The writers’ strike is in its fifth month. I support the writers in their efforts to share in the wealth of the film industry, and I understand their concern about being terminated (so to speak) by AI.
But I think they have some things very wrong.
The thrust of the sign in the image above is that AI isn’t human and therefore can’t re-create humanity. Another sign that I saw the other day said something like, “Chat GPT is stupid and can’t write like I can.” But that misses the point. First, from the creative standpoint, there doesn’t look to me to be any reason to assume that we are at the end of the AI road. In fact, I think we are somewhere near the beginning. The question, at least for the foreseeable future, is not “Is AI alive like we are?” It isn’t. But that doesn’t mean it won’t be able to recognize and reproduce and creatively enhance stories of human experience. It is already sounding very human, even if there are definite limitations to what it can produce right now. If we teach it variations on a theme, it will learn those variations, and, by what they have in common, it will learn the theme. If we teach it what we mean by creativity — and that is something that can be personalized, since your idea of what is wonderfully creative may not be the same as mine — then it will have a very good chance to be creative.
But that, too, is not the main point I wish to make here.
I hope that the writers’ negotiators are aware of the larger danger to the entire industry. The studios themselves should be every bit as worried about AI as the writers are. Because with the advent of text to image, and text to video on its way (since video is a series of images), it is going to be increasingly difficult for studios to justify their massive expenditures on production values. This isn’t to say that it’s going to be easy for someone to tell a machine to make a movie for them. In order for that movie to be good, it’s going to have to have all the elements that we enjoy in movies — interesting characters, visual effects, plot twists, climaxes, story structure, good dialog, a combination of the surprising and the expected, and so on. But all of these things are either right now teachable to the language model, or they will be before long. If innovation turns us to something that supersedes large language models, the same will hold true.
Large language models work by examples. If we provide them with good examples, they will be able to create additional examples. And when text to video becomes sufficiently impressive, the studios will face an existential threat.