Benner *****

I love popular science books, but it's a quiet sort of love: it's not often a book like this gets me really excited - but Meet the Neighbours (sorry, I know it's Neighbors, but it grates) does just that.  Once upon a time there was quite a market in books speculating about life on Mars, but of late most Martian titles have focused on getting to and colonising Mars (though often avoiding the 'c' word). Steven Benner drags us back to see the evidence for life having existed there, and does so with a vengeance. (Almost literally - he really lays into those who are dismissive of what evidence there is.) Do we have proof? Well, no - but proofs are for mathematicians. As Benner says 'science does not deal in proof... Rather, it relies on a 'preponderance of evidence' standard to shift the burden of proof.' Benner points out the weird reality that since the Viking landers, which arrived in the 1970s, there hasn't been a mission looking for organic sig...
Post a Comment
Read more

Data Empire - Roopika Risam ****

The central thesis presented by Roopika Risam is that information gives us (and particularly countries) the power to organise, control and dominate others. Although I have a couple of issues with the presentation, this is a genuinely interesting trip through the history of our use of stored information from the earliest tallies to the latest information technology. I loved a quote from Lisa Gitelman that data is is always 'cooked' so 'raw data is an oxymoron'. This neatly underlines Risam's thesis that data and information are not neutral facts, but rather tools that (like everything from fire to electronics) can be used for good or evil. As we are taken through the historical context, it can sometimes be a little difficult to judge whether Risam regards a particular example as bad or good, even when the outcome is disastrous. One thing I didn't like too much is the adherence to a popular science writing approach that has got distinctly hackneyed: opening chapte...
Post a Comment
Read more

The Selfish Gene (50th anniversary edition) - Richard Dawkins ****

Remarkably for those who came across it in the early days, this classic of popular science writing is now 50 years old - and like many of us who reach a certain age it has become somewhat overweight: a reasonable sized paperback has morphed into a 500-page monster. Even so, we have to recognise the importance of the book, despite its message not being as clear as it once was. One major change is that Dawkins, with his God Delusion hat on, was once the hero of the provisional wing of the progressive movement, where now he has been pushed aside for daring to suggest that there are only two sexes. The other change, of course, is that the science has moved on and some of the genetics certainties of Dawkins' day have reached the scientific stage of 'it's more complicated than we thought.' In a couple of chapters that weren't in the original book and about 120 pages of discursive end notes and an epilogue, Dawkins attempts to clarify some scientific developments and to ...
Post a Comment
Read more