Digital Economy Dispatch #300 -- Who Decides the Future of AI?

Will AI destroy civilisation, or save the planet? Will it make us cleverer or hollow us out? Does it end work, or merely rearrange it? What becomes of being human when the machines get good enough? The argument about AI's future has never been louder, and it has rarely been less useful. Nobody knows the answers, and the confidence with which so many are delivering them is usually a decent guide to how much speculation is being positioned as a revealed truth.

Leaving that aside, what has caught my attention in recent weeks is a small group of serious voices who have stopped asking what AI will do to us and started asking who is deciding. Three of them make that case from very different attitudes and altitudes. Reading them together has changed how I look at the future of AI.

Who Sets the Pace

Let’s start with one of the most vocal AI commentators, Cory Doctorow. He is both brilliant and infuriating. His writing can be insightful one moment and crass the next. He makes me laugh out loud, then yell in despair. Reading his latest book on how to survive AI, I threw it onto the sofa in disgust several times, only to pick it up again in eager anticipation of what came next. What can I say?

I can say this: The Reverse Centaur's Guide to Life After AI is going onto my reading list for senior leaders anyway. Not because I agree with it. Because it goes after the assumptions sitting beneath almost every AI strategy document I read, and it does so at the only level most managers can actually act on.

Doctorow's title rests on a piece of automation theory he has made his own. A centaur is a person assisted by a machine: human on top, horsepower underneath, whether that means a hearing aid or a car. A reverse centaur inverts the arrangement. The machine sets the pace and the human becomes its peripheral, feeding it, checking it, keeping up with it.

The example he highlights is radiology. In the centaur version, a radiologist works alongside an AI system, and the pair produce more accurate analysis than either would individually. The result is better outcomes. But the radiologist is still there and still paid. In the reverse centaur version, the system does the work and the surviving humans are demoted to checking its output at volume. A much cheaper and less demanding role for the human. Also, predictably, more error-prone, because a human asked to rubber-stamp a machine's work at speed is a human who will eventually stop looking properly.

Notice what is doing the work here. The technology is identical in both scenarios. The capability is the same. What differs is the operating model wrapped around it, and the operating model is a management choice we control, not a technical inevitability. Doctorow argues that the business case pushes relentlessly towards the second version, because that is where the savings are, even though the rhetoric for much of AI deployment is the first one.

This observation reframes the entire adoption debate. We spend enormous energy arguing about capability. Is the model good enough yet? Has it crossed some threshold? Doctorow suggests capability is close to beside the point. The same system, performing identically, produces opposite outcomes depending on whether the professional using it retains the expertise, responsibility, and authority to say no.

His more provocative claim follows. The industry's most valuable product, he contends, is not any model at all. It is a story told to investors: that human labour is about to become optional, so buy the machine and demote the worker to its minder. Whether the system can genuinely do the job matters far less than whether the boss can be persuaded that it can. Dorian Lynskey, reviewing the book for The Guardian, pointed out that Doctorow is not animated by existential risk, deepfakes or AI psychosis, all of which he treats as side effects. His quarrel is with the revenue model and the investment capital that sustains it.

Why Belief is Not Enough

Carissa Véliz makes a related argument at greater depth in Prophecy: Prediction, Power, and the Fight for the Future, from Ancient Oracles to AI, longlisted for the Financial Times Business Book of the Year. Her thesis is surprisingly simple and considerably more direct than Doctorow's. She maintains that prediction has never really been about knowing the future. It has always been about power over other people, and the technology companies making confident forecasts about AI stand in a very long tradition running through oracles, astrologers and the early statisticians.

Making predictions, she believes, is a way for those in power to force their vision on the rest of us. With the result that predictions about human beings have a habit of making themselves true. Once a forecast is used to decide who gets the loan, the job or the medical treatment, it stops behaving like a weather report and starts behaving like an instruction. The person refused is then told the system merely saw what was coming.

Put the two books together and Doctorow's argument gains the mechanism it was missing. He tells us the boss only needs to believe the machine can do the job. Véliz explains why belief is sufficient. Predict that radiologists are becoming obsolete, restructure the department on that basis, stop investing in the training pipeline, and in five years you will have radiologists who are effectively less capable than they were. The forecast will have been vindicated. It will also have been the cause.

That is the mechanism our AI business cases often rely on, and almost none of them acknowledge it. We treat workforce projections as observations about an external future rather than as decisions that help bring that future about. When a leadership team announces that a particular skill is going the way of the typewriter, it is not reporting the weather. It is writing a script and then hiring to it.

Véliz also supplies what Doctorow’s description lacks, which is a usable alternative. Her prescription is that a resilient society needs less prediction and better preparation. Preparation means building the capacity to cope with several futures. Prediction means committing to one and calling it inevitable. The second is cheaper, more confident and far easier to put on a slide, which is precisely why it keeps winning.

Who is to Blame

Which brings me to the third voice, and to the piece that finally resolved one of my concerns about Doctorow's book. He is intent on considering the future of AI as good versus evil, and workers being exploited for profit. I believe that the issues are more structural than personal. It is a view well-articulated by Nobel prize winning economist Joseph Stiglitz.

Stiglitz has been exploring these issues for a while. In a Columbia Bizcast interview last year, he examined what freedom means in the age of AI. But it is in a Financial Times essay this month that he sets out what he calls a progressive AI agenda, and arrives at conclusions close to Doctorow's by an entirely different route. His argument is structural. For investors to see anything like the returns now priced in, three conditions have to hold simultaneously: competition must stay limited, deployment must move fast, and the macroeconomy must be well managed through the disruption. The trouble is that these conditions are in conflict with one another.

Consider what this means in practice. If competition is genuine, and Stiglitz notes that barriers to entry look lower than the valuations assume, given how the lead keeps changing hands and how quickly prices have fallen, then profits get competed away and the returns never appear. If competition is suppressed and the monopoly profits are realised, the wider economic benefit shrinks and inequality widens, which weakens the very consumer demand those profits rest on. If deployment is slow enough for workforce displacement to be manageable, investors are disappointed. If it is fast enough to satisfy them, the resulting unemployment undermines demand and profits crater anyway. In all cases the arithmetic does the work.

Most useful of all is what Stiglitz does with competition policy. Doctorow offers antitrust as the fix. Stiglitz agrees it matters, then says the part that Doctorow leaves out: with real competition, investors do not get their profits, and the bubble pops. Enforcement is not the alternative to the crash. It is one of the things that brings it on. That is a harder and more honest position, and a warning to us all.

What both Doctorow and Stieglitz agree on is what gets left behind if AI investment collapses. A great many people will lose a great deal of money, and what survives is a powerful tool and the infrastructure to run it.

Three Voices, One Direction

These three writers bring three different perspectives. Doctorow on what happens at the workstation, Véliz on what happens in the business case, Stiglitz on what happens to the economy. All three arrive at the same place: this is being decided, and the people deciding would much prefer you experienced it as an inevitability, not a choice.

Their remedies are largely for other people on other timescales, and mostly American ones at that. Our levers here in the UK are different and arguably better suited to the problem. We have procurement at national scale, sector regulators with real teeth, and an employment law tradition that already treats work intensification as a legitimate subject of scrutiny. If the reverse centaur is the problem, procurement rules that ask who sets the pace are a more direct instrument than competition law will ever be.