Artificial Intelligence capabilities are developing so rapidly that annual reports – such as the AI Index that was recently released by Stanford University’s Human-Centered Artificial Intelligence department – can barely capture the speed of progress. But here are the highlights.
Some of the fastest-thinking AI models can “meet or excel” baseline human performance in PhD-level science exams. Some 53% of Americans now use a large language model (LLM) – achieving that scale of adoption faster than either the personal computer or the internet. More than 130,000 AI patents were registered in the world last year, up from 40,000 in 2020.
In the space of five years, this new kind of so-called generative AI has progressed from crude and expensive chatbots mimicking human responses to questions, to models that can use logic to solve problems, and move from answering questions to performing tasks.
The technology is, of course, still spectacularly bad at many things. Models that can outperform a PhD student in exams still cannot tell the time from a watch face, or find Christmas Day on a picture of a calendar. Even LLMs specifically designed to understand video fail to do so more than a quarter of the time.
And when faced with “Humanity’s Last Exam” – a genius-level pub quiz with 2,700 questions – no AI model has so far got more than 38% of answers right.
Meanwhile, humanoid robots, which look so cool in their controlled demonstration environments, are still completely useless when placed in unpredictable real-life situations.
As the public finally wakes up to the possibilities, risks and dangers of this new technology – including the distinct possibility of a financial market crash once the hype phase is over – it’s right to ask whether the politics of AI are in the right place.
The UK government has now launched Sovereign AI – a £500m state-owned investment fund that will support British-based AI companies, buying them time on a supercomputer, fast-tracking visa applications for highly skilled staff and giving them support to commercialise their products.
It comes on top of a string of initiatives: AI growth zones, with streamlined planning both for datacentres and related energy supply; a National Data Library, which is supposed to store information generated from health, crime and other services centrally and securely, ensuring responsible use; and £2bn committed to expanding supercomputer capacity in Cambridge, Bristol and Edinburgh by up to 20 times its current power.
But Labour’s most strategic move revolves around something it has avoided: regulation. While the EU passed a comprehensive AI Act, focused on mitigating risks and protecting citizens’ rights, the UK government has remained focused on targeted interventions – such as the ban on nudification, and the confrontation with Elon Musk over Grok.
In the end, Keir Starmer’s administration took on and defeated Musk by quietly raising the prospect of getting his products suspended from the Apple and Google app stores.
But this case-by-case approach, focused only on the so-called frontier AI models – the biggest and best – leaves risk mitigation in the hands of hard-pressed regulators. This includes Ofcom, the Information Commissioner’s Office, and senior officials in Downing Street, who are forced to play whack-a-mole with risks as they arise.
It creates a stark regulatory contrast with the EU, where certain AI models can be banned outright, and where tech companies have to prove their AI is safe before deployment. And while the UK has attracted a slew of corporations to expand their operations here – with Anthropic set to quadruple its staff in London to 800 – there is no guarantee that the British state will not be overwhelmed by the challenge of light-touch regulation, as it was with banking in run-up to the 2008 crisis.
Just as with railways and factories when they first appeared, it is inevitable that the state will need to impose general regulation. I would like to see less breathless boosterism from Labour ministers and more consideration of the principles behind the regulation they will eventually need to impose. Because AI is coming at us fast, and the benefits could be massive.
So far, much government thinking has been about attracting capital and talent, and investment in computing and energy infrastructure. But a social democratic party also has to think about redistribution of the gains that the technology will bring.
Probably the most fundamental philosophical challenge AI will pose to the centre left in Britain is in its attitude to efficiency and work. Labour is, after all, what it says on the tin – a party for workers and of employment. I’ve sat on countless platforms in the Labour movement where there is indifference and hostility to the idea that information technology should reduce the need for work and workers.
Suggested Reading
The truth about Trump’s Iran downfall? The worst is yet to come
The “dignity of labour” narrative is rightly deeply embedded in UK social democracy. But I am not sure it can survive what’s coming.
The question is whether the left’s job is to fight automation, or to embrace the efficiency gains of AI throughout all kinds of public infrastructure and services – including health, education and social care.
In some spheres – for example customer support agents – the productivity gains are already spectacular, with entry-level workers achieving more than 30% increase in output, according to one study. The Stanford report says there is now clear evidence that AI is replacing younger, less experienced workers in software development, marketing and accountancy.
So the Labour tradition needs to decide where it stands on this: the government is right to fight for the UK to retain a sovereign AI industry; and it is right to treat public data as a public asset. But if we really want to exploit the upsides of this new technology, we should start directing its use to socially beneficent ends.
What if, for example, we gave the 4.2 million adults on disability benefits a login to the “Pro” version of an LLM for free? It would cost less than a billion pounds. What if we then allowed them to perform paid work from a home computer or smartphone, some of which can be done through voice rather than a keyboard?
What if we opened up the UK’s low-altitude airspace to drones – allowing people in remote or run-down communities to fly delivery drones in London, or to control the autonomous boats that will soon service windfarms?
At the very least we should be rolling out a mass AI safety and user training programme: there is evidence that early AI use can dumb-down young professionals and force their thinking into uniformity rather than creativity. That can be mitigated by training and use-policies within firms and public bodies – and by “mass”, I mean everybody should get a go.
I have no doubt that, at some point, there’ll be an AI-driven financial crash. In its aftermath, the business models of the AI firms will change, and the state will gain greater traction over the way the technology is used.
In the meantime, I want the government to focus just as much on accelerating the benefits for everybody – in terms of productivity, creativity, quality of life and employability – as it does on subsidising startups.
Ultimately, if the threatened jobs massacre actually happens, the Labour tradition will need to look once again at a policy it’s been avoiding for decades: embracing the severance of work from wages and providing a universal basic income, alongside universal basic services, to a population with diminishing opportunities for work.
