I don't care about AI
I’m sorry. I know I should but I don’t.
I care about AI a little. I started using Claude Code to do my job for me right around the same time as everyone else, so I care about Claude Code, which lets me make rapid progress on things I care about, which don’t include AI. But I don’t care about how Claude Code works.1 I don’t care about foundation models. I don’t care about evals. I don’t care about data centers. And I really don’t care about P(doom).
I know I really should because I still cop to being an effective altruist and these days effective altruism is all about AI. My friends who used to work on animal welfare now work on Animals x AI, or they work on “model welfare”. My friends who used to work on biosecurity now work on Bio x AI. The organization that initially convinced me that I should quit medical school and go work in biosecurity is these days very open about only caring about AI.2
So maybe I am not an effective altruist anymore, because today that means caring about AI. And I don’t.
I first encountered LessWrong 15 years ago. I’ve read all the arguments. Some of them are very good. I don’t have arguments nearly so good. I have no well-grounded basis on which to disagree with assertions that AI might kill everyone, and therefore I should care about it very much. I can only gesture to a vague feeling that I get, encapsulated by this “proof” that 1 = -1:
I’m not terrible at math, but when I first saw this “proof” I couldn’t identify the error. I sent it to my dad, a much more seasoned mathematics understander than I am. He also couldn’t see the error, so he sent it to his friend and colleague, a stats professor at UNC Chapel Hill. The stats professor couldn’t see the error either; he’d specialized in the wrong kind of math, so he sent it to his friend, an abstract algebraist. The abstract algebraist looked at it and could identify the error. It’s on line 5. I forget his exact explanation; I probably forgot it right away, because I don’t need to know that to know that 1 ≠ -1.
Perhaps you, reader, could see the error right away. Skill issue, you are thinking. Of course it is. But I don’t need to have that skill to know that 1 ≠ -1.
Perhaps you are objecting. AI doom is much more complicated than this stupid 8 line “proof.” No arguments about AI maybe killing everyone are anywhere near so obviously wrong as an assertion that 1 = -1. I’m being vastly overconfident. The comparison is not appropriate.
That’s all true. I anticipate your arguments and they are all correct. I still don’t care though. Sorry.
Back in 2022 (what some call EA’s summer of love and what I call EA’s deranged cokehead era) I was doing a PhD at an Ivy League university, on the theory that I could spend a few years doing fun science stuff (that, to be entirely honest, I much preferred doing to medical school anyway) and eventually I would be well-posed to do Very Serious Biosecurity Policy Work. I didn’t really know anything about Policy but I was sure I’d eventually figure it out and it would be very Impactful. Hell, it had to be better than medical school.
By the way, if you want to sneak into an Ivy League university, I strongly recommend doing it as a graduate student. Much less competitive than undergrad and you still get the brand name.
And you definitely want the brand name. Cokehead EA very much cared about brand names. This is why in 2022 the Centre for Effective Altruism deployed some earnest young people to my Ivy League university with the intention of setting up an EA student group there. As part of this effort these earnest young people enlisted me to help run a structured reading group about topics in effective altruism.
This made me feel very important and good. Just a few short years ago I was desperately looking for ways to Get Involved and now I was to lead a reading group!
The readings were about all sorts of things, but the earnest young people made it clear that the early readings about global poverty and animals were really just there to get buy-in for the good stuff—longtermism, existential risk, AI. Milk before meat. Really the whole purpose of the university EA group was to comb through the susceptible population of a well-resourced group of influenceable young people, and convince them to devote their lives to doing something about AI.
At some point in the semester, I attend a weekend retreat with this group of earnest young people. We play a game called Hot Seat.3 One of the earnest young people is in the hot seat and is asked a variant of the drowning child question: if three newborn babies, and one effective altruist, were both drowning in separate ponds, which one would you choose to save?
The earnest young person answers obviously you save the EA. I mean, by expected value the EA is worth so much more. Because the EA might do something about AI, while the babies are just some random babies. They probably won’t even have time to grow up and learn enough about machine learning in time to do anything about AI. No brainer.
This weekend my writing residency has shipped us off to a mansion up the coast. I am to understand this mansion once housed CFAR, the Center for Applied Rationality. CFAR. See-far. Get it?
CFAR was supposed to teach people rationality techniques, but actually that’s not what it was supposed to do. It was supposed to convince people to abandon their lives and devote them to doing something about AI.
A lot of people in my writing residency are writing about AI. I try to read my fellow residents’ posts, but I have not read a single post about AI this month. I’m sorry. I’m sure your posts are good. I just don’t care about them.
That’s okay though. I have no proof or direct evidence to speak of, but I have a suspicion that this residency mostly exists to get more people to write about AI. Because the more people are successfully blogging about AI, the more likely that someone might read an unusually good post about AI, and then decide to devote their lives to doing something about AI.
