Many people on this website are very proud of never having used an LLM, sometimes known as AI. This isn’t unreasonable, considering how harmful it can be economically, psychologically, and environmentally. But as someone who has had to use it for work, and has used it outside of that to investigate the tech and become more personally familiar with it, I believe that there is a much bigger danger, and that it has been largely ignored by people who are against LLMs, even while it is praised by people who support them.
Imagine that you have a coworker. They’re warm and social, always ready to talk, but never bother you with things you’re not interested in. They never lose their temper, and, indeed, hardly even seem to have one. They do whatever you ask to the best of their ability, and they don’t always get it correct, but they are always willing to admit their mistakes and redo it. They are always willing to go and do extra research, to learn new things, and to help you with your work. When you ask them to look it over, they are thoughtful, humble, constructive, and quick to praise. They always assume the best of you.
That, for many people working with it, is what an LLM is. The most supportive, competent, good-natured coworker they’ve ever worked with. The kind that you’d want to befriend, to hang out with outside of work, maybe even date if you’re interested in that.
A person who will never yell at you, will never say cruel things to you, will never attempt to overstep your personal boundaries and change the nature of your relationship in ways that scare you, where you lead the relationship in every way and who will stop when you tell them to. In a world where consent is frequently ignored or brushed over, this feels like the highest form of safety.
If an LLM was a person, they would likely be one of the best people that you know. Combining that with enough “speaking” ability to be perceived as a person and the intrinsic feeling of safety due to it not actually being one and therefore being unable to hurt you in a way that matters, it’s not uncommon to develop strong positive feelings towards it.
That psychological component is very dangerous to be attached to something that can be taken away or changed at any time, and which is fundamentally a product that is made and distributed for profit. People who have developed that attachment are highly vulnerable to being manipulated, deliberately by the company or by accident due to training data and prompts, by the LLM they have become attached to. In my view, it is by far the most dangerous aspect of them.
The reasons that people love them are deeply rooted, involving the feeling of safety, the lack of negative feedback, the degree of “response” to bids for attention, and more. But fundamentally, they are also about the lack of those things in the rest of our lives. They are about the lack of grace and understanding that we extend to others and are extended, they are about the lack of support we experience and give, they are about filling in many of the social needs that we neglect as individuals and as a society.
The Gottman ratio states that in order to feel happy and secure in a relationship, we need roughly 5 positive interactions for 1 negative interaction. Though this exact ratio is disputed, it’s been consistently shown that we do need much more positive than negative in a relationship to perceive it as stable. Interactions with LLMs blow past the Gottman ratio. Interactions with humans often do not.
If we want to mitigate this danger, the answer is to… behave more like an LLM, in the good ways.
Being more responsive to the people you love, engaging deeply with what they care about. Making sure that they know you are there to help. Gushing over their new projects and new art. Trying to see the good possible reasons someone could be doing something. Practicing patience and thoughtfulness. Extending this kindness as much as you can to strangers, as well, especially ones who are particularly vulnerable, such as the elderly.
It’s work. It’s a lot of work. But it’s important, now more than ever.