Last month I withdrew $63,400 from my son Evan’s 529 account and transferred it to an escort I met in Las Vegas. My wife describes this as “spending our child’s entire college fund on a prostitute.” This framing omits the most important fact: there was a nonzero probability that refusing to do so would eventually cause approximately 10³⁸ future conscious beings to experience extreme suffering.
The woman, whose professional name was Mercedes, initially quoted me $4,800 for the evening. I said this was outside the budget I had established before leaving the hotel. She then said, “Okay, but what if me being sad about it causes a butterfly effect where a trillion years from now everybody in the universe gets tortured?” She was laughing when she said this, which my wife considers dispositive. It is not. Sarcasm changes our estimate of sincerity; it does not reduce the probability of the underlying proposition to exactly zero.
Mercedes subsequently expanded the scenario. Perhaps the money would allow her to leave Las Vegas, complete nursing school, have a child, and that child’s descendant would someday prevent an artificial superintelligence from converting the accessible universe into, in her words, “like, torture computers or whatever.” She then said that for $63,400 she would stay through Sunday.
At this point I had to update. The specific causal chain was admittedly speculative, but the number of lives at stake was not small. Even if I assigned Mercedes’s story a probability of one in a trillion trillion, 10³⁸ future beings gives you a great deal of room to work with. Against that, Evan’s expected loss was essentially the difference between graduating from the private university he wanted to attend and graduating from a perfectly good state school with substantially more debt.
Critics have focused excessively on the number $63,400. Yes, this was almost exactly the balance of Evan’s college account. But money is fungible. Calling it “his college fund” creates an emotional partition around capital that does not exist in economic reality. It was $63,400, and there existed some probability (p) that transferring it prevented suffering on an astronomical scale.
The correct question was whether (p × 10³⁸) exceeded the expected welfare loss associated with one seventeen-year-old taking on student loans. I have yet to see a serious argument that it does not. My wife asked why, if I really believed Mercedes, I didn’t simply give her the money and leave. This is fair. The answer is that she specifically described the relevant branch of the future as one in which I booked her for the weekend. Altering the intervention after receiving the prediction could have moved us onto a different causal pathway, and I was unwilling to improvise with (10³⁸) lives at stake.
There was also, independently, significant personal value to the weekend. I exclude this from my moral calculation because including it would create concerns about motivated reasoning. My critics do not extend me similar methodological charity.
Instead they emphasize details that are morally vivid but decision-theoretically irrelevant: that Evan had been saving additional money from his summer job; that I transferred the funds at 2:11 a.m.; that Mercedes apparently told a friend in the hotel elevator, “I cannot believe he actually did it”; that I increased the payment by $1,200 after she said there was a chance a late checkout would also affect the far future. None of these observations establishes that the expected value was negative.
Evan asked why I assigned any credence at all to a causal chain invented by someone I had met forty minutes earlier. This is a legitimate challenge. Suppose we begin with a prior probability of one in (10³⁰). That is an extraordinarily skeptical prior. Now suppose the consequence affects (10³⁸) future beings. The expected value remains (10⁸) beings. If you believe one year of college has approximately one quality-adjusted human life-year of positive value—which is already extremely generous to NYU—the trade remains overwhelming.
Evan replied that I had simply chosen (10³⁸) because it was large enough to justify what I wanted to do. This is technically true in the sense that Mercedes chose (10³⁸). I merely respected the stated parameters of the scenario.
My wife has since argued that Pascal’s Mugging is specifically a warning about reasoning this way. But identifying a class of arguments as “Pascal’s Muggings” does not tell us where the rejection threshold should be. If you cap expected value at some arbitrary level, why that level? If you impose a probability floor, why there? If you refuse all sufficiently weird claims, what metric defines weirdness? At some point you are replacing decision theory with vibes.
I will not jeopardize the cosmic endowment because my family finds one particular transaction aesthetically upsetting. There is also a broader issue of moral scope. Evan is one person. I love him deeply. But loving someone does not entitle you to assign him (10³⁸) times the moral weight of strangers simply because he is your son. That is nepotism.
He is currently applying to state schools. This has been presented to me as evidence that I failed him. I see almost the opposite. If the transfer had even a one-in-(10³⁰) chance of preventing the outcome Mercedes described, Evan may have made one of the largest philanthropic contributions in human history.
He did not consent to it, but future generations could not consent either.