every conversation about ai right now is exhausting, and i understand why. it has become a referendum on everything at once, jobs, art, truth, power, the shape of the next decade, all of it loaded onto one word until the word can barely hold a sentence. there is real fear in it, some of it earned, and there is hype on the other side that is just as tiring, the breathless kind that promises everything and respects nothing. i do not want to stand on either. the loud optimism is dishonest and the loud despair is its own kind of comfort, a way of being done with the question. neither is really paying attention.
but underneath the noise there is something i privately believe and have not stopped believing, even on the tired days, and it asks to be weighed rather than declared, so let me try to weigh it.
put the costs on one side. some are about misuse, the ordinary list, deception, dependence, the concentration of power, and those are real but at least partly correctable, because they are about how the thing is governed and used rather than about what it is. some are subtler, and i take them more seriously, the ones the rest of my own thinking keeps circling, what it does to attention, to honesty, to the work of making our own judgments. those are not small, and i do not want to file them under nothing. they go on the scale at full weight.
now the other side. a great many of the hardest problems we have, especially in medicine and the sciences, were never bottlenecked by how much people cared or tried. they were bottlenecked by bandwidth, by search spaces too large for human patience and patterns buried in more data than any mind can hold at once. cancer is not one problem, it is thousands, each a different argument between a cell and its own broken instructions, and understanding even one has taken decades of people. a tool that is genuinely good at finding faint structure in enormous data is aimed straight at that bottleneck. it will not cure anything alone, it is not magic, and the people who speak about it like magic should be ignored on principle. but shortening the years between a question and an answer, letting one researcher try a thousand ideas in the time that used to buy ten, is not a toy and it is not a threat.
so the weighing. it comes out lopsided, though not for the reason the hype gives. the upside is not certain, it is a chance. but it is a chance at something enormous, years taken off the suffering of a very large number of people, and a modest probability of an enormous good can outweigh a high probability of a bounded harm. you do not need the optimism to be guaranteed. you only need the magnitude to be real and the chance to be more than zero, and both of those i think are true. discount it hard for everything that could fail, and the sum still does not come out even.
i am not naive about how it goes wrong, and the worry is not foolish. it is rational, and it is cheaper, which is part of why there is so much of it. worry asks less of you than hope does. but the imbalance itself, the way nearly all the energy goes to imagining the harm and hardly any to sitting with how much better things could get, is a distortion, and it subtly tilts the scale before anyone has finished weighing.
so when i have spent a day buried in the discourse and finally step out of it, what i am left with is not a forecast. it is a question i have decided is the right one to hold. if there is even a real chance that somewhere, in a lab i will never see, the thing that was going to take another twenty years takes five instead, and someone keeps a person they were going to lose, what exactly are we setting against that, and are we sure it weighs as much as we have let ourselves believe?