AI Competes With Connection

I came across this excellent Derek Thompson piece documenting the collapse of American happiness since 2020. It reminds me of the challenge I have with the Steven Pinker kind of argument that, by the numbers, everything is getting better. Measurably fewer wars, less poverty, longer lives. And yet, as Thompson shows, people feel measurably worse: lonelier, less trusting, more adrift. Both things are true simultaneously, and perhaps that’s because we can only count what we can see, and the instruments we’ve built can see very little.

This feels parallel to so many conversations in philanthropy, where we keep building better monitoring and evaluation frameworks. They get more sophisticated every year, and in a way, the gap between what the tools show and what the field teams know only widens. Overall numbers improve, but the community isn’t more capable than before. This is not new. Deming had it right when he said the most important things can’t be measured. The revision I would bring is that the most important things can’t be measured without being changed by the measurement.

Which brings me back to AI.

Reed Hastings supposedly once said that Netflix competes with sleep. I think AI competes with connection. Not directly. I don’t mean it’s intentionally attacking connection in favour of a capitalist agenda. It just offers a better-designed substitute that is faster than your closest friend, more patient than your therapist, more available than a colleague, and less judgmental than any of them. At the individual level, this substitution works because you feel heard and supported.

But connection was never just about the individual feeling connected. It is the thing that builds trust between strangers. It is the thing that lets communities hold disagreements. It is the thing that makes collective decisions stick because people say, “We might not fully agree, but I’m in.”

AI produces the feeling of connection at the individual level while, best case, never building connection at the collective level, and worst case, hollowing it out. The individual gets more capable while the collective gets more brittle. Thompson’s data is the macro version of this. The dashboard says the economy is working while the people say something is broken. Best case, the instruments can’t see it because they were never built to. Worst case, no amount of instrumentation can measure what must be felt.

We need more people in the room, not better tools to read the room from outside.

Originally written for LinkedIn on 30 April 2026. View original →