Generative AI: Make Me Hate It!

Kamden Baer

July 19, 2024

Modern architecture stands behind older buildings with more character, even though they notably don't belong to any style of architecture, given they were created by ChatGPT and DALL-E.

The point of new knowledge is to say interesting things, right? We go through life and find meaning in the new information we get every day, either through our own experiences or by learning of it vicariously—whether it be through friends or family or the news or social media. Can generative AI create new knowledge? That is something I’ve struggled with since the debut of ChatGPT to the public, and I’m not really sure at the moment it can. Certainly, it knows more than any one person, but when we look at the realm of our pooled knowledge, is AI just dipping into it and hoping that the well never runs out? But this focus on interesting things to say led me to think about what qualifies as something interesting, after all, what may be interesting to one person doesn’t mean it holds interest for another.

I’ve been thinking about this nuance at the level of society, and how interesting things are said, particularly through art. I remember hearing the argument (probably from an Instagram post of a screenshot from Tumblr, if I’m being honest) that modern architecture has become so enraptured with universalism and utilitarianism that it has become devoid of personality and meaning. For example, streetlights have gone from ornate fixtures that set one city apart from any others, to simple metal poles with an LED light shining down from the top. There is a subtle joy in stumbling upon a row of stately Victorian or Colonial homes, while the sprawl of cookie-cutter suburban homes creates some sort of aversion, at least personally. I believe a similar sentiment echoes in the culinary arts, where there is always something new to discover in the kitchen—whether it be your own or that of a restaurant. I pity the bodybuilders and picky eaters who choose to regiment their diet with the same rice and chicken, or whatever else it is they eat day after day. That’s to say that there is interest in the personal, even at the expense of the universal.

The lesson that I’ve leaned into and that I feel can provoke action from this feeling is that we can and should create things that are original and personal enough to make people capable of hating them. I’ve found it’s been a good reminder when a friend dislikes something that I really enjoy or vice versa, but it also speaks to the emotional responses and connections hidden behind the surface of our actions. Both loving or hating a piece of art or any other interesting thing means there was an emotional connection and some sort of challenge to one’s worldview! That’s exciting, even if we don’t like that thing.

I think generative AIs like ChatGPT and Claude detract from this excitement of the novel a little bit, because they compile what is already known and repackage it as new. Even though they are the product of many sources of knowledge, they are only one output, and as such have their own singular tendencies and patterns. In a way, they serve as a regression to the mean, an appeal to everything and nothing. We can see this  synchronization in academic papers , where Andrew Gray, a librarian at the University College of London found an uptick in certain words more common in LLM lexicons than in more human ones. This rise is found based on data of academic papers from 2023 and can be attributed to the public release of ChatGPT at the end of 2022. The prevalence of certain words rose extraordinarily, with “meticulously” growing 137%, “intricate” growing 117%, and “commendable” growing 83% in usage. Given all the acclaim generative AI has as relatively unbiased and universal, removing singularity and personality from it seems like an exceedingly difficult challenge. Do we want AI to say something new, or do we want it to be used commonly across the world? It seems we cannot have both.