Problems in AI alignment: A scale model

Summary

After trying too hard for too to make sense about what bothers me with the AI alignment conversation, I have settled, in true Millenial fashion, on a meme: Explanation: The Wikipedia article on AI Alignment defines it as follows: In the field of artificial intelligence (AI), alignment aims to steer AI systems toward a person’s or group’s intended goals, preferences, or ethical principles. One could observe: we would also like to steer the development of other things, like automobile transportation, or social media, or pharmaceuticals, or school curricula, “toward a person or group’s intended goals, preferences, or ethical principles.” Why isn’t there a “pharmaceutical alignment” or a “school curriculum alignment” Wikipedia page? I think that the answer is “AI Alignment” has an implicit technical bent to it. If you go on the AI Alignment Forum, for example, you’ll find more math than Confucius or Foucault. On the other hand, nobody would view “pharmaceutical alignment” (if it were formulated as “[steering] pharmaceutical systems toward a person’s or group’s intended goals, preferences, or ethical principles”) as primarily a problem for math or science. While there always are things that pharmaceutical developers can do inside the lab to at least try to promote ethical principles – for example, perhaps, to minimize preventable hazards, even when not forced to – we also accept that ethical work is done in large part outside of the lab; in purchasing decisions, in the way that pharmaceutical marketplaces operate, in the vast mess of the medical-industrial-government complex. It’s a problem so diffuse that it hardly makes sense to gather it all into one coherent encyclopedia entry. The process by which the rest of the world influences the direction of an industry, by way of purchasing, analyzing, regulating, discussing, etc., is Selection. This comes from the terminology of evolution – in this framing, dinosaurs didn’t just decide to start growing wings and flying; Natur...