[kyleplum]

When someone uses Photoshop to create artwork, nobody believes that the Photoshop algorithm has any claim to be the author of that artwork. I don't see how the code behind e.g. an AI image generator is fundamentally different than the Photoshop algorithm from a "who created the produced work" perspective. Similar to Photoshop, the AI art generator is still just running code at the behest of some human.

The more interesting question I think is whether any of the creators of the training data used by the AI have any claim over the generated artwork.

[shon]

Yes that is the more interesting question and the answer is most probably yes, especially as these systems are trained on billions/trillions of samples from the net. But, due to the opaqueness of how these models really work, it might be very hard or impossible to prove a claim like that, unless the resulting work is an obvious copy of prior art.

[0thgen]

if the models are too opaque to give appropriately weighted credit to all parties involved in the dataset the model learned from, would that make it reasonable to just push all AI generated inventions to the public domain?

(at least in regards to black box models)

[mherdeg]

> The more interesting question I think is whether any of the creators of the training data used by the AI have any claim over the generated artwork.

Feels like we covered this back when we decided whether sampling music creates a derivative work.

[Kye]

There are absolutely people who think the computer does all the work with digital art. They might even be a plurality. That kind of feedback is a common complaint from digital artists. Especially when it comes to commissions. "How can you charge so much? The computer does [most, all] of the work!"

[960design]

Only if your third grade math teacher has claim to your algebraic code comparisons. Current trends are iterative training with positive first model, negative second model and let the AI 'scale' the learning from there.

[CharlesW]

> I don't see how the code behind e.g. an AI image generator is fundamentally different than the Photoshop algorithm from a "who created the produced work" perspective.

The difference is that Photoshop doesn’t create 2D works any more than Blender creates 3D works. You may be ascribing too much magic to Photoshop.

[Dylan16807]

In terms of filling up a canvas, I'd say content aware fill is in the same ballpark, if you add some google image search to help it.