The Value of Scarcity in the Age of A[I]bundance
In a time of accelerating AI development, I think we should take a step back and understand why and for what purposes we want to use generative models.
Throughout this post, I will distinguish between two different generations of generative models. The first, and current, generation is good at imitating human behavior, but struggles with out-of-distribution generalization, which is necessary for actual intelligence and creativity. The second generation, which is probably still very much out of reach, will close this gap to us humans.
There is no doubt both generations will provide opportunities to build things, many of which create great value. The possibilities are vast, and I will not focus on obvious upsides in this post. Instead, I want to argue that the creative potential, even of second generation models, is smaller than one might think.
Humans, probably for evolutionary reasons, enjoy things that are scarce. Take the Harry Potter books, for example. Why are they so beloved? Sure, they tell a great story. Every child would love to go to Hogwarts. But another reason is that, by the time you encounter the books, you have already read many other, mediocre ones. That is because good stories are scarce!
This principle extends beyond Harry Potter. As I write this post, OpenAI just released its its 4o Image Generation model. People love it and use it to draw pictures in the style of Studio Ghibli. Why? Because Studio Ghibli creates beautiful, hand-drawn movies. Anybody watching them is aware of the amount of work and dedication going into every single frame. Few other studios create comparable pieces of art. Once again, it’s (at least partially) about scarcity! Second generation generative models would eventually eliminate this scarcity. Once that happens, how will our appreciation of the final result change?
I also believe that, at least subconsciously, we humans value the creative effort another person invests into making a piece of art. We relate to the artist and appreciate the fact that we couldn't have created the piece ourselves. In contrast, the prospect of an AI generated movie feels hollow, cold. After all, for many, a movie is more than just a sequence of images.
Similarly, blog posts aren’t just strings of words. They hold value in the fact that somebody took the time to gather their thoughts, sat down and decided to share them with others.
This makes me skeptical that even second generation generative AI will create a lot of creative value. Even if it reaches human capabilities, its output will be missing the human component - something that gives tremendous value to the creations we connect with.
To summarize, I see two main inhibitors for generative models in creative fields. First, humans value scarcity, and cheap generative AI might eventually lead to abundance. Second, we appreciate each other's creations, and second-generation models would diminish the human influence.
Amid all the buzz, we’ll need a bit of time to collectively reflect on the role we want generative AI to play in creative tasks. Regardless, we can't - and probably shouldn't - slow down progress anyways. So sit back and enjoy the ride!
Edit: if you want to read more about scarcity, AI, and art, I greatly recommend Scott Alexander’s The Colors Of Her Coat or Erik Hoel‘s Welcome to the semantic apocalypse. Give these posts a read, they are amazing!