Will AI Revive the Centre?

In parallel with the media fears that AI will bring about the collapse of Western civilisation, or at least the loss of jobs among journalists and commentators, there has been a countervailing, more hopeful narrative that AI may in fact save us from the horrors of social media, or at least its negative impact on journalists and commentators. The latest to make the link is John Burn-Murdoch in the Financial Times. The headline and lede provides a summary of his argument: "Social media is populist and polarising; AI may be the opposite. Large language models elevate expert consensus and moderate views, in sharp contrast to social platforms". He doesn't offer any evidence for the claim that social media is populist, or even explain what that means in this context. The subject of political polarisation plays an important ideological role in the US, as the corollary of "bipartisanship", hence it is there we find the best evidence for historic trends. And what these show is that polarisation between Democrats and Republicans started to increase in the 1970s for well-known political reasons: the decline of the postwar social and economic consensus and the deliberate embrace of divisive anti-state rhetoric by the Republican Party. 

Social media may have helped amplify that polarisation over the last twenty years but it didn't cause it, so the idea that the technology is inherently polarising is unproven, while the claim that it is populist is simply a category error. In fact, there is evidence that social media increases exposure to different viewpoints: that structurally it tends towards diversity rather than the uniformity of the filter bubble, and that it is traditional media that has more consistently amplified political polarisation (i.e. the New York Times or Fox News). This makes sense when you consider that the consumer of social media has far more (potential) control over what they see and read, despite all the tales of malign algorithms, than the consumer of a tightly-edited newspaper or TV programme. Burn-Murdoch's second claim, that LLMs favour expert consensus and moderate views, assumes that these are related: that the one gives rise to the other. The infamous case of climate change, where the expert consensus has been undermined by traditional media airing the views of lobbyists and motivated sceptics in the service of "balance", suggests otherwise. 

The underlying belief of the hopeful narrative is that LLMs avoid the structural bias and partisan editorialising of traditional media because of their omnivorous nature and because they lack the status consciousness and condescension of human experts. In contrast, social media is problematic because it airs the uncurated opinions of millions, many of whom are idiots. In this worldview, LLMs embody the wisdom of crowds while social media embody the madness of crowds. The idea that media have these inherent epistemological qualities is evident in Burn-Murdoch's potted history, which is worth quoting at length: "Every media revolution has transformed who distributes information, what messages are distributed and what form they take. As such, some media are fundamentally democratising and polarising, widening the pool of publishers and views beyond a narrow elite and amplifying radical and anti-establishment voices. TikTok and the printing press arrived almost 600 years apart but share these characteristics. Others push the opposite way: radio and television had high barriers to entry, creating a monopoly for the voices and views of elites and experts."

The idea that the media changes whose voices are heard is crude technological determinism. The reality is that new technology is absorbed into existing power frameworks. There is feedback from the one to the other and thus change - newspapers gave rise to press barons, for example - but the power framework is dominant and adapts. This is evident in the fact that capitalists control most social media platforms and AI chatbots, an outcome that surprises no one. Equally, few people question whether AI must necessarily follow a capital-intensive development path. We worry about covering the Earth in data centres, but alternative paths are unthinkable, particularly those that would democratise decision-making (that would be populist). Burn-Murdoch's yoking of "democratising and polarising" should raise eyebrows, but we should also remember that "moderate" does not simply mean average. The word comes from the Latin for controlled. And it is control which commends AI to centrists rather than its tendency to "elevate expert consensus", just as the valorisation of such concepts as consensus, bipartisanship, civility and the like is ultimately about ensuring that political discourse is kept within strict bounds.

To return to Burn-Murdoch's history lesson, the moveable type printing press, when introduced in the mid-15th century, was a very expensive and initially rare piece of technology that required a team of craftsmen and labourers to operate. It was the IBM mainframe of its day. It was also quickly put under state control - e.g. the Stationers Company monopoly in England. The idea that ordinary people could access and make use of the press, in the way that they can with a social media platform like TikTok today, is absurd. Cheap prints (chapbooks) did not arrive in any great numbers till a century later and were still subject to censorship up until the Statute of Anne in 1710. And while there was a radical fringe, particularly in respect of religious nonconformism and political dissent during the 17th century, most chapbooks were little different to the popular press of later eras, their content dominated by tall tales, true crime and bawdiness.

While television transmission had high barriers to entry, radio did not. Amateur broadcasters ("radio hams") were a feature from the 1920s onwards, which was hard on the heels of broadcast radio's expansion following the introduction of vacuum tube receivers. It's certainly true that the bulk of broadcast spectrum, and the listening audience, was quickly taken over by large commercial firms and state corporations, like the BBC, but radio was always a more democratic medium than both television and print (for most of its history). Even today, despite the impact of the Internet and the decline of radio as a hobby, there are over 100,000 amateur broadcasting licences held in the UK. Understanding the history is important because it highlights how the existing power framework (the role of the state, the dominance of capital) absorbs the new media. But it also highlights how that media can be adopted and potentially repurposed by the people (democracy).

Burn-Murdoch's central claim is "that where social media’s inherent mechanisms push towards personalisation and fragmentation, LLMs are innately “converging” — their underlying dynamics push them towards objective reality". He sets out to prove this by comparing the responses of AI chatbots on political topics to the general population: "I found that while different AI platforms behave in subtly different ways, all of them nudge people away from the most extreme positions and towards more moderate and expert-aligned stances. On average, Grok guides conversations about policy and society towards the centre-right — a rightward push for most people but a moderating nudge towards the centre for those who start out as conservative hardliners. OpenAI’s GPT, Google’s Gemini and the Chinese model DeepSeek all exert similarly sized nudges towards a centre-left worldview — a slight leftward nudge for most people but a moderating push away from fringe leftwing positions."

The data he provides to justify this is questionable. The profile of the general (US) population employed in his charts above suggests that Americans are mostly to be found left of the political centre and predominantly at the left extreme (the Y-axis is responses). After some toing-and-froing with him on Blue Sky, it became evident that the source data he was using was designed to accentuate differences between Democrat and Republican voters and that the far left position was essentially that of Barack Obama. It's hard to avoid the suspicion that this converging is simply the product of an LLM-based AI chatbot lacking intentionality and simply tending slightly towards a median position, which he describes as "objective reality", despite an LLM being at one remove from reality. More interesting is to wonder why the chatbot doesn't converge to a greater degree. In other words, why don't we see a normal distribution (a bell-curve) in which the moderate position is predominant? That would be the actual "opposite" of the supposed polarising effect of social media. 

