Distinguishing Generation from Creation

It’s time to stop referring to “generative” AI as creative. That fact of the matter is that there is nothing creative about Gen AI, and we all know it. The abstract models instantiated in computers don’t produce anything new, don’t genuinely create anything, and in principle never can. The reason they can’t is straightforward.

The idea behind these models is that they have massive amounts of data from which very general inferences might be drawn; this is called “training,” and it’s sometimes bolstered by human feedback which corrects for inappropriate or incorrect inferences drawn due to limitations in the data or problems with the underlying models. Those inferences are expressed as percentages and used to predict and “generate” the next most probable output given some input.

What’s clear from this is that while the output may be epistemically surprising (certain researchers like James Evans talk about this in terms of CS Peirce’s abductive reasoning), the model itself isn’t doing anything more than performing combinatorics. There is information in the sense used by Shannon manipulated by the model and supplemented by humans at the input and during training, which produces the output. Nothing the model does, though, actually generates new information. That’s because, as Raymond Ruyer pointed out in Cybernetics and the Origin of Information, Shannon’s model assumes the existence of information and merely describes the transmission of that information. All computers can do without human intervention is perfectly transmit or transmit in a degraded mode; they can’t create anything wholesale.

So stop calling these in any sense “creative.” They may produce something new and surprising to you or to us, but that’s not the same as genuine creation. The one is epistemically novel while the other is metaphysically so.

P.S. No, we can’t just say that they’re 'practically’ the same. Making that equivocation differentially favors and harms people. Businesses need people to “innovate” yet they also hate when people don’t do what they’re told or aren’t efficient enough for their purposes; they’re simultaneously anthropophilic and -phobic. But the people who reap the rewards and those who get kicked to the curb don’t balance out. Some people’s creative actions are valued while other people’s aren’t, and the ones who’s creativity isn’t valued are getting replaced with these generative models by those who are valued.

We have admit that Gen AI just exacerbates existing social, political issues, and that the only way to address these problems will be by political means.