Tuesday Variety Pack

Hot links for the cool crowd.

Mark Newgarden, Imagine

Well, I guess we’re all still here…

1. David Deutsch on Lulie Tanett’s new podcast Reason is Fun. Deutsch talks about why he’s not worried about AGI, and why current AI isn’t on the path of AGI and is, in fact, basically the opposite of AGI. This interview with Deutsch by Naval Ravikant on the Tim Ferriss show is a better introduction to his overall worldview, but I like Lulie’s lighter, less formal, approach.

2. Peli Grietzer “Patterns of the Lifeworld” in Aeon Magazine. Grietzer continues to stake out a plausible claim to being the first great literary theorist of the machine learning era. This one is a bit harder going than “A Theory of Vibe” but continues to mine a rich vein that runs through art, philosophy, and AI. “At my worst,” Grietzer says, “I believe that poetry directly gets at the material bedrock of our games of sense – gets at the real affective, cognitive and physical dynamics that give stakes to sense and nonsense. […] poetry can’t be explained using manifest-image terms because it is a kind of backstage tour of the manifest image, giving us a glimpse into the constitution of these terms themselves.”

3. This lecture by Gregory Chaitin at Carnegie-Mellon from 2000 is a fantastic overview of how the most abstruse philosophical questions about the foundations of mathematics led to computers and therefore all of… this. It culminates in a beautiful image of the pure incompressibility of certain mathematical entities, which cannot be explained as the result of any underlying pattern and therefore exist as a kind of event horizon at the edge of meaning. Chaitin’s palpable delight in leaning, dangerously, over the lip of this precipice is infectious.

4. Alex Tabarrok suggests putting an AI NPC in a game and seeing if it takes over the world. Zvi Mowshowitz thinks this is a good idea. I’m not so sure, for the same reasons that I’ve always been skeptical about studying fiscal policy or epidemiology in World of Warcraft. Or rates of recidivism in Grand Theft Auto, if you see what I mean. Games, even simulation games, are not simulations. They work by crafting connections, shaping motivations, establishing patterns, constructing characters, and inviting players to push and pull at the specific dynamics of the resulting microworld. Games are not a neutral vessel that will reveal your true character. I thought we already litigated this at the final tribal council of every season of Survivor?

5. g, a Statistical Myth. This Cosma Shalizi post from 2007(!) kinda blew my mind. I had always assumed that IQ skepticism was mostly motivated by aversion to the sketchy political issues attached to the subject but that some version of some kind of “general intelligence factor” existed and was commonsensically obvious. This deep dive into the statistical methods that underlie the concept convinced me otherwise. Two things make it incredibly relevant to contemporary questions about AI. First, it bears on the issue of whether or not a singular, all-purpose, scalar “problem-solving capacity” exists. But even better, the argument itself is a demonstration of the kind of statistical reasoning that modern machine learning type AIs do (and/or fail to do). What does it mean, mathematically, to extrapolate from data to explanatory model? Shalizi shows us, and it strikes me as very important that he does so from the perspective a committed practitioner of statistics and a teacher who cares about the norms and ethical commitments of his field.

6. My book is coming out! Holy cow. I wrote a book! I’m pretty excited about this, can you tell? I just got through fixing typos on a pdf and now it’s a real book with a coming soon page and you can even pre-order it on Amazon. Yes, I would love to appear on your podcast!