Chat with Anna Riedl - On Drug Addiction in AI, Language Models, and Scaling Academia
TL;DR: I did an interview with Anna Riedl ( riedlanna.com ) where we talk about some of the things that I've written about in the last months. Great fun!
It was a delight talking to Anna! Check out her webpage and keep your eyes open for future videos. Here’s an excerpt from our conversation on different scientific mindsets:
Anna: Another thought that came to me right now - because you mentioned ‘doubt’ being a very central thing in academia. That's something I often think about when I think of mental health problems in academia. Questioning yourself and questioning your ideas seems to be so central to the scientific method - but you always wonder ‘How is it wrong?’. And you always have to anticipate ‘What would another smart person criticize about what you're doing?’. You basically internalize this way of thinking. It seems pretty straightforward that you have a very harsh internal monologue after doing this every day.
Jan: Yeah, that is a super interesting observation. I guess every individual step that you're describing is this ‘Karl Popper’, classic falsificationism, where you try to disprove every hypothesis that you come up with. This is taken to an extreme in the interaction with reviewers, which can be super uncomfortable (speaking from experience). And there you get this cognitive dissonance where you try to model what the ‘adversary’ is going to say to whatever statement you're making.
But research practice… I mean there are other theories of how science is working. One prominent opponent of Popper is Thomas Kuhn who's saying that there are ’scientific paradigms’. Instead of ‘at every individual step question all of your assumptions and trying to break them’ you rather have a set of core theses and then around that a set of auxiliary theses. You try to push those core theses forward and just see how far you can get by assuming them. And whenever you encounter evidence that goes against your construct of theses, you start modifying your auxiliary theses instead of modifying your core theses. I think that’s a lot more congruent with how the human mind wants to work in many cases.
So of course this Kuhnian framework can only work if there are multiple paradigms in parallel interacting with each other - otherwise there's no correction mechanism anymore that tells us at what point we're running in a wrong direction. Also, this way of thinking has been criticized as ‘p-hacking’. ‘P-hacking’ is this practice of tweaking your parameters until you ‘finally’ get your significant result and you can start publishing. Which is terrible for science at large and a big contributing factor to the reproduction crisis in psychology…