Lord Willetts is President of the Resolution Foundation. He is a former Minister for Universities and Science.
The arrival of ChatGPT has shown how fast AI is moving. Search is being transformed. The new large language models can deliver data to us in forms indistinguishable from normal human communication. The implications are potentially enormous.
It is not perfect, however. It is untethered from any link to underlying reality. So it can “hallucinate” – generating plausible but untrue accounts. These are traps for the unwary. Knowing more in advance is the best way to get the most value out of search.
AI is not a monster from the deep either. It is not going to take us over. It feeds back to us a distilled version of what we say. That is how it generates nasty and shocking statements. But it does not have a deliberate intent to rule it: it is more mirror than an agent.
The most striking implications will be in the jobs market. Jobs involving the interpretation and use of data are very vulnerable to this next stage of automation. The journalist writing up the sports results or economic statistics or the lawyer ploughing through documents to find the key evidence could all find themselves displaced.
That will be a heavy blow for some. But it need not lead to a general fall in employment as there are always jobs to be done – even if it is hard to predict what they are. When I started my adult career, the Treasury had typing pools and there were a quarter of million coal-miners. Those jobs have gone but there are more people in employment now than then. The job of Government is to help people through these transitions, not to try to block them.
It would be wrong to assume that as the future is in AI so future jobs will be in AI. AI is automating some classic digital jobs – such as writing software. The future may lie with the jobs that can’t be digitised. The speed at which AI is advancing is a contrast with the slow arrival of driverless cars.
There is a message here. Analysing symbols is being automated with more and more types of symbol being included: after words we are also seeing it reach music and painting. So the jobs of symbolic analysts are most vulnerable. But the physical world proves as resistant and obdurate as ever. Jobs with things will do better than jobs with words.
What does it mean for the UK? There is bad news and not so bad. These advances in AI involve powerful computers analysing massive data sets. It is all about scale. I used to argue that Britain has distinct skills in writing smart software so we could outwit America’s vulgar focus on raw power.
I am not so sure now. Just using massive compute power can get you to the best answer quickly. Some commentators on the Ukraine war have explained Russian tactics of just throwing large numbers of soldiers at the front-line by quoting Stalin: “Quantity has a quality of its own.” Something like that may be true of AI. We do have major players such as Deep Mind. I was sorry it was sold to Google – another example of our failure to recognise the real value of our most innovative start-ups. But DeepMind reply that their work requires enormous compute power, costing hundreds of millions of dollars a year, and it was not clear than any company or public agency in Britain could meet that cost.
If anything the real constraint on AI is the rapid increase in compute power it needs to do so much analysis of so much data. One estimate I heard in Cambridge last week is that the compute power needed for AI is doubling every three and a half months.
But at the same time we are reaching the limits of Moore’s Law. The capacity of our chips is not doubling every eighteen months. This is a moment for real innovation in the design and operation of computer chips.. It is just possible that in Britain now are the exciting start-ups which are going to lead radical innovation. The Government has just published its semi-conductor strategy. There is no point trying to match the rest of the world in fabs for today’s silicon wafers. But it does make sense to nurture the truly innovative companies trying to do things differently – using new materials and compounds in novel ways to produce chips with different properties.
We do have one potential big data-set: the integrated patient-records of the NHS. The Americans don’t have anything like it. We should be able to link modern genomic data with patient records going all the way back to childhood and other social data too.
That would be a data-set of global significance. It would enable these powerful new AI tools to track the origins and patterns of disease, linking genes and experiences in ways very few health care systems can do. It would require establishing an ethical regime.
But it would also require that the NHS really did have genuinely usable integrated patient records. Lots of AI experts approach the NHS hoping and believing it does but go away empty-handed because the raw data is a mess. This ambition has been around for years now. But we have not been able to deliver it. Sorting that mess out is the best single thing we could do to ensure Britain benefits from the AI revolution.