Delegating the Self
On mocotó, ChatGPT, and why you can't outsource an opinion.
A friend once told me about this intern who asked a language model to write an essay about how he (he being the intern) was feeling. If you feed a reasonable prompt into any language model, it will come up with something that sounds articulate and coherent. If you ask the same question of a human being, the odds are against you. The human mind can be messy. And it's in that mess that we find who we are.
Language models are seductive because they're so good at presenting information in a well-structured way. They give you the impression of value even when they're in no position to answer the question. When the question is about you, there is no external authority. The answer doesn't exist until you make it. So why do we delegate our thoughts and feelings to silicon?
Often we don't know for sure how we feel or what we think about something, and not knowing is uncomfortable. So we hand the discomfort to something that will always produce an answer. A few months ago, a client of mine asked ChatGPT whether his project was "beautiful," because he didn't know whether he found it beautiful. He wanted certainty faster than he could produce it. So do the rest of us. Until we put them into words, our thoughts and feelings can be a disorderly, jumbled, half-formed mass. The thing is, we form a point of view in the process of grappling with that mess, finding the themes that repeat and putting it in some order.
Let's take taste as an example. Someone asks you how you feel about a Brazilian dish, like mocotó (a stew made from cow's feet). Maybe it's the first time you've tried it. You don't know how you feel about it. It's different from what you're used to. Kind of weird. Then you give it another try. It starts to grow on you. The taste of the beans, the meat of the sausage, the thickness the collagen gives the broth. You… kind of like it. This second time, your answer is a little different from the first. A week later, you're arguing about how much chopped parsley belongs on top.
Whether it's a feeling or a preference, the answer is constructed, not retrieved. Knowing ourselves takes time. It's OK to use LLMs as tough sparring partners for stress-testing your ideas. You just have to remember who's giving the final answer.
They can't do this for you because the process is the product. You don't actually want the answer. You want to have become someone who has that answer.