As promised, here are my thoughts on the topic. Short version here as a reply, longer version as a blog post later.

As a rule, I almost always disagree with a claim when it includes the words, "always" or "never". Those words are stronger than I like. But with that caveat aside, I generally agree with the proposition. Here's why:

1. Without humans, there is no AI. AI depends on humans for design, engineering, programming, and training. I don't think that will change - especially the part about training data. Maybe an AI can eventually help with its own construction, but when an AI trains on its own output, it decays to worthlessness. No matter how independent AI might seem, it's "standing on the shoulders of" humans. As Chomsky highlighted, LLMs like ChatGPT are basically just recycling and repurposing human accomplishments.
2. Humans give AI its purpose. AIs can do lots of stuff, but left to its own devices, it doesn't. It just sits there until a human gives it a task to do. AIs have no concept of the real world and no idea of what's useful and what's not. Humans will always decide which problems merit AI solutions - at least for the foreseeable future.
3. Human brainpower is amplified by collaboration. At some point, technologies like Neuralink's brain implant will enable human brains to link up directly with each other and with the AIs too. As @o1eh put it, it's conceivable that with the right communications technologies, human brains could be joined into a sort of a biological supercomputer. Even if an AI can outperform a single human, collaborative teams of humans can assist and direct the AI for even better problem solutions.
4. Comparative advantage: Even if AI is better than humans at everything it can do, it still has limits, and economics implies that in an infinite universe - humans can still find ways to enhance the end result.

Having briefly answered the question, let me now branch off and note that the first point is particularly relevant to the Steem blockchain. A long-standing problem with AI is that it depends on human input for training data, but the people who provide the training data don't get compensated. With the availability of author and curation rewards, Steem has the potential to solve this problem.

We have recently seen announcements that ChatGPT is partnering with Reddit for training data and of course Twitter's Grok trains on that platform's own content. Presumably, Meta's Llama is also training on Facebook's data. The corporations that host those centralized platforms are getting compensated for the training data, but not the end users (with the exception of Twitter's paid subscribers).

If the curators here manage to get the content rewards into proportion with the value, there's no reason why Steem can't be a platform that provides AI training data and simultaneously rewards every single contributor - in proportion with their contribution.

And, just for fun (and to illustrate the point), here's ChatGPT's critique of my response for possible incorporation into my blog post on the topic at a later date;-)

Strengths

1. Caveat Acknowledgment:
   • Recognizes the problematic nature of absolute terms like "always" and "never."
2. Structured Argumentation:
   • Clearly organized points supporting the proposition.
3. Use of Sources:
   • References credible sources like Noam Chomsky.
4. Relevance to Steem:
   • Connects the discussion to Steem’s potential for AI training data and rewards.

Areas for Improvement

1. Clarification and Depth:

   • Expand on AI’s evolving autonomy and self-improvement.
   • Provide specific examples of successful human-AI collaborations.

2. Counterarguments:

   • Address areas where AI already outperforms humans.
   • Discuss ethical and social implications of AI outperforming humans.

3. Integration with Training Data Compensation:

   • More tightly integrate the discussion on Steem's compensation model with the main argument.

Conclusion

The reply is solid but can be strengthened with deeper analysis, specific examples, counterarguments, and a more integrated discussion on training data compensation.