[GaggiX]

A human can be less or more creative, there is always some creativity, but for a machine it is controversial to say if there is any creativity at all. Going from not very creative to very creative seems to be a much easier task than creating a machine with some creativity to begin with.

[freehorse]

I think there is creativity as a process (which is not what I think the machines do) and creativity in the output as a perception/judgement on a work. What I meant is that I see machines as emulating human creativity, as in producing something that has some of the characteristics of work that is usually seen as creative. The quality of creativity, in this sense, is to be seen in the output, not the process. AI is trained on human (creative) output, thus results of it can seem "creative" because of that. AI does not transform multisensory inputs and lacks any sort of experience or coupling with the world. It cannot be really creative not (just) because we have not figured the right algorithm, but also because it is not transformative enough. But the emulation itself can produce outputs where humans will see as containing creativity.

Maybe the conclusion is that we do not really need creativity for some tasks, what we need is certain aesthetic and other rules that the AI can learn through is training. If that makes a more or less boring future where everything is the same and nothing really changes is left to be seen.

[GaggiX]

I disagree, I don't see how a model that is trained with multimodality (input and output) and that can explore the environment will acquire creativity, I can see being more creative, but that suddenly these specific characteristics should make the model creative instead of emulating seems odd. I also disagree with "not transformative enough", they are very transformative.

[freehorse]

I never said that a model "trained with multimodality" will acquire creativity just because of that.

How transformative they are is probably a subjective judgement depending on what one means. For example, the way I see it, when we do poetry, we transform our experience into words. What speaks to people through poetry is not the words themselves, but the pointers to the connected experience they contain. Poetry is not about writing a nice poem by predicting the next token based on statistical correlations. Current AI is not transformative enough because it does not transform lived experience into words, it predicts tokens based on what tokens it was trained on.