Selection happens many times prior to publication. Of all the ideas generated, surely most do not even approach publication, because they are judged too badly by the person who thought of them. Either initially, or during the formalization process that turns ideas into philosophy
Yeah, it's definitely a part of the creative process, but it seems sufficiently intertwined with other parts of the creative process to consider it together with them, separately from external selection. At least that's how I was thinking of this.
(In the biology analogy, there's also selection in the form of DNA repair mechanisms to remove mutations. But natural selection doesn't refer to that.)
Creativity implies novelty (in multiple sense of the term). Current-gen LLMs are not capable of such, but if new architectures are employed wherein an LLM is just one piece of a truly “agentic” whole, we may see some interesting results.
I'm not sure I agree; if I tell a model to write a poem in the style of some obscure author who never wrote poems, and the model makes a reasonable response, surely that's novel in a meaningful sense? Doesn't mean it's creative as such — we probably agree that the creativity was in me telling the model to do this. But I don't think novelty per se requires agency, even if creativity perhaps does.
Well done, a thoroughly modern "modest proposal." Best wishes for your entry in the essay contest, I trust the judgments will be Swift.
Selection happens many times prior to publication. Of all the ideas generated, surely most do not even approach publication, because they are judged too badly by the person who thought of them. Either initially, or during the formalization process that turns ideas into philosophy
Yeah, it's definitely a part of the creative process, but it seems sufficiently intertwined with other parts of the creative process to consider it together with them, separately from external selection. At least that's how I was thinking of this.
(In the biology analogy, there's also selection in the form of DNA repair mechanisms to remove mutations. But natural selection doesn't refer to that.)
Maybe you need 100 billions AI that each do a simple task, but it is the combinaison that lead to complexity.
Creativity implies novelty (in multiple sense of the term). Current-gen LLMs are not capable of such, but if new architectures are employed wherein an LLM is just one piece of a truly “agentic” whole, we may see some interesting results.
Almost certainly not before 2026, however.
I'm not sure I agree; if I tell a model to write a poem in the style of some obscure author who never wrote poems, and the model makes a reasonable response, surely that's novel in a meaningful sense? Doesn't mean it's creative as such — we probably agree that the creativity was in me telling the model to do this. But I don't think novelty per se requires agency, even if creativity perhaps does.
The novelty necessary for true creativity is a novelty of perspective; that is not possible with current model architectures.
"Analogy" is analogous to "exaption," ie. when an adaption is repurposed. Footnote1 thus wrong.