Who gets the benefit

Suppose it goes well. Suppose the scientific promise is real and a genuine breakthrough arrives.

A breakthrough does not distribute itself.

AlphaFold is the useful case here precisely because it went about as well as these things go. The predictions were released free in a public database covering more than 200 million protein sequences. Any researcher with an internet connection can use them. That is a deliberate decision, it was not the only option available, and it made the benefit unusually wide.

Now follow the chain further, to where the good actually lands. A predicted structure helps a researcher, who needs a laboratory, which needs funding, to develop a candidate, which needs trials, which need regulators, and manufacturing, and distribution, and a health system that can pay. Every link is a place where access narrows.

So a medicine that reaches people is the output of that entire chain, and the AI improved one early link in it. Improving one link speeds things up. It does not widen the ones further down, and the later links are where most of the narrowing has always happened.

The same pattern shows up closer to home. A tool that makes skilled professionals more productive tends to benefit people who already have skills, positions, and the time to learn it. That can compress the gap between a beginner and an expert, which is genuinely levelling. It can also widen the gap between people who have access and people who do not.

Which way it goes is not determined by the technology. It depends on price, on who is trained to use it, on what infrastructure exists, on whether the useful version is free or a subscription, and on decisions made by people who could decide differently.

The honest position is that capability creates possibility. Distribution is a separate problem, and it is a political and economic one rather than a technical one.