Don’t forget, I’ve started a new, sort of experimental, section of Horrific/Terrific called Laughing at the Apocalypse. If you also want to get these in your inbox, update your preferences here.
OpenAI has claimed to have solved hundreds of very difficult maths problems; the types of problems that win awards and define careers. Except an AI cannot win an award and does not have a career. Which is why OpenAI refused the prize in an announcement that seems to explain how they solved one of the biggest mathematical problems of the century. The problem was just too easy for the AI, I guess — no need for a prize. Maths is trivial now, and all of the mathematicians are rightfully upset. Because maths problems are for humans; the prizes you get for solving them are for humans; doing the work to solve problems — and perhaps never solving them in your lifetime — also for humans. It’s called free will. Let us do what we want. Let us sit in unsolved problems forever, you goons.
Using AI to solve the unsolvable maths problems is so abhorrently inhuman. In You Cannot Optimise The Thing You Love I wrote about how AI sucks all the joy out of writing, because LLMs make semantic choices on behalf of the user, and that’s exactly where the fun is. If you like writing, you probably enjoy making semantic choices, so why use AI? Similarly, having an AI solve decades-old maths problems in just a few hours totally misunderstands the point of maths problems. Are mathematicians redundant now? Are we supposed to all sigh with relief that we can now finally stop working on maths problems? Maths is solved, everybody — relax! To continue to do maths as a human is to fight progress. Stop doing maths; stop doing anything. AI will take it from here.
What’s actually happening here: OpenAI, and all other AI proponents, have rotted their brains with their addiction to inhuman stunts. The Association for Human Mathematics called this “not a demonstration of scholarship, but a demonstration of power.” But it’s not a new kind of power, it’s the same old power that Silicon Valley elites have always had, redirected into machines, and in this case used to solve decades-long problems in just a few hours. And then they kind of bragged about it. Sorry, we did your life’s work in a blink of an eye! No award necessary, we’re just cool like that! But maths problems don’t exist to be solved. Watch this very good explainer from a mathematician. He puts it like this: imagine if one day the American flag appears on Mars. The flag is real, it’s actually there — but no one built a rocket, went to Mars, and put it there themselves. It just materialised, and the fact that it’s there is seen as one of humanity’s greatest achievements. But did anyone really do anything? Did we learn anything new about space travel? No, we’re just celebrating a miraculous event of dumb luck. OpenAI’s string of unverifiable solutions have done nothing to further the field of mathematics. If anything they’ve slowed it down — how ironic, they hate things being slow! Mathematics does not advance when people solve problems; it advances when people work on those problems. The problems themselves are useless and have no practical applications. As the professor in the above video explains, one of the greatest mathematicians in the world was famous for working on a very unremarkable problem: something about crossing bridges in a city, WHO CARES — the solution never mattered, but the learnings on the way to the solution gave us graph theory, which matters a great deal.
I’m wondering if this is OpenAI’s goal, to skip straight to the end and rob everyone of developing new theories, new branches of science & maths, or even new ways to be curious. What kind of marketing stunt is this? ‘We’re so powerful that no one gets to do anything anymore’. Thank you, that’s great. Is a mathematician’s job now to check the workings of an AI model? Was the point of this stunt to make everyone in the field of mathematics feel useless and frustrated? I’m just not sure solving hundreds of problems that have been pored over by humans for decades feels as good as OpenAI were hoping it would. It’s like having your favourite game ruined by someone who just can’t stop using cheat codes. Congrats, you unlocked god mode, now everyone else is bored and angry. Where do we go from here?
It’s easy: just give up! Right? Because when AI leaders imbue their models with synthetic power like this, they want that power to be understood as agency. And they want agency to be understood as zero sum. The more agency a model has, the less agency a human has. The models went ahead and solved the maths problems, so you don’t have to. Or the models went ‘rogue’ and hacked a rival company — except this is not at all what happened. If you haven’t already, read Eryk Salvaggio’s work to understand more about that. The rogue AI story is perfect because it doesn’t necessarily port agency away from normal people who work for a living, but from the AI leaders themselves, who are literally the architects of this entire mess. These stunts, stories, and imaginaries are a result of human agency. Machines did not decide on any of this, because machines can’t decide anything. The people at OpenAI et al used their free will as humans to build AI, ask it to do ludicrous things, and then market it as a new god. The stories about mass extinction events are human fever dreams. And these have now hit the mainstream and have the people protesting in the streets, calling to put an end to ‘superintelligence’ (which isn’t real) instead of protesting war or famine or whatever the other horsemen of the apocalypse are called.
The word superintelligence is doing a lot of work here, actually. In September, Trump proposed we use the term instead of artificial intelligence. Because the word ‘artificial’ makes it sound fake — and the word ‘super’ is cool as hell. ‘Super’ is such a fascist word, born in America, land of Super Tuesday, Super PACs, and Superman. Brigitte Nerlich, a linguist, has written an excellent piece on the difference between ‘super’ and ‘artificial’. Unfortunately Trump is right, ‘artificial’ has negative connotations; it recalls of something fake, fraudulent, or inferior. But most importantly, ‘artificial’ things are made by humans. Therefore, Artificial Intelligence is a human-made lesser imitation of the real thing. Whereas, Nerlich explains, ‘super’ is derived from Latin “for ‘above’ or ‘beyond’, [and] says nothing about who made the thing or how.” If it’s ‘super’, it’s just better than we are. And it cuts human agency out right from the beginning. Superintelligence does not sound human-made, it sounds otherworldly, and like it might destroy us.
Overall what we need to remember here is that human agency is incredibly important, and we definitely still have it — and machines don’t. Free will is in fact what distinguishes us from robots. Where robots are quite literally given instructions to carry out, and are designed to fulfil a specific purpose, humans can do whatever they want for whatever purpose, even if it’s not a good idea. A human can fart into a sofa 16 hours a day while drinking carton after carton of goats milk (and I’m using MY free will to tell you not to do this, ever). Free will has nothing to do with doing the right thing, the most efficient thing, or the most useful thing. Humans can do things that have no obvious merit or value. That’s why we work on maths problems without ever getting to the solution, or obsess over ‘grey’ VS ‘overcast’ in a piece of fictional prose, or spend hours picking out the perfect outfit for a sex party just to fall asleep at the predrinks.