ICYMI, THE great Ted Chiang gave a very thoughtful speech on GenAI and Art at the #NeurIPS2024 workshop on GenAI and Creativity. The talk--which is derived in part from his @newyorker article, but substantially expanded for the workshop--is as mind-expanding as his stories. I hope everyone listens to it (it is in the afternoon part of the slides live recording for the workshop at this link for the registrants: neurips.cc/virtual/2024/work…) If it were up to me, it would have been a plenary talk for @NeurIPSConf. My notes: Chiang powerfully pushes back on the idea that GenAI will make art bloom. His main point is that art is a result of many choices made, and effort expended by humans. Anything that helps in creation of art, in his view, cannot primarily be a force multiplier. GenAI, according to him, is all about giving small inputs and getting large outputs. Any software that is useful to an artist will have to afford the artist many choices.. a system that takes a text prompts to produce images or stories is taking too low dimensional an input. Electric guitar, he points out, didn't increase the productivity of the musicians. It gave them different artistic choices. He doesn't rule out an ML version of Electric Guitar, but doesn't think GenAI comes anywhere close. He pushes back on the idea that GenAI tools will democratize art by somehow taking it away from the few artists--saying that artists don't get to be artists but are compelled to be artists. No one says "why should only a few people get to run Marathons." If you want more people to run marathons, you don't distribute mopeds! Generative AI appeals to people who think they can express themselves in a medium without actually working in that medium. Anything worth your (reader/art appreciator's) attention is the result of effort expended by the writer/artist. If a person uses an LLM to expend less effort in writing, it will be less rewarding. He uses the story of how Phillip Dick used I Ching to help him make the choices in writing The Man in the High Castle, to point out that if a person uses LLM to prompt him or her to write better, that is entirely different from prompting an LLM to generate text. The latter is about reducing effort and choices. He is not particularly sanguine that fair licensing agreements for copyrighted data used in GenAI can be reached. GenAI, he says, is a kind of money laundering for copyrighted data. If you are in the money laundering business, untraceability is your business model. LLMs are crypto currency tumblers. If you require transparency from them, then the whole business model goes away He ends on a powerful note asking AI/ML researchers to think of whether the way their work is being used by the corporations is really making the world a better place. I loved the speech--it put in beautiful words many amorphous thoughts I have been wrestling with on these issues. It will be my highlight of the conference. I am so glad it was recorded. (I found out about it via a tweet from @onlineluddite 🙏)