Collateral Has Names

Sam Altman recently suggested that society should tolerate some harm from artificial intelligence in exchange for its benefits. He said it plainly — the way someone speaks when the calculation is already done and the conclusion already settled.

I keep returning to that word: acceptable.

When someone decides harm is acceptable, they've already resolved two things. First, that the benefit outweighs the cost. Second, that they won't be the one paying it. The person who calculates the trade is almost never the collateral. That cost always lives somewhere else — in a job that quietly disappears, in a student who stops reaching for an answer because the answer arrives before the thinking does, in a conversation that used to require effort and now requires nothing at all.

What makes this particular moment different is that the harm is harder to see. When a factory closes, the town knows. When a highway cuts through a neighborhood, the displacement is visible. But when an AI system quietly reshapes how people think, how they create, how they relate to one another — the loss is interior. It doesn't make the news. It doesn't have a displacement statistic. It just slowly changes what it means to be a person who does things, and we wake up one day not quite sure what we lost.

We've been here before. Every era has its version of this speech. The factory that would revolutionize production. The highway that would connect a continent. The platform that would bring everyone together. Each came with a footnote — some people will be displaced, some communities will fracture, some things will be lost that won't return. And we accepted it, because progress always sounds more convincing than the names of the people it steps on.

What unsettles me isn't the trade itself. Every choice has a cost. What unsettles me is how easily the word acceptable slides off the tongue when the speaker isn't in the category of the harmed. How quickly a cost becomes theoretical when it belongs to someone else.

There's an idea older than any technology — that every person carries a worth that can't be reduced to output. Not a utility score. Not a contribution metric. A worth that doesn't fluctuate with the market or the model. The moment we begin speaking about acceptable harm, we've quietly switched to different math. We've started measuring people by what they produce rather than what they are.

I'm not opposed to AI. That's not the argument. The argument is about what we're willing to say out loud, and what it reveals about who we think is expendable.

When harm becomes a line item, it stops feeling like harm. It starts looking like efficiency. And that's the most dangerous moment — not when we cause damage intentionally, but when we stop recognizing it as damage at all.

Who gets to decide what's acceptable? And would the math still hold if we were the ones being counted as cost?