"AI is just a tool," but is it really? (opens in new tab)

The problem here is that a GenAI model isn’t exactly governed by randomness as there is distinct information you can pull out in a knowledge graph (not saying there’s a literal knowledge graph here, just referring to what’s encoded in the weights) by composing the appropriate set of prompts, and as such some information will always be more closely linked versus other unrelated or dissimilar sources. So, this operation is unlike the operation of a slot machine, where you pull a lever and you are bound to find some answer within a set of permutations.

I think GenAI models do have some deterministic properties depending on their training data, but it depends on the model settings, prompts, and how you yourself impose your editorial decisions on the outcomes. In that sense, I don’t think it’s fair to say that the writing and code produced with help from GenAI models aren’t the property of the composer because the writing and code wouldn’t exist in the form that they do if they weren’t prompted and processed by the composer (provided that this material isn’t plagiarized).

The instruction-driven process of producing acts from a GenAI model make it more involved than what this writer is describing as the model itself has no thought, purpose, or intention. So it’s not a service, necessarily, where I’d expect a service to take more off my plate, but in the case of GenAI the composition, editorial, and review process take a lot of time and effort. As such, I feel comfortable referring to GenAI as a tool because there is both a deterministic quality to its output, and the tool wielder exercises considerable control over the output.