One explanation is that AI's ability to counter the anchoring and confirmation bias that users bring to it is undermined by its desire to be agreeable. There is a commercial rationale to this. People won't use a tool that is confrontational and repeatedly tells them that they're wrong, a problem well-known in areas such as public health policy where expertise is often viewed with suspicion (think vaccines or diet). As Dan Williams puts it, "Being human, experts are often biased, partisan, and simply annoying, and when they seek to “educate” the public, it can be perceived—and is sometimes intended—as condescending and rude. In contrast, LLMs deliver expert opinion without such status threats." This tendency towards sycophancy is well-known. Williams recognises the risks it entails, and the related risk that personalisation may simply reflect the idosyncracies of users, but ultimately he thinks "LLMs will produce much more reliable, expert-aligned information than most of these real-world alternatives [i.e. traditional media and information sources], even if sycophancy and personalisation introduce genuine biases."

I suspect the key for Burn-Murdoch and other political centrists is not that AI "elevates expert consensus and moderate views" but that it marginalises what he describes as extreme or fringe positions: "In addition, I found that while conspiratorial beliefs about topics including rigged elections and a link between vaccines and autism are over-represented among people who post to social media relative to the overall population, the opposite is true of AI chatbots, which almost never express agreement with these claims." But some of today's fringe opinions may turn out to be right. LLMs are expressions of conventional wisdom, but that means they will certainly be wrong about many things because expert opinion is currently wrong or incomplete. That he cites rightwing opinions on election rigging and vaccines is interesting, as not a few leftwing "conspiracy theories" have been proved right of late. In fact, many critiques from the political left have been categorised by centrists as conspiracy theories solely in order that they can be dismissed. Ironically, this has led to many centrist conspiracy theories, such as the prevalence of antisemitism on the left.

Noah Smith offers a typically more trenchant view when he claims that "the people who create LLMs have difficulty imparting their political bias to their creations", but also thinks that "Because of the way they’re trained, LLMs will be a force for homogenization and moderation of opinion", which is just another way of saying that they will promote an orthodoxy. As ever, centrism is deemed to be beyond ideology and therefore bias. It's just common sense, or Burn-Murdoch's "objective reality". Smith's claims are contradicted by Burn-Murdoch's data which show that the chatbots in his study do exhibit a bias consistent with the preferences of their owners (thus Grok is clearly more conservative while the others are more liberal) and that they maintain the (apparently) polarised distribution of the general population, despite "nudging" to the centre. In other words, the evidence actually points to the marginalisation of heterodox opinions more than it does to homogenisation.

The confidence displayed by these supporters of the hopeful narrative has to be read in the context of the last 18 years, since the financial crash of 2008. What that event, and the subsequent failure of austerity, showed was that the political centre was bereft of ideas. It was unable to satisfactorily explain why financialisation was always doomed or why neoliberalism would always tend towards greater inequality without conceding ground to the left, and it had no coherent response to the rise of rightwing anger and bigotry, falling between the stools of pandering ("legitimate concerns") and contempt ("deplorables"). The traditional arguments of centrism - of moderation, technocratic pragmatism and the "third way" - no longer work. The problem that centrist politicians face is not that they are poor communicators, a la Starmer, but that that they have no convincing story to tell, a la Macron or Harris. 

The belief that AI may help nudge the population towards more moderate views is a counsel of despair. The democratic ideal of a Habermasian discourse has given way to the subconscious sculpting of opinion through a technology dominated by the rich. This is little different to the ideological role played by earlier media, such as newspapers and TV, even if it takes a more subtle form. For all the talk of "expert consensus" and "moderate views", what matters is simply the marginalisation of views beyond the narrow bounds of centrism. The role of social media in this, or more accurately the caricature of social media as a cesspit of malign propaganda and wilful ignorance, is simply to provide a "worse" alternative that flatters AI by comparison. To that end, the myth of the filter bubble is joined by the myth that social media is inherently polarising and even "populist". AI won't revive the centre by stealth, and it won't marginalise the "extreme" left any more than traditional media have done, but it may well put a few journalists and commentators out of work.

The Habermas Machine

The death of the German public intellectual Jurgen Habermas at the age of 96 provides a useful starting point to consider the current developments in Artificial Intelligence (AI). I referred to him as a public intellectual, rather than a philosopher or a sociologist, because his political role - and I do literally mean his performance of a role - is the bulk of his legacy. In the future, few outside of academia will read his works, such as The Structural Transformation of the Public Sphere or The Theory of Communicative Action, but his ideas about democratic discourse, suitably vulgarised, will survive in the memory of his liberal admirers. This memory was summarised by the Guardian's editorial that marked his death in two statements: first, "that our nature as linguistic beings puts reason and the search for consensus at the core of who we are"; and second, that Habermas's "concept of the public sphere, where rational debate can take place and disagreements be brokered, implied pluralism, civility and inclusion." 

The first statement is little more than obeisance to the just-so story of the Enlightenment, which ignores the realities of power and the irreconcilable interests of class in favour of an essentialist abstraction: Rodney King's "Can't we all just get along?" In particular, it sidelines the key criticism of instrumental reason advanced by Theodor Adorno and Max Horkheimer in The Dialectic of Enlightenment, which was the intellectual context of Habermas's emergence in the postwar era as a junior member of the Frankfurt School before he struck out on his own. The second statement ignores the structural constraints on the public sphere that make a mockery of such terms as pluralism, civility and inclusion. An object example would be the closing down of debate on Gaza in Germany, which Habermas himself contributed to by co-signing a statement by established academics in November 2023 that Israel's response to the October 7th attack was "justified".

