#AcademicRunPlaylist - 5/27/25

It really warmed up today, and while I was mostly inside I was able to get out for a bit and listen to books and talks for my #AcademicRunPlaylist!

First was an intriguing talk by Blake Richards on large-scale brain decoding at the Kempner Institute at Harvard University. I'm concerned about the ethics of some of the cross-species future work mentioned here, but this approach has promise https://www.youtube.com/watch?v=pGzJmaSaofs

Next was "Open Innovation Results" by Henry Chesbrough. While the concept has somewhat fallen out of favor (and this is addressed in the book), open innovation remains a potent organizational tool to drive long term success. Chesbrough reviews the economic literature to illustrate the necessity of improving innovation to reverse declining economic growth trends, and while I'm skeptical that open innovation specifically will move the needle here, the process changes necessary to support it will likely have positive effects, as this book shows. There are a wide range of case studies here, and they add to the survey and experimental results that are examined as well. Sometimes the book loses focus, with a fairly random chapter on China standing out here. Overall, however, this is an excellent book for academics and practitioners alike. Highly recommend https://academic.oup.com/book/32327

Last was "Noise" by Daniel Kahneman, Olivier Sibony, and Cass Sunstein. This book presents the best and worst of behavioral economics. At its best, it exposes the flaws in human decision-making and reasoning and the implications for those flaws across a wide variety of areas. At its worst, it falls into the trap of failing to interrogate the nature of accuracy, data labels, and systems that cause outcomes. This emerges from the shockingly uninformed usage of many thoroughly debunked metrics as supposedly “unbiased,” “objective” performance metrics - intelligence tests and bail data particularly stand out.

The time spent trying to unconvincingly distinguish what they call “noise” from bias could have been devoted to addressing ethical issues that come with scaling up a single metric of performance, issues of distribution shift, etc. From the extremely limited coverage of the actual issues in AI systems, however, I'd be unsurprised if they were mostly unaware of these issues. If you’re already familiar with these problems this book can provide some helpful cases and framings, otherwise reading this book will probably paradoxically increase your misunderstanding of these fundamental issues https://readnoise.com/