401. Why Science is Fundamentally Irrational feat. Michael Strevens
unSILOed with Greg LaBlanc
Explaining Predictions in Machine Learning
The chapter explores the distinction between explanations and predictions in the context of machine learning algorithms. It emphasizes the importance of understanding concepts beyond prediction accuracy and touches on the historical shift towards empirical testing for understanding the world.
What can we learn looking back on the paths of influential thinkers like Popper and Kuhn today? How are the motivations and passions of scientists left behind in the pursuit of scientific progress??
Micheal Strevens is a professor in the Philosophy department of New York University and the author of several books. His latest work is titled, The Knowledge Machine: How Irrationality Created Modern Science.
Michael and Greg discuss the unspoken motivations and aesthetic judgments fueling the progress of science. They explore the delicate balance between rigorous empirical data and the broader intellectual landscape in which it resides, offering insights into the irrational but inherently human elements of scientific inquiry. Michael shares his own experiences and the profound joy found in understanding causal models in a field where explanation often trumps prediction.
*unSILOed Podcast is produced by University FM.*
Episode Quotes:
The rule of science
10:11: I think science has its problems at any point, but it's in reasonably good shape in the sense that there's a sort of an agreed set of rules for playing the game, the iron rule, as I call them. I mean, even something like p-value null hypothesis testing, that has its downside, of course, but it is a rule for doing science. And if you just think of it as a kind of a set of boxes you have to check to have research that you can stick into a journal, then I think it mostly actually does what it's supposed to do. It's possible to game it and for things to go south in certain kinds of situations. But as long as you don't take it too seriously, it's actually telling you something about the intrinsic quality of the data. Basically, really just formal threshold the data has to pass. It's like a legitimate move in chess. Okay, a move can be legitimate and also a really crummy move, and likewise, data can satisfy these rules and still be terrible data.
Michael's motivation to tackle motivation in the world of science
03:46: Perhaps the most important things about modern science were more connected to the psychological or the sociological, to the institutional framework of science, rather than to the kind of thing that, more traditionally, stood out for philosophers of science—stuff to do with the method in a kind of logical, intellectual sense, a reasoning sense. And so it was a kind of switch in my thinking from arguing and logic to questions more subtle and background questions about motivation.
Diverse attitudes in science
11:34: I think there's a huge range of attitudes in science. There are a lot of scientists who it's just their daily job almost, and so they go and do the job and don't spend a lot of time worrying about it otherwise. And then there's some scientists who really feel like they want to make some breakthrough and come up with some revolutionary discovery. All of them have to, as it were, play the same game. And the game works ultimately independently of their personal motivations, simply by generating enough facts with enough systematicity and attention to detail that if there is some problem with the big framework, science will ultimately find it.
How do we optimize research efforts for maximum ROI on the frontier?
41:10: In a world where doing science costs millions and millions of dollars, it's not so easy to just leave it up to the judgment of scientists as a whole. It's a tough problem, but on the whole, I think it's good not to try to pick out just a few projects and funnel everything towards those few projects. I'm afraid I don't have too many good ideas about the alternative, apart from that, except to allow diversity to flourish in one way or another.
Show Links:
Recommended Resources:
- Karl Popper
- Thomas Kuhn
- Newtonian Dynamics
- Quantum Mechanics
- Arthur Eddington
- Albert Einstein
- Isaac Newton
- René Descartes
- Robert Boyle
- Bruno Latour
- Philosophy of Science
- P-Value
- Case Method
Guest Profile:
- Faculty Profile at New York University
- Strevens.org
- PhilPeople.org Profile
- Profile on the Guggenheim Memorial Foundation
His Work:
The AI-powered Podcast Player