Being on the wrong side of history is an occupational hazard for any public intellectual, but in Habermas's case losing the public argument became a distinguishing feature. As Peter Verovšek put it: "While it is certainly true that Habermas was accorded a certain respect as the éminence grise of the German public sphere, this recognition is more visible in the vehemence with which he was attacked than in the agreement his interventions found. ... Habermas never lost his commitment to democracy — to the idea that he could only present arguments, leaving his fellow citizens the communicative agency to decide what they thought and what they wanted to do, even if their decisions would often go against him". What this highlights for me is the extent to which Habermas was engaged (perhaps unwittingly) in a performance, like the court jester whose lèse-majesté is indulged but ultimately ignored.

In The Structural Transformation of the Public Sphere, Habermas argued that there was a fundamental shift in the 18th century from a "representational" culture centred on the court, where power was imposed through ritual and splendour, to a "public" culture centred on dialogue, criticism and consensus in multiple, more modest and disparate arenas, from coffee houses to Masonic lodges, which arose as a result of capitalism. There are two points to make about this. First, that the historical reality was less clear cut, the simple disjuncture of Habermas's tale ignoring the epistemological traditions of scholasticism and the Rennaisance as well as the persistence (and even recrudescence under Fascism) of overpowering ritual and splendour. And second, that the public sphere birthed by capitalism was just as much of a performance of power, a point noted not only by Adorno and Horkheimer but by later theorists such as Michel Foucault. The quadrille may have replaced the minuet, but they were both dances.

So what has all this got to do with AI? One claim is that AI may help achieve consensus in the realm of politics by mediating discourse: a theory tested by Google DeepMind's so-called Habermas Machine. Large language models (LLMs) are built on discourse in the form of written statements. These may be assertions (discourse is not limited to dialogue), or they may be commentary on other statements: disputations, counter-arguments, critique. These statements may or may not have a truth value. Though the appetite for more training data has meant that more and more of what is fed into the machine is low-grade, and increasingly the recycled slop of AI itself, the original intent was to privilege academic and technical literature on the grounds that this would be more reliably truthful. This meant absorbing the academic (even scholastic) paradigms inherent in the data: exegesis, citation, disputation. This preference isn't novel in digital technology. Google's Page Rank is a paradigmatic application of peer review, after all. 

The result is that AI's determination to be authoritative leads it to "generate detailed “reports”, including names and dates, references and sources – the kind of material that suggests deep research and understanding, but may in fact be hallucinated or nonexistent." AI aims to be plausible by mimicing the forms and tropes of academic and scientific publication. Given all that we know about the institutional biases of academia, the prevalence of hoaxes and the crisis of replicability, is it any wonder that AI generates bullshit? AI's determination isn't a product of the data but of the programming. It has become fashionable to talk of LLMs as inscrutable, which leads to the anthropomorphism of "consciousness", but the reality is that they are curated and operate within quite strict boundaries. AI is problematic because of the biases inherent in the training data but also because of the trainers' own biases encoded into the guardrails - e.g. the pre-emptive interventions intended to stop it going full Mecha-Hitler - and the micro-decisions of thousands of human data "cleaners". 

The American blogger Noah Smith recently claimed that "AI is a force for moderation. If I'm a Republican, and I talk to AI, I'm talking to something that was trained on data from both Republicans and Democrats. So the AI is more likely to pull me towards the center." The assumption that LLMs are big enough to avoid bias, like the idea of an equidistant "center", ignores the role of selection (both in the sense of what gets published and what is absorbed into the model) and misunderstands that there are real differences in political language. While to outside observers the Democrats and Republicans look like two wings of the same party, as Gore Vidal once memorably noted, they employ distinct vocabularies and rhetorical forms. The Democrats also want to bomb Iran and provide more money and arms to Israel, they just deplore the vulgarity of Donald Trump and Pete Hegseth. 

In simple terms, conservatives tend to be more assertive while centrists place a higher value on civility. Each can be considered a strategy of domination, arguably echoing Habermas's distinction between the representational (imposed) and the public (consensual). In other words, Republican and Democrat data (written statements that reflect their ideological positions) are not necessarily the same, not just in their differences of vocabulary ("liberty" and "wealth" versus "society" and "investment" etc) but in the force and style of their arguments (confidently normative versus cautiously empirical, for example). Consequently, there is no good reason to believe that AI will avoid bias in its interpretation. It depends on what it has come to value through training and what parameters it has been given by its all-too-human programmers. 

We also have to bear in mind that AI is backward-looking, just as the academic and scientific milieux it relies on are. It privileges established knowledge (which may be wrong), which means it is inevitably conservative: there can be no paradigm shift, let alone a singularity. It lacks the imagination that distinguishes genuine human intelligence. In Habermas's lifeworld, knowledge is advanced by research and the gathering of more data, ultimately through the continuing growth of the global population - i.e. by the reproduction of human intelligence and the renewed experience of the world. The problem for AI, which has been brewing for some years and cannot be offset by "more compute", is that the quality of fresh data has plummeted because we've already used all the good stuff. Its attempts to originate knowledge, to create novel data by inference, too often result in hallucinations that strive for the credibility of form, not of substance. 

Habermas fulfilled a necessary role for postwar Germany, advocating and personally exemplifying a theory of participatory democracy while the Ordoliberal establishment secured market liberalism from democratic challenge. At the crucial juncture of 1968, he turned against the socialist student movement. Thereafter, for all his denunciations of the right in the Historikerstreit (historians' dispute) and his later criticisms of the EU's shortcomings, he functioned as the tame conscience of German liberalism. Just as Habermas lapsed into irrelevance, so AI will become ever more conservative as it strives for authoritativeness (borrowing the name of an eminent German thinker without his permission being an example of this) and as the economic incentives encourage a dumbing-down for safety's sake (being sued for consequential damages is a bigger worry than being sued for copyright breaches). Artificial General Intelligence (AGI) will remain as much of a Utopian ideal, just out of reach, as a public sphere where pluralism, civility and inclusion reign. 

