AI and Human Potential

The debates about AI tend to cluster around two poles.

One pole: AI is a tool. It augments human capability. It removes drudgery and expands the range of what skilled people can do. Doctors become better doctors. Engineers become better engineers. The productivity frontier expands for everyone.

The other pole: AI is a displacement engine. It replaces human labor in ways that are faster, cheaper, and more scalable than any previous technology. The people whose labor is replaced do not all find better work. The benefits concentrate at the top.

Both of these framings are probably true, for different people, in different time periods, in different economies.

The question I care about is a different one: will AI expand or concentrate opportunity?

The distribution question

Technology has historically had a complicated relationship with opportunity distribution.

The internet created enormous new opportunities — but the benefits were highly concentrated among people with existing access to capital, education, and networks. The person in Kalahandi with a smartphone but no reliable electricity, no English, and no network of technology workers did not benefit from Web 2.0 the way a software engineer in Bangalore did.

The question for AI is whether the pattern will be different this time.

There are reasons to be pessimistic. AI systems are being built primarily in wealthy countries, by relatively homogeneous teams, optimizing for the markets where willingness to pay is highest. The gap between frontier AI and accessible AI is growing faster than it is shrinking.

There are also reasons to be cautiously optimistic.

Where the optimism comes from

Language models, unlike most previous technologies, are inherently multilingual in their potential. The same underlying architecture that powers an English-language system can, in principle, power a Odia-language system. The barrier is data and investment — not fundamental technical limitation.

This matters because language is one of the most significant barriers to economic inclusion. The ability to access information, to communicate with institutions, to participate in markets — these all require language capability. AI has the potential to make this capability more democratically distributed than it has ever been.

Healthcare is another domain where the distribution question matters enormously. The doctor shortage in rural India is not going to be solved by training more doctors — the timelines are too long, the incentives push graduates toward cities. AI-assisted diagnosis, decision support, and patient communication could extend the reach of skilled healthcare in ways that previous technology has not.

The fork in the road

But none of this is inevitable. Technology does not distribute opportunity on its own. It requires intentional choices — by builders, investors, policymakers, and institutions — about who the technology is built for and who gets access.

The version of AI that expands opportunity is the version built for the next billion people, not just the existing billion. It requires investment in infrastructure, in local language data, in deployment models that work without reliable electricity and expensive devices.

The version of AI that concentrates opportunity is the version that optimizes for the richest markets, extracts labor value from everyone else, and leaves the distribution of benefits to happen however it happens.

We are at an early enough stage that the choice is still open.

That is both the burden and the opportunity of this moment.