Artificial stupidity

Are you an ex-imperial power whose blue water navy can’t project power the way it used to? Economy flagging? Fertility drooping? Don’t worry: do we have the solution for you! Artificial Intelligence. No longer the stuff of early 2000s blockbusters — science fiction has become science fact. Once you have “unleashed” AI it will “turbocharge growth”. Forget about your wilting reputation abroad, after you’ve “mainlined AI into the veins of your enterprising nation” you’ll feel 200 years younger. Don’t get left behind: “move fast and take action to win the global race.” Be a “changemaker rather than a change-taker”. Once you’ve “backed AI to the hilt” you’ll find your mid-sized nation “irresistible to AI firms”, should you encounter any in dimly lit hotel bars.  

Amazingly none of this language came from an advertisement for erectile dysfunction drugs, but was in fact used by the Labour Party and Keir Starmer in promoting their “AI Opportunities Action Plan”. Having ruled out Keynesian stimulus (as I noted earlier this month), Starmer has precious little room for maneuver. He can’t cut taxes, he can’t seriously shrink the state and he is unwilling or unable to intervene in the economy and set industrial policy. With few options and worsening prospects, AI serves as a literal deus ex machina. According to its proponents, it can boost growth and productivity across the economy. And forget ultimate economic payoffs — if the government can quickly integrate AI into public services, it can potentially fix stagnating public sector productivity and improve public services without having to spend money it doesn’t have.

In some ways, it’s welcome that the government is embracing innovation. But the problem with our political classes is that they have a very limited understanding of science and technology and tend to engage in magical thinking about it, especially when a novel technology arrives on the scene. It’s one thing to say you want an innovative economy, but a survey found that only 17 per cent of MPs in the 2015 parliament had a STEM background. A 2012 survey suggested the majority of MPs were unable to solve very simple questions about mathematics and probability. More basically, does anyone think that Keir Starmer understands what an LLM (large language model) actually is, or how it works? Does anyone in the cabinet? Whether we should greet new frontiers in digital technology with fear or optimism, the main problem we currently have is that we greet it with ignorance. 

Whilst AI may well have beneficial applications, boosterish claims for its ability to “supercharge” productivity should be taken with a pinch of salt. Digital technology has certainly generated vast profits for those at the commanding heights of the industry, but it has offered no equivalent of the broad-based wealth and social mobility of the post-war boom in manufacturing and consumer goods. One of the main reasons productivity gains are anticipated for AI is that it is a way of automating jobs in white collar sectors. The promise made about the West’s deindustrialisation was that jobs in the “knowledge economy” would ultimately fill the gap — now, AI aspires to automate jobs in administration, medicine, publishing, law and many other professional fields. 

But let’s leave aside, for the moment, the desirability of automating hundreds of thousands of middle class jobs out of existence. Can it actually be done? The answer, as Keir Starmer is finding out, may well be no, or at any rate, not easily. Only two weeks into Labour’s ambitious scheme to revitalise the public sector, ministers have been forced to shut down or drop multiple AI prototypes. Little has yet been revealed about what is going wrong, but it isn’t terribly hard to make a guess. 

What, after all, does AI “automation” look like? William Morris, writing of the excesses of industrial automation, noted the problem of the “manufacture of useless goods, whether harmful luxuries for the rich or disgraceful makeshifts for the poor”. Cheap mass production can seem an unalloyed good, but in substituting automation for the labour of skilled craftsmen, it can easily be a vector for wastefully producing vast numbers of low quality and useless items. Human care, intelligence and taste can be bypassed and on a massive scale. 

Rather than experimental AI, much of the public sector could do with straightforward, modern standardised software

In exactly the same way, a company might save itself money by outsourcing, say, the writing of its instruction manuals to an AI. It isn’t great literature, so what does it matter? But what the customer loses is a concerned human mind designing words to be easily understood by another person. As more and more aspects of design, packaging, advertisement and sales become automated, physical goods themselves, and their use and presentation, could be doubly devalued and degraded as multiple layers of automation are applied to it, and human care and diligence is removed from the equation. 

The application of AI in areas entirely involving human social organisation is more alarming still. The government is currently working with American AI firms like OpenAI, Anthropic, Google and Microsoft, with bold promises to improve efficiency. But do we want to automate processes that involve children in care, disabled people receiving benefits, the allocation of social housing, Visa applications or medical records? For anyone who has experienced our catastrophic public sector, the very last thing we need is more paperwork, more process, and more data entry. Rather than experimental AI, much of the public sector could do with straightforward, modern standardised software and easier public access to human beings empowered to make decisions. 

Proponents for AI assure us that the technology will only get better, but it is worth examining the limitations of not only our PPE-brained politicians but also mechanistically-inclined tech bros. Sam Altman, one of the founders of OpenAI, has predicted that we will have an AI “superintelligence” in a mere “few thousand days!” For those who have a limited understanding of the technology, it’s easy to confuse the wildest sort of futurism with realistic predictions from those in the know. It’s worth noting that the basis of these kinds of speculation is not scientific, but metaphysical. 

Altman anticipates that “astounding triumphs – fixing the climate, establishing a space colony and the discovery of all of physics – will eventually become commonplace. With nearly-limitless intelligence and abundant energy – the ability to generate great ideas and the ability to make them happen – we can do quite a lot.” He assumes, blithely, that “artificial intelligence” will become capable of creativity, innovation and sometime at least comparable to human sentience. We should by no means think that computer scientists and engineers, many of whom have no deep grasp of the basic sciences, let alone neuroscience or philosophy, have a good basis for saying that even the most sophisticated computer or software is capable of replicating a human mind or the achievements of a human mind. As mechanistic materialists, they are much given to transhumanist fantasies of uploaded brains and sentient machines, but all of this speculation is built on a childish analogy between machines and brains that is demonstrably wrong, even on a sheerly material level. 

The result of a non-human “intelligence” producing writing, images and organising data is not some slow dawn of machine intellect, but rather, as we are all discovering — AI slop. The coinage, produced as the first AI images and text was unleashed on the internet, evokes the almost sickly sensation we have at the reconstituted garbage that LLMs reliably churn out. The ability of any idiot to hit a button and produce a picture or an essay has not been some economic or cultural renaissance, but in fact the degradation of public forums and an existential threat to education. AI is demonstrably making us stupider, and it isn’t intelligent — nor capable of being so — itself. 

AI is undoubtedly a leap forward in digital technology, but our political class is severely misjudging its potential and who is in control of it. First of all, it is too late for Britain to lead in AI — we have failed to capitalise on our strengths in computer science over the course of decades, and it will take years to correct this, if we ever do at all. In the meantime, embracing AI means giving a vast strategic advantage to American-based firms, who will be integrated into the functioning of the British state should massive adoption of AI take place in the public sector. 

Nor is being first to adopt a technology the same as being early to invent and market it. In fact, being an “early adopter” can be a disadvantage, as you experience all the errors, problems and higher costs of a prototype technology, whilst later adopters can invest less cash for a finished product. This is not UK PLC innovating to first place in the global race, this is Beta Test Britain, hoping to make a few quid injecting untested drugs into its weary veins.