A case for AI taboos – Writings and rehearsals by Nathan Schneider

Skip to content

A case for AI taboos

We probably all know the person, and maybe it is ourselves, who has inundated themselves in AI use, and say they are fully aware of all the dangers, but are so, so careful as to not fall into such traps themselves.

They know AI can incline us to inhibit thinking, poison our relationships, and evade responsibility. They know it is a crutch and a running joke, and even a war crime. But that is not how they use it. No, never. Except, from the outside, everyone else can see: they are not so different from the more ignorant and susceptible people they imagine themselves aloof from.

Part of the function of culture is to protect people from what they cannot protect themselves from alone. We fall into temptations before we know it. Sure, most times you run next to the pool, or have whiskey for breakfast, or eat unwashed food, or get your phone out at the dinner table, or wear maximally revealing clothes, you’ll come out okay. But each comes with risk, and in some cases the risk involves self-delusion.

Cultures draw arbitrary lines around those risks, which colonial ethnographers came to call taboos. They often seem excessive or absurd. But they are excessive precisely because they are realistic about the frailty of human judgment.

A pivotal moment in the acclimation of culture to AI will be the development of AI taboos.

Perhaps the most relevant set of historical taboos to AI are those related to the sin of idolatry—the sin of treating as divine the things that humans ourselves have made. In the Abrahamic traditions, idols distract people from worshipping God and from doing justice to each other. The threat is so grave, and so tempting, that the taboos far exceed the bounds of the sin. No statues allowed, whether you worship them or not. Mosques typically prohibit humanlike depictions of any kind.

There are costs to exceeding the bounds of the sin. But there are also epiphenomenal benefits. The prohibition on graven images has produced an extraordinary tradition of Arabic calligraphy. Prohibitions on certain foods or attire produce in-group solidarity. In any case, taboos are something that humans always do. They arise without formal legislation or decree, whether we ask for them or not, although formal institutions often adopt them. And it often seems like the people who like them least are those who need them most.

I don’t think we entirely know what taboos are needed in societies that include talking, writing, reasoning, self-organizing machines. But the cringey breakdowns of self-control can be a guide. So is the phenomenon of “slop.” Whatever is obviously cringe and slop is a starting point for discerning what might be a helpful taboo.

Allow me to suggest a few preliminary candidate taboos. I’m not sure I’d stand behind them in the end, but I offer them as contributions toward the greater project of cultural wising-up. I’m not the biggest fan, for instance, of human supremacy over machines; like many taboos, these proposals have a whiff of superiority to them. But perhaps that is a necessary form of self-defense.

Share not your raw outputs.

A young software developer recently told me about having asked a higher-up at work for feedback; they got back a response pasted from and attributed to a chatbot. This might feel clever to someone established in their career, for whom AI is a novelty. But it is infuriating for a young person looking for guidance and mentorship—who is perfectly capable of asking a chatbot, and in this case did beforehand, but now was seeking human judgment.

Draw a line. Don’t share raw outputs. The temptation to being inordinately impressed with the results is too great, and you’ll probably insult the other person in the process. Before communicating an idea to someone, you have to explain it on your own terms, in your own words. Even if you are working on AI-assisted code or art, you have to explain what you were doing and why.

Ask not for attention you are not willing to give.

This is an extension of the first. But sometimes the prohibition of raw outputs is not enough. People send each other hybrid slop—material that is partly their own but also bot-augmented, so as to require far more attention to interpret than it took to create.

It is better to send someone a short postcard or a few bullet points than to have a bot unravel it into something far larger, more time-consuming, and intention-refracting. There is simply far too much content in the world right now for us to be able to sustainably tolerate disproportionate expectations on our attention. If you ask for more attention than you give, you shouldn’t be taken seriously.

Name not your bots with personlike names.

Sure, maybe you think you can distinguish your chatbot from a person. Maybe you can’t. But kids and elders (and everyone in between) are definitely developing relationships with these things, and that...