We are reaching peak hype cycle for LLM as a result of some comments made by the NVIDIA CEO during his keynote speech and I just want to make a friendly reminder to everyone that AI cannot (yet) create like a human being, and as a result cannot yet program, and doing so would require rethinking how we approach AI altogether. Let me explain.
Programming is not strictly an engineering exercise but a creative one. It’s not a matter of time invested, intelligence or effort to whether or not AI will replace programmers. It’s a fundamental limitation to how AI is currently organized in relation to its users. All AI in its current form can do is synthesize and recombine existing ideas created by humans. It requires our ideas to function, it feeds on them like nectar, without which it would stagnate.
Think of AI in its current organization like a parrot or a DJ, not a musician. It does not create new ideas on its own because the creative process doesn’t come from just mimicking other artists’ ideas. The creative process is rooted in an embodied agent reflecting and subverting their relationship to their environment which results in reputational risk and possible societal reward. In other words, we are in the real world, we transcribe whats actually happening here into forms like text, which is a profoundly lossy down sampling of information when you think about all the things we do on a daily basis.
I will sample the cover of the cover of the cover of the cover of "Imagine"
AI doesn’t really have reputational risk, it doesn’t have embodiment and it doesn’t have a relationship to its local environment because its environment is the entire social sphere of those who interact with it according to how its RLHF feedback is specified. Until those things exist AI can’t really create, and its responses will be a muted average or pastiche of societies collective sentiments, not the reflection of an independent agent striving to orient themselves in relationship to its environment.
Because of this IF everyone stops creating and depends on AI to create, like the scaremongers and doomsayers are saying, then its tap of ideas which it can plagiarize is cut off and all human intellectual endeavor is stunted. AI and ourselves both die a slow death of lack of ingenuity.
Can AI become embodied and achieve agency? Sure! But it’s a much different conversation if we are talking about the neighbor’s android spewing misinformation on the internet than if we are talking about ChatGPT doing so. A much more human conversation.