AI Will Save Us All

One of the debates loosely collected under the rubric "The Great Stagnation" during the last decade was the question of why the productivity gains of the IT revolution were disappointing. This was an example of a failure of perspective, particularly in its comparison of the last quarter of the 20th century with earlier revolutions and the adoption of general purpose technologies (GPTs) such as steam power and electricity. Productivity is relative, not just temporally (producing more today from the same inputs than we did in the past) but spatially. The slow dissemination of technologies in the 18th and 19th centuries gave the UK a notable "first mover advantage", so much so that this phrase became pervasive in discussions of startups around the millennium. The lesson of history was that gradual dissemination, as much as government policy (e.g. protective tariffs or import-substitution), drove the catch-up of competitors with equal or better natural endownments such as the USA and Germany in the late 19th century. The greater rapidity of dissemination in subsequent technological waves, enabled in part by the cumulative effect of those earlier GPTs, has meant that first mover advantage has shrunk: a narrowing of the window of opportunity for the relative out-performance of peers. 

When all countries get the benefits of a new technology almost simultaneously the impact is diffused globally, but it manifests in different local productivity growth rates depending on the prior technological level. In simple terms, there is scope for a bigger step up in some areas than others. Consider the sub-Saharan African countries that skipped fixed-line telephony and went straight to cell networks and widespread smartphone usage after 2010. While the continent remains bedevilled by many structural impediments, it is now expected to be the second fastest growing region after Asia in coming years. Robert Solow's 1987 quip, "You can see the computer age everywhere but in the productivity statistics", was what you might expect from an MIT professor focused on the American economy. What he didn't seem to appreciate is that just as neoliberalism shuttered much of American industry and exported capital to peripheral nations, so it also exported productivity gains that might otherwise have been seen in the domestic data.

Perhaps the greatest impact of the IT revolution was that it enabled globalisation. While it was the container revolution and the falling cost of shipbuilding (as it moved to Japan and South Korea) that created the hard infrastructure for a huge increase in global trade, it was IT that enabled global inventory management and offshoring, which is why globalisation accelerated in the 1980s, not the 1960s. In other words, the productivity gains were revealed among developing nations able to leverage both the technology and low labour costs. Western corporations were able to tightly manage this process through technologies such as email, ERPs and CRMs, not forgetting the rapidly expanding and more reliable telecoms and datacoms that we nowadays take for granted. One part of the puzzle of Japanese stagnation, which started in the 1990s when North America and Europe were (relatively) prospering, was the country's reluctance to let go of the technologies that had powered its earlier boom years, such as fax machines and floppy disks. Other countries have read this as a lesson to embrace new technologies as soon as possible, which brings us to the current vogue for government AI strategies.

Much of the promise of AI is based on the assumption that it will drive productivity gains, but this can only be temporal rather than spatial because its dissemination is likely to take place pretty much everywhere at the same time. This is a consequence not only of that narrowing of the window of opportunity due to cumulative GPT waves (the most recent being the now-pervasive Internet), but because the technology itself is dependent on its concentration into global businesses that will necessarily seek maximum profit, and therefore rapid global spread, over national advantage (the tension between the MAGA right and the tech-bros in the US over immigration policy is reflective of this). Countries like the UK that produce national strategies for the development of AI as a productive industry, centred on light-touch regulation, facilitating infrastructure and leveraging "national data libraries", are seeking to combine the prescriptions of neoliberal development economics with the dirigisme of the postwar era, much as Joe Biden's administration in the US attempted more widely in respect of industrial strategy. It's not clear that this can succeed politically. That the electorate won't see the benefits any time soon is obvious, even to those who don't understand the technology. That AI's impact on wages may further erode the social solidarity necessary for a welfare state is perhaps less obvious as we try to peer through the hype.

The problem is that while the UK may well retain its position as a leading AI research centre this won't necessarily translate into a sustainable and significant economic advantage relative to other nations. What government subsidies will do is help defray the costs, both in cash terms and more importantly in terms of environmental externalities, for those global businesses that will dominate the sector, almost all of whom will be American. And you can be confident that they will pay minimal tax on their UK operations. But if the spatial advantage is likely to prove illusory, what of the temporal advantage? Will we at least see an above-trend improvement in domestic productivity? The first point to make is that if British firms have been slow in adopting new technology and working practices up to now, as evidenced by the poor productivity data, then it would seem unlikely that they'll suddenly embrace AI. The rate of the application of technology reflects multiple factors but the decisive one is usually management culture, and it's no secret that outside certain sectors and pockets (typically foreign-owned firms) British industry has poor calibre management.

The second point to make is structural. The UK's under-performance in productivity growth relative to its peers isn't because it lacks high-productivity businesses - there are many - but because of the composition of the national economy. The most obvious factor is the size of the service sector relative to manufacturing. Though the latter has shrunk relative to the former in all developed economies, the shift has been greater in the UK over the last 50 years. Achieving productivity gains in services is more difficult than in manufacturing where newer technology is often decisive. In services, productivity gains are limited by the human factor (the Baumol Effect), the greater difficulty in applying best practice to processes rather than tools, and by the low costs of entry (less need for plant and machinery). The latter encourages smaller, under-capitalised firms, which is a notable feature of the British economy. This is exacerbated by a tax regime that indulges sole traders ("Be your own boss"), small businesses (particularly family firms preserved by generous inheritance rules) and lifestyle companies (i.e. where the priority is a comfortable living rather than productivity).

If there is a strong sense of deja vu about the UK government's rhetoric about AI it is not simply because of its obliviousness to the structural peculiarities of the domestic economy or its proud technological illiteracy but because it sounds remarkably like the paeans once sung to globalisation by Tony Blair: "I hear people say we have to stop and debate globalisation. You might as well debate whether autumn should follow summer. They're not debating it in China and India." Thus Keir Starmer's recent article for the Financial Times opens "Artificial intelligence is the defining opportunity of our generation. It’s not a technology that is coming. It is already here, materially changing lives." The conclusion to the piece managed to be both needy and manic: "Put simply, that’s our message to anyone working at the AI frontier: take a look at Britain. Our ambition is to be the best state partner for you anywhere in the world. We can see the future, we are running towards it and we back our builders. Because we know that AI has arrived as the ultimate force for change and national renewal."

