Generative AI: Surprisingly Unsurprising

We have reached the stage of generative AI where artists, authors, and performers are convinced that AI is stealing their lives — and, perhaps more importantly, their livelihoods. This dystopian theme is showing up in lawsuits, strikes, and, of course, Black Mirror episodes.

Yet, as far as I can tell, generative AI models do not try to copy individuals. Perhaps the developers of generative AI models are concerned with the legal or societal implications, but I suspect that the real reason is far more mundane: relying on an individual source is simply too risky.

Many people have criticized the output of ChatGPT and its rivals as relentlessly boring. I hope that generative AI developers see this as a compliment. After all, it is far better for their output to be boring than exciting — and wildly wrong. Indeed, generative AI receives the most attention when its output looks like the result of an bad acid trip — or, as we say in the AI biz, “hallucination”.

When we seek information, boring is good. We do not expect textbooks, encyclopedia entries, or weather forecasts to be works of art. We do not focus on their individual expressions of style. Most importantly, we do not expect the information itself to be unique. Indeed, if the response to an informational query veers too far from the consensus, that is usually a reason not to trust it.

Generative AI models work hard to avoid surprising us. Even when they mimic a specific artist’s style, they do so through an abstractive synthesis that waters down the style, reducing it to its lower common denominator rather than expressing the boldness of any individual work from that artist.

By trying hard to avoid surprising us, however, generative AI models are unlikely to ever delight or inspire us. Playing it safe has its benefits, but it is a recipe for respectability at best, and mediocrity at worst.

That is not to say that generative AI models could not aim for individuality. But that would be a risky strategy, and it would produce a high failure rate. Indeed, people who veer too far from the safe middle take the same risk. Being an outlier often —perhaps usually — means being a crank or a crackpot. Most of us choose to play it safe.

Yet a tiny minority of outliers turn out to be the people who change the world. Brilliant scholars. Genius artists. Unique performers. People who buck the consensus despite the risks and bring the world along for the ride.

I do not expect generative AI models to supplant this kind of creativity, They will raise the bar for the rest of — but that is what automation has always done. Perhaps the most surprising outcome of all of this existential angst about AI is that the output of generative AI is not that surprising.

So, if you want to resist our new robot overlords, then be different. Take a chance and do something outside the consensus. Seek out authors, artists, and performs who push boundaries, and take delight in their bold, individual works. Surprise yourself, and surprise the world.