4 Comments
User's avatar
Kyle Sossi's avatar

Thanks for this review and for pointing people, including me, to this book!

One add is that a tool’s scientific authority is partly social. Trusting observations from microscopy, distinguishing artifacts from phenomena, posed recurring problems over centuries. So maybe the paradigm concept shows up within tool use too.

This seems to be happening around AI model weights and scientific discoveries. How much do other researchers need to inspect the tool itself to accept its results? That will be answered through a social process and might vary substantially across fields.

kalim's avatar

I haven't read the full piece yet, but I saw the TLDR on LinkedIn. Reminded me of this piece by Sayash Kapoor and Arvind Narayanan, titled "Could AI slow science?" In which they highlight that the geocentric model of the universe —despite producing more accurate results—was rendered obsolete after theoretical breakthroughs. It's a fascinating read.

Link: https://www.normaltech.ai/p/could-ai-slow-science

Anyhoo, looking forward to reading this piece over the coming weekend.

Alan Trachtenberg's avatar

Actually, I believe there is a very strong way that innovative toolmaking is encouraged, more motivating, perhaps, than publication and professional prestige. That is the patent system, offering the possibility of significant monetary rewards for new tools and methods, especially as the new tools find new uses through the marketplace.

Conor Griffin's avatar

Thanks Alan - are you mainly thinking of reforms to academic patenting here, as a way to incentivise more tool-making?