Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
The writer is an economist and the former chair of the Financial Services Authority
AI experts hail the potential for super-rapid prosperity growth. Sir Demis Hassabis, founder of DeepMind, foresees “an era of radical abundance, a kind of golden era”. Technology investor Marc Andreessen believes “absolutely stratospheric” productivity growth will make worries about inequality irrelevant. Anthropic’s economics team reports a median expectation of 8 per cent per annum GDP growth by 2030.
But the astounding technological advance of AI does not mean it necessarily accelerates growth in measured GDP or human welfare, and it is almost certain to exacerbate concerns about inequality.
As societies get richer, an increasing share of economic activity is devoted to zero-sum competition, with gains for some coming at the cost of losses for others. Ever more sophisticated AI-enabled cyber attacks, for example, require ever-increasing cyber defence expenditures with no benefit to end consumers. In legal services, AI will automate some activities but could unleash a dramatic increase in litigation; “agentic flooding” is already overwhelming the British state and legal system with complaints and injunctions generated by AI.
Conversely, AI will deliver huge benefits that come almost for free. Multiple scientific breakthroughs promise accelerated progress in drug discovery, perhaps eventually giving us drugs that ensure 100 years of disease-free life. In the short term, GDP will be swollen by research expenditures, manufacturing costs and pharma company profits. But once the drugs are off patent and manufactured in AI-enabled factories, profits, wages and the contribution to GDP will collapse.
At the limit, if anything is made infinitely abundant by AI, its measured economic value becomes infinitely small.
That in no way diminishes the human welfare benefit. But tax revenues to cover interest on accumulated debt are paid from nominal GDP, and AI contributions to human welfare will not mitigate political disputes about who pays for public expenditure.
Great benefits, moreover, will be balanced by AI-induced harms. Setting aside the risk of human extinction, deepfakes and AI-enabled manipulation are driving an explosion of online scams and increasing the threat to adolescent mental health. This harm, like many AI benefits, will be invisible in measured GDP.
As for inequality, AI seems likely to deliver increasing returns to high skill — whether of scientists, business executives or lawyers — but poses a threat to workers of average skill. The optimistic response is that many goods and services will be so cheap that wage stagnation will not matter.
That optimism is misplaced. As societies get richer and technological progress reduces the real cost of many goods and services, individuals devote a rising share of income to buying things inherently in short supply. The most important of these is housing in desirable locations, driving the increases in house price-to-income ratios observed across the world. AI will not increase the abundance of attractive locations for housing.
In a world of ever-increasing intelligence, an essentially weightless and location-independent attribute, it is inevitable that the most physical and location-specific thing of all — land — will increase in relative price.
Concerns about inequality are therefore more likely to increase than decline. No amount of cheaper computer games, entertainment or manufactured goods will offset the sense of restricted opportunity felt by average earners unable to afford housing in London, Paris, Seoul or Shanghai.
AI can both increase human welfare and degrade it. But the balance will be determined by the policy choices societies make. And many of the most important effects — for good or ill — will never show up in GDP.