8 Comments
User's avatar
Pamir Sevincel's avatar

Great take!

Here is a question: do you think ML models will ever replace scientific theory, especially those areas of science where the human brain may be inherently limited to explore? In other words, instead of descriptive equations showing how dark energy works, could ML take its place without the need for us to explain how dark energy exactly works, especially if we can make accurate predictions using that ML model about dark energy?

Expand full comment
Christoph Molnar's avatar

Great question!

Replace theory -> No, at least not completely in the near future. But it many areas we might acknowledge more and more that phenomenons can't be explained by interpretable equations (like a linear model).

I see science as becoming more accepting that due to interactions and non-linearities we won't be able to "understand" all relations. But machine learning models can often approximate such relations really well and become part of the science toolkit.

Expand full comment
Pamir Sevincel's avatar

Christoph - thank you for your thoughtful response.

Would you say that there is a resistance in the scientific community for Replace Theory? Also, do you think that the physics of the cosmos is statistical in itself or determined by laws with perfect mechanistic relations?

Expand full comment
Christoph Molnar's avatar

I'm not familiar with state of the art of modeling physics of the cosmos, so can't say.

About resistance: when people argue not to use ML because "it's a black box" and "it's not interpretable", then I would see that as a resistance against ML, so yes you can observe that. But mostly I think it's a tendency to sticking to the scientific methods that have always been used in a field. change always happens slowly

Expand full comment
Pamir Sevincel's avatar

It will be exciting to watch for sure.

Looking forward to your book and future newsletters!

Expand full comment
Carl Hutton's avatar

Predictive models teach us a lot about the system we are observing. My fair value bitcoin model does a poor job of predicting price, but it teaches us a lot about the human emotions behind market prices.

Expand full comment
Rainer Lange's avatar

Great project! I think one of the central questions is, what are scientific theories for? Is science "just" about prediction, or is it also about explanation, and how are these related? And how is "explaining the ML model" related to "explaining the world" (ie, what science is supposed to strive for)? Cf. the discussion by Henk de Regt on eplanation, and the use Mario Krenn et al make of this concerning the role of AI in science (https://doi.org/10.1038/s42254-022-00518-3).

Expand full comment
Christoph Molnar's avatar

That paper wasn't on my radar. That's a perfect reference, thanks a lot for sharing Rainer! Will dive right in. If you have more references like this, don't hold back 😉

Expand full comment