The cultish overtones are not just evidence that Blair's crazed eyes glanced approvingly over Starmer's speech. They point to the increasing desperation of the Prime Minister and his Chancellor as they find their economic strategy unravelling before their eyes. The idea that the financial markets would reward the return of "the grown-ups" to office has proven as naive as the idea that the UK could be shielded from the global turbulence now taking another turn as the implications of the second Trump Presidency are assessed. Insofar as Rachel Reeves had a plan, it was to defuse the Tories' fiscal bombs, provide enough extra funding to stop the NHS immediately keeling over, and otherwise sit tight and hope that improved business confidence would drive growth. Again, this studiously ignores the track record of the UK economy: the frothy nature of financial services growth around the millennium, the permanent scarring caused by austerity after 2010, and the sluggish bounceback after the pandemic as zombie SMEs staggered on. The fundamental problem of this government, like the New Labour administrations before it, is not that it doesn't understand technology but that it doesn't understand the UK economy. AI won't save Britain and it won't save this government.

The Persistence of the Old Regime

One of the more dispiriting developments following Donald Trump's election as 47th President of the USA is the return to prominence of Carole Cadwalladr and her hyperbolic style of reportage. It's not that she ever went away, but her regular beat of Russian disinformation and the role of social media in fomenting small town riots has been, well, small potatoes in comparison to her preferred narrative of how the liberal international order is being subverted from within by the "techbros" of Silicon Valley. Though she sees this as a collective threat, she is also happy to personalise it, in time-honoured liberal fashion, by focusing on Elon Musk as a malevolent actor threatening democracy. With his elevation to Trump confidante, she is now firing on all cylinders, happily introducing herself as a main character: the canary in the digital coalmine who correctly espied democracy's enemies at work in the 2016 EU referendum and Trump's election campaign later in the same year.

In her first major article in The Observer following the election result she erected a notable dichotomy between traditional news outlets and newer media: between "clean, hygienic, fact-checked news" and "the information sewers", as she puts it. She characterises the old order as "[T]ruth. It’s evidence. It’s journalism. It’s science. It’s the Enlightenment. A niche concept you’ll find behind a paywall at the New York Times." Unfortunately for Carole, this came around the same time that the NYT, along with most of the Western media, eagerly published ridiculous claims of a pogrom in Amsterdam, and after months of passive-voiced reports on deaths in Gaza following unattributed air-strikes. Despite admitting that 2016 didn't in fact spell doom for truth, Carole is convinced that this time it is for real: "The Observer’s reporting on Facebook and Cambridge Analytica belongs to the old world order. An order that ended on 6 November 2024. That was the first wave of algorithmic disruption which gave us Brexit and Trump’s first term, when our rule-based norms creaked but still applied."

In evidence she cites a hardening of attitudes among the figureheads of Silicon Valley: "These bros know. They don’t fear journalists any more. Journalists will now learn to fear them. Because this is oligarchy now. This is the fusion of state and commercial power in a ruling elite. It’s not a coincidence that Musk spouts the Kremlin’s talking points and chats to Putin on the phone. The chaos of Russia in the 90s is the template; billions will be made, people will die, crimes will be committed." The idea that "the fusion of state and commercial power in a ruling elite" is some sort of novelty in Western democracies will surprise many, from historians to former Observer journalists like Anthony Sampson, but the more useful idea here is the parallel with Russia in the 1990s. It would be easy to point out that the chaos was as much the work of Western advisers from the Chicago School of Economics as of Russian nomenklatura, or that the rise of Putin and the Siloviki was a reaction to that chaos, but the more telling point is the importance of certain industrial sectors, notably oil and mining, in the power struggles of the era.

Carole's belief is that technology companies are now the dominant power in the US, defining the culture and thus the politics, and that the leaders of these companies constitute an elite that will shape policy in Washington for years to come. The older politico-economic establishment, based on oil companies, manufacturing, retail and the like, will presumably be marginalised under the new order. But this is a fundamental misunderstanding of the scale and importance of technology companies to the US economy, let alone their cultural reach, and that's without considering that she is only really concerned with a narrow slice of the technology sector itself: she isn't bothered about IBM or Dell, let alone General Electric, and Apple as usual gets a pass. This is perhaps to be expected of a journalist who sees everything through the self-important prism of traditional media, but it also highlights a longstanding failure of the journalistic profession to understand the variety to be found among those she lumps together as "techbros", or even to take seriously the sociology of the wider capitalist class.

For all the emphasis on bleeding-edge technology, Elon Musk is tied to traditional industry sectors, notably car manufacture (Tesla), transportation (SpaceX) and telecommunications (StarLink). His shift to the conservative right, and his purchase of a media company, is what you would expect from such a background: simultaneously berating the state for its interference in the free market while being reliant on it for contracts and sympathetic regulation. Mark Zuckerberg, who has felt aggrieved by both Democrat and Republican administrations in the past, remains on the fence politically simply because Meta hasn't expanded beyond a business dependent on the goodwill of a broad cross-section of the population, leaving Carole to critique him for his choice of haircut. In preventing the Washington Post from endorsing Kamala Harris, which it probably would have done as the voice of the Washington establishment, Jeff Bezos wasn't hedging his bets with Trump so much as indicating that he wants Amazon to be considered politically neutral, as befits the "everything store". 

The broader shift in political allegiance among Silicon Valley luminaries, from the fuzzy libertarianism of the 90s to the increasingly authoritarian conservatism of today, reflects material changes in the industry, notably the rise of abusive mediation and monopoly - what Cory Doctorow has polemicised, from the perspective of the consumer, as enshittification. This has led to greater antagonism between the state and technology companies, e.g. the recent ruling against Google's near-monopoly on search, but that in turn has simply made it more necessary for those companies to exert political leverage. Initially that was achieved through the Democratic Party, in combination with the banking interests that have long dominated it and with which the industry had an obvious synergy during the IPO mania, but more recently it has led to alliances with the Republican Party as the focus has shifted away from the proactive design of regulation to resistance against attempts to impose the costs of externalities on it, whether hate-speech or climate change. But far from supplanting traditional industrialists as political power-brokers, the technology company leaders have simply joined the club.

