Cynicism about artificial intelligence is here, but it is unevenly distributed. There is not enough scepticism about the technology’s applicability in a variety of fields in the business world, and too little awareness of how transformative it can be, and indeed already is being, for states.
More scepticism should be applied to how usefully AI can be deployed in creative fields. Chris Cocks, the chief executive of the entertainment conglomerate Hasbro, which owns among other things the tabletop role-playing game Dungeons & Dragons, recently made some eyebrow-raising and self-contradictory remarks about the use of AI in his own field.
He believes that century-old companies like his own have an advantage today because large language models and generative AI work more effectively the more information you can feed them, and “legacy” companies like Hasbro have plenty of information. Or, as Cocks put it: “D&D has 50 years of content that we can mine. Literally thousands of adventures that we’ve created, probably tens of millions of words we own and can leverage.”
There’s a tension here: Cocks is arguing both that the use of AI will reward large organisations which have their own data and resources with which to feed it, and that the costs of using it will remain low enough for this to be worthwhile. But for the likes of Hasbro to have a structural advantage in the age of AI, copyright and intellectual property will have to be strictly enforced. This has implications for both the cost and effectiveness of the technology. In evidence given to the UK House of Lords, OpenAI warned that it is “impossible” to train AI without the use of copyrighted materials.
The trouble is that the history of court judgments arriving at an expansive definition of what is already someone else’s intellectual property is long. Similar court cases might well drastically raise the cost of designing and deploying AI at scale, at least in the private sector. Cocks’ argument is that using AI for storytelling can be cheaper than using humans: but whether regulators and governments will oblige with the narrow definition of intellectual property this would require is an open question.
Add to that the cost — both financial and in terms of emissions — of running powerful enough processors, and there is reason to doubt that AI will decisively transform most workplaces and business ventures. But we should be less sceptical about the technology’s ability to transform entities that don’t have to worry about copyright, or any of the other limitations that might slow the rollout of AI, or make it exorbitantly costly: namely, states.
States need not worry about copyright or data protection for two reasons. First, since they write the laws in the first place, they can give them as much or as little autonomy as they want. Second, and more importantly, most of the relevant information for public policy is already in their hands. And although political pressure will be brought to bear on states to ensure that private companies can’t simply exploit other people’s intellectual property, there are votes to be won by automating bureaucracy, since that frees up government resources to be spent elsewhere.
If you are Hasbro, putting half a century of Dungeons & Dragons into a LLM might open you up to awkward questions about where, exactly, you end and the work of JRR Tolkien or Bram Stoker begins. But if you are the British government, then the only thing stopping you from doing the same with NHS patient data is your own failure to collect and maintain good data sets.
The UK government is already moving to improve the way it collects and stores data. It is rolling out schemes trialled at Chelsea and Westminster hospital in London to the rest of NHS England — for many ministers these are a model for how to innovate well in public healthcare. The trials involve general practitioners’ notes, discharge summaries and the clinical coding of records all being automated, which has more than halved the amount of time clinicians spend discharging patients. In the US, AI is helping officials from the Department of Housing and Urban Development to speed up the planning process.
The showy bits of AI — the stuff of corporate presentations — might well never be particularly cost-efficient or effective. But the back-office, time-saving functions that are already here are speeding up businesses. And given that almost everything governments do is in some way tied into back-office functions, AI has the potential to transform how states operate. While it might never be cheap enough to replace you at work, it already is changing how you are governed.
stephen.bush@ft.com