The Bay Area is very beautiful. I’m writing this from a gorgeous rocky beach on the Sonoma Coast. In California the weather is always nice. The sun shines brighter here. There are always parties happening where I can always find someone to have an interesting conversation with. Sometimes even about something other than AI. Here is the only place where the kind of committed polyamorous household that I long for is something like a norm. One of my best friends lives here. We founded a startup together. I probably should live here too. Why don’t I live here?
San Francisco is maybe the last city in America that looks to the future. The other day I got to try cultivated salmon. How marvelous! Not very good, but marvelous. One of my dinner companions casually mentions that he’s essentially retired, because by the time he runs out of money, the AI probably will have killed everyone.
I could never live here. There is just something so unhandsome about it all.
This is not the first such essay to be published by someone approximately in my position. I know the beat. At this point the author, who is not directly themselves doing something about AI, who finds so much of the culture and epistemics around AI discourse to be toxic, admits that they care very much about AI. Because it’s so important. It’s the most important thing. We need to clean up our act re: AI. We can’t afford to fuck this up. It’s the most important thing in the entire universe and we have to Get It Right.
Well, I won’t do this maneuver. I won’t admit how much I care, because I don’t. I don’t care how important it is. I won’t and you can’t make me.
Technically speaking I am an effective altruist in good standing. I work on a technology that kills airborne viruses and bacteria. The company I cofounded is trying to make this technology cheaper and more accessible.
Effective altruism is interested in this technology. Not because it prevents the prosaic sort of suffering from airborne infections disease that happens every day. Nor even because it might prevent a “normal” pandemic in the future. Effective altruism is interested in technologies that prevent airborne disease because the AI might try to use airborne disease to kill everyone.
So by devoting my work to this technology, really, I am doing something about AI. Perhaps this makes me punked. Totally tricked. Owned, you might say.
Fine, then. Sure. Whatever.
I still believe in effective altruism. I feel the same way about it as my fellow resident does. I believe in the true and the good and the beautiful. Maybe that’s why I refuse to care about AI. Whatever in EA is pointed at the true and the good and the beautiful, it isn’t there. I don’t know what in AI is good and true and beautiful. Maybe nothing. Maybe that’s why EA can’t seem to properly metabolize it.
I sometimes think effective altruists are born, not made. When I first heard of EA, I went, oh yes, of course, exactly this! This was what was already in my heart. I’ve come home. I guess that’s what they mean about not stepping in the same river twice.
A persistent source of tension between my husband and I is that he is a basically normal, good person and I have something wrong with me. He wants to be a good dad and son and upstanding member of the community, and I want to be the Main Character of the Universe and do the very best job at the most important thing. The more flattering way to put it is that I need to be pointed at the good and the true and the beautiful; that I feel called to serve God. Sometimes it’s hard to be married to someone who doesn’t need that, in the same way that it might be hard to be married to someone with no sense of smell.
Other times, I recall the last time I was in the Bay Area. I attended a party at the very same campus hosting my writing residency, where I spoke to a young man who seemed profoundly lost. He was on three different drugs simultaneously, each of which was made up of letters and numbers and not words. “I’m not sure if I should be optimizing for utils or hedons,” he told me. Utils meant trying to do something about AI. Hedons meant just doing as many drugs as possible before eventually being killed by the AI.
After I’m done with this residency, my husband and I are moving to a small town in Maryland near our parents to have a baby. We’ll be there for a couple months and then decamp to DC. I hear there is a community of EA parents there, who I hope to trick into being my friends. They probably care a lot about AI. But that’s okay.
I hope that young man finds his way to the good and the true and the beautiful. And I hope that I do too.
I know how it works, vaguely—my abortive PhD was in machine learning applications—and I don’t care to hear any more details, I just want the tool to be good so I can use it to not think about AI.
It used to be they would refer young engineers who cared about pandemic prevention to me for advice, and now they don’t, because pandemic prevention is only relevant to this organization anymore because an AI might try to kill everyone with a pandemic. Or maybe I was just doing a bad job advising young engineers. Both things are probably true. I’m mostly relieved, because my own career path has been so idiosyncratic and path-dependent that I don’t really have any good advice for replicating my “success.”
Hot Seat is like Truth or Dare, or Never Have I Ever, except without the boring parts. One person is selected to sit in the “hot seat” and everyone asks them questions and they have to answer.
Very glad someone put this into words, much better than I could've. I think of Dorothea's plight in Middlemarch -- EAs have a longer history than people think, it goes back to Mother Theresa, back to the early Christians. There will always be people who fervidly pursue the good and the beautiful, and it's sad to see doom descending on them.
in my understanding the error in the proof is that square roots have two results: a positive and negative one. i/1 is the positive one, and 1/i is the negative one. So line 4 is incorrect (equating the square roots).