AI and Human Potential

AI tools arrived in executive search in 2022 and 2023. I watched how the recruiters I knew used them.

Some used AI to automate outreach, more messages, more CV screening, more templated emails. Volume went up. Quality went down. The candidates on the receiving end noticed.

Others used AI differently: to analyze patterns across hundreds of searches, identify what was causing pipeline degradation, draft outreach that was specific rather than templated. They could do things that weren't feasible manually, not because the AI was smarter, but because it could hold more context and iterate faster than any person working alone.

Same technology. Different question asked of it. Outcomes that were not comparable.

That gap, between AI used as a volume machine and AI used as an intelligence multiplier, is where the entire debate about artificial intelligence and human potential actually lives. Not in the abstractions about replacement. In the specific choices people make about what they ask the technology to do.

The question I care about is larger than my own industry: 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. IMF research published in 2025 found that AI's impact on economic growth in advanced economies is likely to be more than twice that in low-income countries. AI is accelerating labour market polarisation, bifurcating employment between high-skill, high-wage roles and low-skill, low-wage positions, and the gains are flowing to major technology firms before productivity benefits reach most businesses. AI systems are being built primarily in wealthy countries, by relatively homogeneous teams, optimising for the markets where willingness to pay is highest.

IMF research: AI's economic impact in advanced economies is projected to be more than twice that in low-income countries. The gap is structural, not incidental.

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 an Odia-language system, a Swahili-language system, a Bhojpuri 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. A first-generation professional in rural Odisha navigating a government portal, a farmer in Rajasthan asking about crop disease, a migrant worker in Mumbai trying to understand a rental agreement, these are not edge cases for AI. They are the majority use case for the country.

Healthcare is another domain where the distribution question matters enormously. 80% of India's doctors are concentrated in urban areas, and 70% of specialist posts at Community Health Centers, the facilities meant to serve rural populations, are vacant. That shortage is not going to be solved by training more doctors: the timelines are too long, and 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 community health worker with an AI diagnostic aid is not a substitute for a specialist. She is a multiplier for the specialist's reach.

The labor question nobody wants to answer honestly

There is a harder version of the distribution question that the optimistic framing tends to skip past: what happens to the people whose labour is displaced before the new opportunities become accessible?

This is not a hypothetical. India's IT-BPM sector employs 5.4 million people and contributes 7.5% of GDP. AI agents can now handle up to 95% of customer queries without human assistance, and call centre management is projected to see an 80% productivity enhancement. The roles most at risk are concentrated among people for whom the transition to "higher-value AI-enabled work" is not straightforward. The retraining narrative, that displaced workers will learn new skills and move up the value chain, is comforting and largely unsupported by historical evidence at scale.

India's IT-BPM sector: 5.4 million employed, 7.5% of GDP. AI agents now handle up to 95% of customer queries autonomously. The displacement math is not abstract.

The honest answer is that the transition will be painful for some people, and that the pain will not be evenly distributed. The question is whether institutions, governments, companies, civil society, will respond to that pain as it materialises, or whether they will keep pointing to the long-run optimistic case while ignoring the near-term cost.

This is a reason for urgency in getting the distribution question right, not a reason for pessimism about AI overall. But it requires being honest about who bears the cost of transitions that, in aggregate, produce gains.

The fork in the road

None of the optimistic outcomes are 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 specific choices that matter: whether AI systems are built to function in low-bandwidth, low-literacy environments or only in high-connectivity, English-speaking ones. Whether the data used to train systems reflects the full diversity of the populations they will serve or just the populations with the most data to offer. Whether the deployment models price AI tools as premium products or as infrastructure.

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 optimises for the richest markets, extracts labour 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. The decisions made in the next five years, about what gets funded, what gets built, and who gets to use it, will shape the distribution of AI's benefits for decades.

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

The question isn't what AI can do. The question is who it does it for.

Sources: IMF: AI Adoption and Inequality, 2025 · Gulf News: AI Agents and India's BPOs · India Data Map: Doctor-to-Patient Ratio · Knya Med: Doctor Shortage India

See also: AI and India, AI and Education, AI and the Future of Talent Sourcing

Sources

IMF Working Paper: Artificial Intelligence and Economic Growth: A Cross-Country Analysis (2025)

World Bank: World Development Report 2024: The Middle Income Trap

McKinsey Global Institute: The Economic Potential of Generative AI (2023)