We debated calling this episode "An ode to Michael," because we set out to do an AMA but only get through his first two questions. But never fear, there are only 20 questions, so at this rate we should be done the AMA by the end of 2024. Who said we weren't fans of longtermism?
Questions:
- Hey do you guys have a Patreon page or anyway to support you?
- (Michael) Not clear that humans are universal explainers. Standard argument for this is "to assume o.w. is to appeal to the supernatural," but this argument is weak b/c it does not explain why humans could in principle explain everything. But all Deutch's ideas rests on this axiom. It's almost tautological - there could be things humans cannot explain, but we wouldn't even know about these things b/c we wouldn't be able to explain them. I think this argument that humans are universal explainers and thus can achieve indefinite progress needs more rigor.It might be a step jump from animals to humans, but why could there not be more step jumps in intelligence beyond human intelligence that we do not even know about? I'd love to get your thoughts on this.
- (Michael) Another pt I'd love to get your perspectives on is the idea of the "creative program." Standard discussion is "humans are special because we are creative, and we don't know what the creative program is." But we need to make progress on creativity at some point and it kind of feels like we are using the word "creativity" as a vague suitcase word to encapsulate "everything we don't yet know about intelligence." Simply saying "humans are creative" without properly defining what it means to be creative in a way that we can evaluate in machines is not helping us make progress on developing creative AI. It's unsatisfying to hear critiques of AI that say "this AI model is not 'truly intelligent' because it is not creative" without also proposing a way to evaluate its creativity. In this sense, critiques of AI that say AI is "not creative" are bad explanations because these critiques are easy to vary. Without a proposing a proper test for creativity that can actually evaluated, it is not possible for us to conduct a test to refute the critique. I'd love to get your thoughts on how we can construct evaluations for creativity in a way that enables us to make scientific progress on understanding the creative algorithm!
References:
Contact us
Support
You can support the project on Patreon (monthly donations, https://www.patreon.com/Increments) or Ko-fi (one time donation, https://ko-fi.com/increments). Thank you!
How much explaining could a universal explainer explain if a universal explainer could explain explaining? Tell us at incrementspodcast@gmail.com.
Support Increments