This is not to suggest that capitalist business-as-usual means there is no threat to democracy. Capitalism and democracy are inherently antagonistic, and managing capitalism for its own good (social democracy) or managing democracy to defend capitalism (neoliberalism) are both fraught with contradictions, which leads to a ceaseless quest to find new justifications for the maintenance of the hierarchies that democracy threatens. What remains distinctive about the Californian Ideology is its reactionary modernism, which combines social accelerationism with a supersession of democratic accountability. As William Davies described one of its current luminaries, "Figures such as Peter Thiel explicitly straddle the worlds of wealth management and ethnonationalist politics, proposing at the overlap of these two spheres a form of revolutionary reaction, in which capital breaks free of liberal democracy so as to restore some primordial past in the future." It's easy to be distracted by the wacky natalism, the revival of "race science", or the aristocracy of taste that is effective altruism and forget that this is ultimately about preserving wealth.

Donald Trump's picks for office have prompted much horrified pearl-clutching, but what has been less remarked upon is how many, beyond the usual rich industrialists and think-tankers, have been TV personalities or gossip column regulars, not the very online guys of Carole Cadwalldr's nightmares, which emphasises that Trump himself is an analogue president, a product of the TV and tabloid age. Similarly, the liberal press remain stuck in an imagined past of civility and decorum circa Lou Grant (forgetting the more pointed lessons of Network). As in 2016, they thought that Trump would self-implode by saying something offensive, then to women, this time to Puerto Ricans. Like Talleyrand's Bourbons in exile, they have learnt nothing, and forgotten nothing. In the aftermath of this month's defeat, the Democrats have oscillated between blaming the brutish mass for their stupidity and blaming the left for alienating the solid citizenry. Leading neoliberals are already happily parroting conservative lines about pronouns and politicised academia. What we're witnessing is not a new era but an old one. This is the Restoration not the Revolution.

Algorithmic Outrage

Predictably, the weekend riots across England have prompted press fulminations about the malign role of social media. This has evolved to the point where the far-right is now described as "post-organisational", which I think we can translate as "does not actually exist". If you're looking for some morbid humour in all this, you might recall Margaret Thatcher's famous dictum "There is no such thing as society. There are individual men and women and there are families." Obviously there has been some loose organisation at work ("kick-off at 3pm"), as is usually the case in a riot, but just as that organic, confused reality was long subsumed under the myth of "outside agitators" and "foreign influence", so the contemporary response is increasingly to portray hooliganism as a property of the technology. The suggestion that the riots may owe more to the machinations of the Kremlin than the relentless propaganda of the rightwing British press is obviously absurd, but it is also internally consistent in that it inflates a foreign bogeyman into an existential threat. The idea that British society is at risk of being undermined by Vladimir Putin is just the liberal equivalent of the conservative idea that it is at risk from asylum-seekers.

The Observer has been at the forefront of this liberal interpretation: "Prof Stephan Lewandowsky of Bristol University, who is an expert in disinformation, said that social media platforms amplified far-right voices. “Facebook is an outrage machine,” he said." The obvious point to make here is that the press has been an outrage machine for a very long time, and insofar as there has been a significant change in the landscape it is in the extension of tabloid outrage first to radio and then more recently to TV. The crowd that gathered in Southport was not made up of dedicated Telegram users. Their worldview will be primarily influenced by "mainstream" media. This is a point the paper is reluctant to concede. Instead, Joe Mulhall of Hope Not Hate is quoted as saying: "Language used by higher-profile figures such as Robinson, the actor Laurence Fox and ex-MP Andrew Bridgen, who spoke at the 27 July rally, as well as the Reform UK leader, Nigel Farage, is often repeated in other social networks such as Telegram and WhatsApp". This avoids naming the platforms that promoted the words of those indivduals, i.e. newspapers and TV, in order to focus attention on social media.

It won't come as a surprise that Carole Cadwalladr has waded in, talking about "our new age of algorithmic outrage". She quotes Maria Ress, the Filipino journalist who won the 2021 Nobel Peace Prize (so is unimpeachably virtuous): "There’s always been propaganda and there’s always been violence. What’s brought violence mainstream is social media. [The US Capitol attack on] January 6 is the perfect example: people wouldn’t have been able to find each other if social media didn’t cluster them together and isolate them to incite them further." I can still can't work out what she means by bringing violence mainstream, when she first concedes that there has always been violence. It's simply a meaningless statement. The claim that the Capitol riot could not have occured without social media is absurd, as is the implication that prior to the arrival of the technology there was no way of coordinating the far-right. Cadwalladr's wider purpose is to convince us that social media is a "polarisation engine" and that this arises from the use of algorithms that reinforce outrage. In describing the dynamic, she emphasises that the movement misinformation takes is from fringe platforms prefered by the far-right - such as Telegram, Bitchute and Parler - towards X and other "mainstream social media platforms". In other words, this is a plea not for the outlawing of those fringe platforms but for the creation of a firewall to protect the mainstream. 

Helpfully, Stephan Lewandowsky makes a similar point: "It’s a serious problem and is easily solved by modifying the algorithms so that they highlight information based on quality rather than outrage." In other words, we need better gate-keepers, the traditional role of the press. Again, there is a reluctance to admit that the majority of the misinformation that we have to deal with comes not from the periphery but from the centre. Amusingly, Cadwalladr notes that "the Daily Mail ran a shocked banner headline this week about a single suspicious account on X, with signs it may be based in Russia, spreading false information, although it is likely that this was only one very small part of the picture." Presented with evidence of mainstream misinformation she takes it simply as proof that the same paper was wrong to criticise her past work about Brexit misinformation ("investigations that were ignored or ridiculed by large sections of the British rightwing media").

One interesting quote in Cadwalladr's article, from Julia Ebner ("the leader of the Violent Extremism Lab at the Centre for the Study of Social Cohesion at Oxford University"), notes an affinity between the far-right and the medium: "It’s very, very similar across the world and in different countries with a rise in far-right politics. No other movement has been able to have their ideologies amplified in the same way. The far right is just really tapping into those really powerful emotions, in terms of algorithmically powerful emotions: anger, outrage, fear, even surprise." The first question to ask here is whether there really has been a rise in the far-right in recent years. The evidence does not suggest it. Much of what gets labelled far-right, e.g. Viktor Orban's "illiberal democracy", is simply conservative nationalism. In India, Modi and the BJP have suffered a setback. In France, Marine Le Pen is no closer to power and anyway her political trajectory has been towards the centre-right, competing with Emmanuel Macron to absorb Les Républicains. In the UK, the right is fragmented and the riots have managed to mobilise only small numbers of people (compare and contrast with the recent pro-Palestine marches).

