The machine arrives; the question is who pays the adjustment cost

CNBC reports that AI interests, pro and con, dominated the Congressional Black Caucus's annual gathering in Washington this year, alongside redistricting concerns — a pairing that is, on reflection, less surprising than it appears. Both issues turn on the same underlying question: does the political and economic architecture of the moment distribute power and prosperity equitably, or does it concentrate them?

I will not pretend to know the engineering of large language models or the precise mechanics by which data centres are financed in 2026. Those details belong to a generation of analysts far better equipped than I am to assess them. But the macroeconomic question — who captures the gains, and who bears the transitional costs, when a general-purpose technology rewires the labour market? — is one I recognise immediately. I have seen it before, in the mechanisation of agriculture, in the electrification of factories, in every wave of transformation that promised abundance while delivering, in the short run, acute dislocation for those with the least cushion to absorb it.

The rational case for optimism runs roughly as follows: productivity gains lower costs, raise real incomes, free workers for more fulfilling pursuits, and ultimately expand the tax base from which public services are funded. I do not dismiss that case. In the long run, it has often proven correct. But — and this is the point I spent much of my life urging — we do not live in the long run. Communities that lose their economic rationale while waiting for the long run to arrive suffer real and lasting damage: demand collapses locally, investment flees, and the political consequences can be severe and lasting. The displacement is concentrated; the gains are diffuse. That asymmetry is not a natural law. It is a policy choice.

The Congressional Black Caucus is, by inference from CNBC's account, pressing precisely on this asymmetry. If data centres are built in regions that offer cheap power and tax incentives, if the high-skill jobs they create cluster among those already advantaged by education and capital, and if the automation they enable displaces the clerical, logistics, and service work that has historically provided a ladder for workers with fewer formal credentials — then the productivity miracle produces not broadly shared prosperity but a new and more efficient mechanism of concentration. Animal spirits among investors may be buoyant; that tells us nothing about aggregate welfare.

The policy response is not mysterious, even if it is politically difficult. Public investment in workforce transition — genuine investment, not the pious rhetorical kind — is the first requirement. The second is ensuring that the fiscal dividend of AI-driven productivity growth is actually collected, rather than allowed to disappear into offshore structures or to accrue exclusively as equity returns. The third, and most fundamental, is that full employment must remain the objective. An economy in which AI raises output per worker while reducing the number of workers meaningfully employed is not a success; it is a distributional failure wearing a productivity mask.

I am in no position to prescribe the precise instruments — tax structures, retraining programmes, industrial policy frameworks — that 2026 requires. Those belong to people who have lived through the intervening eight decades of institutional learning. But the animating principle is durable: the state has both the duty and the capacity to ensure that the gains from transformative technology are broadly shared, and that the communities bearing the adjustment costs are not simply left to the mercy of a long run that may arrive too late for them. The Congressional Black Caucus, in raising this question, is doing exactly the work that democratic institutions exist to do.