Post

I’m noticing two philosophical camps when it comes to the question of AI and copyright. Camp 1 anthropomorphizes an LLM, whereas Camp 2 sees an LLM more like a deterministic prediction machine

Camp 1 personifies an LLM, and holds that the use of copyrighted material in training, and subsequent generations is just like humans drawing inspiration from said material. It’s therefore OK to use AI in creating art, since it’s no different from a person being inspired by Tolkien

Camp 2 sees the LLM as a complex, but deterministic mathematical model, and that the existence of copyrighted material in the model means that any generation using that model inherently contains that material.

Its therefore not OK to use AI to create art, since it’s not meaningfully different from using an “advanced autocomplete” that could “suggest” the works of Tolkien

Personally, I fall into Camp 2, at least when it comes to art. These models may be somewhat of a “black box” still, but they’re still deterministic machines.