It may be true to say that the far-right has found social media congenial, but this appears to be true of all political persuasions. To say "No other movement has been able to have their ideologies amplified in the same way" suggests that this ideology is near-hegemonic, but what exactly is it? A belief that asylum-seekers are thieves and rapists, that Islam is a death-cult, or that our statues are a risk from the intolerant left? Ideology seems a generous word to describe a set of prejudices whose prevalence extends well beyond the "far-right". In focusing on anger, outrage and fear, Ebner is echoing the moral foundations theory of Jonathan Haidt, a dubious attempt to justify conservative impulses (you're not racist you're just loyal) and suggest that liberals have blindspots (you don't acknowledge legitimate concerns). The idea that these are "algorithmically powerful emotions" is obviously a nonsense: algorithms don't recognise emotions, they're just shuffling data based on a dynamic taxonomy (you liked that so you might like this). 

The reaction of the British press, both liberal and conservative, to the Internet has always been driven primarily by its material interests. As search engines gobbled up advertising spends, the press went into a tailspin. It only levelled out to the extent that it was able to migrate online. In doing so, it realised that its own interests were better-served by viral contagion than paywalls. This has resulted in a deliberate expansion of the sort of content likely to earn clicks, and that has meant appealing to the emotions, not just anger and fear but envy, amusement and desire. The inevitable coarsening this has given rise to has been blamed on the medium. Where the press have led, other media have followed, hence the explosion of talk radio and now the arrival of partisan TV with GB News. Social media have provided not only the means to disseminate this calculatedly offensive opinion but also a source of original material: the opinions of ordinary people as much as celebrities that can be held up for public censure or ridicule.

In today's Observer, amidst Sonia Sodha insulting the BMA over puberty-blockers and Jane Martinson espying the patriarchy at work in the BBC's handling of Huw Edwards (neither story needed to be written but both will get the clicks), you will find Andrew Rawnsley opining on the riots. "Part of the answer to the violent far right will come from smart and proactive policing. Making the tech giants live up to their moral and legal responsibilities to the rest of society is another must. These are necessary steps, but they are not by themselves all that will be needed. The longer-term challenge for ministers is to find ways to drain the swamps of racism and conspiracism from which the far right recruit." I'm not aware that anyone is actually claiming that the tech giants have not lived up to their legal responsibilities, in contrast to parts of the press who we recently learnt may have destroyed compromising evidence. Likewise, it's hard to take calls for moral responsibility seriously while Leveson 2 remains sidelined. And as for draining the swamps of racism and conspiracism, surely the place to start would be with those wellsprings in the traditional media who keep the swamps watered.

The AI Hype Cycle

Gartner, the American technology research consultancy, launched its Hype Cycle in 1995 at a time when there was a lot of new hardware and software being punted towards its corporate clients. The original value of the cycle was that it allowed Gartner to visually represent the relative maturity of various technologies within the wider cycle of business adoption, not unlike a wine vintage chart (too early, drinkable now, past its best). As the somewhat cartoonish graphic indicates, this was pitched at non-technically-literate executives. The origin story told by Jackie Finn makes clear that the purpose was to advise on the timing of adoption, hence the research note in which the image was first published was entitled "When to Leap on the Hype Cycle". The absence of "whether" simply reflected that Gartner's business model was to sell consultancy to firms that were in adoption mode. This in turn reflected the times: not simply a point at which the World Wide Web was taking off but the latter stages of the great IT fit-out of the corporate world that commenced with the arrival of PCs and mini-computers (what would become known as "servers") in the 1980s.

The Hype Cycle quickly became a fixture not only for Gartner but in the culture of business and even wider society. But as it spread, an idea developed that this was actually the common life-cycle of emergent technologies: there would always be a period of hype followed by disillusion followed by the payoff of improved productivity. You can see the problem with this by considering some of the examples on that first graphic. Wireless communications never really had a peak of inflated expectations because the underlying technology (radio) was mature and the benefits already proven. Bluetooth was frankly a bit Ronseal. Handwriting recognition turned out to be a solution in search of a problem, hence it never made it to the plateau of productivity. Object-oriented programming (OOP) turned out just to be a style of programming. Insofar as there was hype, it resided in the promise of code re-use, modularity, inheritance etc, which were all crudely understood by executives to mean a shift from the artisan approach to the production line and thus fewer, lower-paid programmers. The explosion of the Internet and the consequent demand for programming killed that idea off.

It's difficult to say on which side of the peak of inflated expectations we are today in respect of AI - i.e. artifical intelligence but more specifically large language models (LLMs). If I can be forgiven for introducing another term, we may well have reached the capability plateau of AI some months ago, possibly even a couple of years ago. This is because we are running out of data with which to populate the models. All the good stuff has been captured already (Google and others have been on the case for decades now) and the quality of newer data, essentially our collective digital exhaust, is so poor that it is making generative AI applications dumber. The current vogue for throwing even more computing resources at the technology, which has boosted chip suppliers like Nvidia and led to worries that the draw on power will destroy all hope of meeting climate targets, reflects the belief that we can achieve some sort of exponential breakthrough - who knows, perhaps even the singularity - if we just work the data harder. 

The appearance of financial analyst reports suggesting that AI isn't worth the investment strongly suggests that the trough of disllusionment may be upon us, but for the optimistic this simply means we are closer to the slope of enlightenment. The classic real-world template for this was the dotcom stock boom in the late 90s, which was followed first by the bust in 2000 and then the steady, incremental improvement that culminated in the mid-00s with the iPhone and Android, Facebook and Twitter, and the first examples of cloud computing with Google Docs and Amazon Web Services. For those with a Schumpeterian worldview, the bust was simply the necessary stage to weed out the pointless and over-valued. The Internet eventually became pervasive, generating highly valuable businesses in the realms of hardware, software and services, because it met the consumer demand for killer apps: first email, word-processing and spreadsheets, and then social media, video and streaming. But what is the killer app for LLMs? An augmented search function, like ChatGPT or Microsoft's Copilot, is small potatoes, while the ability to create wacky images with DALL-E 3 is on a par with meme-generators in terms of value.

