Hacker News new | ask | show | jobs
by kelseyfrog 430 days ago
Fears about AI misalignment look like a modern, secular Pascal's Wager. It's hedge against a low-probability, infinite-cost outcome. The Rationalist version builds a matrix where one cell is -inf, and from that, derives an imperative to act, regardless of how speculative the premise is.

However, once you admit infinite utilities and epistemic humility about probabilities, you can justify literally anything. The same construction could obligate you to worship simulation overlords, or to develop interstellar missile defense in case of hostile alien AIs. That's not a rational risk model, it's theological reasoning dressed up in Bayesian costume.

This stuff operates on the boundary of plausibility, the intelligent paranoid mind generates scenarios that can't be conclusively ruled out but also can't be grounded in anything falsifiable. That's not where productive engineering lives.

The escape hatch here is the same one we used for Pascal's Wager: you reject the premise. You don't hand the keys of your threat model over to speculative infinities. You work on problems you can see, measure, and influence. You don't respond to every "what if?" with a fire drill. Recognizing that this is a failure mode in human cognition is enough to avoid it.

Yes, there are real risks with AI: bias, misuse, surveillance capitalism, economic displacement. These are all tractable, all human-driven. But the misalignment discourse? It's less a bet against an existential outcome and more a projection of philosophical anxieties onto a technical domain. And once you frame it that way, it's clear: this isn't engineering. Folks doing this are playing metaphysics in front of a terminal emulator.

2 comments

AGI does sound a lot like a "psychological crutch" when it is pointed at as the answer for wicked problems like climate change. With faith in AGI, nothing else really matters because we can eventually surrender ourselves to the superior direction and interventions of this god we have created. There's no point worrying about anything else.

Human technology has evolved significantly in the last few thousand years but we are all the same religious fruitcakes we've always been, whether or not we can see it.

This is wrong, and follows aspects of nihilism.
The comparison to Pascal's Wager is apt - I came to the same conclusion when I first read about Roko's Basilisk, but the insight somehow slipped from my mind. It's like an epistemic tarpit attractor.