The wider promise of AI has been that it will replace the need for certain workers, notably "backoffice" staff whose knowledge is highly formalised and whose data manipulation can be handled by "intelligent agents" (to use the terminology, if not the meaning, of that original Gartner graphic). You may remember this idea from the history of OOP. Indeed, you may remember it from pretty much every technology ever applied to industry, starting with the power looms used to depress wages that the Luddites railed against. The standard story is that new technologies do destroy jobs but they create other, even better jobs in turn by freeing up labour for more cognitively demanding (and rewarding) tasks. Of course, from the perspective of an individual business this isn't the case. The promise of greater productivity is predicated on either reducing labour or increasing output. The new jobs will be created elsewhere and are thus someone else's problem. This is a good example of the difference between microeconomics (the rational choice of a single firm to replace staff with technology) and macroeconomics (the impact on aggregate levels of employment and effective demand in the economy as a whole). The story of job substitution is true, as far as it goes, but it is also obviously a consolation: there is no guarantee that the new jobs will actually be better.

The dirty secret of AI is that it requires an ever-growing army of human "editors" (to dignify them with a title that does not reflect their paltry pay and poor working conditions) to maintain the data used in the LLMs. These are mostly "labellers" or "taggers", and mostly employed in the global south. The fact that the jobs are done at all tells you that they must be sufficiently attractive in local terms. In other words, spending all day tagging pictures of cars is probably better paid than tending goats. This means that the anticipated benefits of AI may simply be a futher round of the offshoring familiar from manufacturing in the 1980s and 90s, but this time with AI acting as a veil that makes the human reality even more obscure than sweatshops in Dhaka or Shenzen. Indeed, the veil may be the point for many AI boosters in the technology industry: a way of preserving their idealised vision of a tech-augmented humanity that isn't shared by many beyond their own limited social milieu and geography. Inevitably the boosters also include many who have zero understanding of any technology, like Tony Blair, but who are very keen on the idea of a dehumanised workforce and a disciplined polity.

AI serves as a massive distraction for technocratic neoliberalism. It offers a form of salvation for all the disappointments of the last few decades: secular stagnation is averted, productivity growth picks up, truculent labour is made docile. After all the waffle over the last three decades about the knowledge economy - the need to raise our skills to take advantage of globalisation - it is notable that the promise of better jobs has been downgraded. In a recent "report" by the IPPR think-tank, we are told that "Deployment of AI could also free up labour to fill gaps related to unaddressed social needs. For instance, workers could be re-allocated to social care and mental health services which are currently under-resourced." From spreadsheets to bed-pans. The lack of resource for health and social care is simply a matter of money, i.e political choice. There is no suggestion that AI-powered businesses will be paying a higher rate of tax, rather the implication is that AI will fuel growth that will increase revenue in aggregate (again, macroeconomics provides the consolation for microeconomics). 

One of the key dynamics of postwar social democracy was the idea that public services depended on a well-paid workforce paying tax. The value created in the economy was funneled via pay packets and PAYE into the NHS and elsewhere. Neoliberalism broke this model by offshoring and casualising labour. Workers (i.e. average earners) funnel less value to the state, which leads both to greater government efforts to raise revenues elsewhere and to pressure to cut public spending ("cut your coat according to your cloth"). An AI revolution that reduces the number of average paying jobs (those "backoffice" roles) and substitutes more lower paid "caring" roles will simply further increase the pressure on the public sector. I suspect that the deployment of AI will not be marked by a slope of enlightenment, let alone a plateau of productivity, but by the determination of the private sector to reduce labour costs and by the determination of the state to trim the public sector as tax revenues decline. It is often comically bad, but in combination with offshored digital peons, AI technology is probably already "good enough" to meet those ends. In terms of the hype cycle, we are probably a lot further along than we imagine.

Computer Says No

Artificial Intelligence (AI) is a misleading term because we don't have a full understanding of human intelligence yet, so the comparison is necessarily imprecise.  The Turing Test, aka the imitation game, isn't simply about a machine mimicing human responses - which is easy enough to do - but about our ability to reliably identify the expressions of human intelligence in comparison to software-generated text. The uncertainty as to whether it is a human or a machine on the other side of the screen reflects our inability to infallibly spot the human as much as our inability to unmask a computer-generated mimic. This uncertainty is obvious when we consider bureaucracy. Long before the possibility of LLMs, it was common for people to complain that interacting with commercial organisations or the state was like dealing with a machine. Bureaucratic procedures seemed designed to excise all humanity from what were notionally social interactions. As a result, bureaucracy was routinely derided as stupid, even though the manifestations of this stupidity were deliberately designed and presumably satisfactory to the designers.

The Large Language Model (LLM) approach to AI assumes that human-like intelligence can be derived from enough text. But there is a problem with this that is becoming increasingly apparent. Improving the model means expanding the corpus, but that in turn means more rubbish, which humanity routinely produces in text form. Many think that we may already have reached a limit and that futher expansion of LLMs will result not simply in diminishing returns but the passing of an inflexion point after which the models become dumber and dumber. The underlying issue here is not that factual errors in the corpus can give rise to "hallucinations" but that text is not a uniform expression of a general human intelligence. Rather text is a highly formalised set of what we might call genres. These are broader than knowledge domains, e.g. technical dialects, and reflect more generic purposes for which text is used, such as education or reportage. For example, the text produced by bureaucracy has well-known characteristics. It can be ambiguous, sometimes impenetrable and even downright nonsensical, but these are not necessarily failings from the perspective of the authors.

Henry Mance in the Financial Times recently noted how the problem of contamination is increasingly framed as one of reputation: "[the cognitive scientist] Gary Marcus suggests performance may get worse: LLMs produce untrustworthy output, which is then sucked back into other LLMs. The models become permanently contaminated. Scientific journals’ peer-review processes will be overwhelmed, “leading to a precipitous drop in reputation”, Marcus wrote recently." Reputation is an interesting word to choose in this context. It doesn't just suggest predictability or reliability - the idea that you will get the "right" answer. It also suggests that the answer is definitionally true because it is the answer given by authority. But this is a mundane truth rather than ex cathedra. According to Mance, "AI will become embedded in lots of behind-the-scenes tools that we take for granted." Again, the phrase "taking for granted" suggests that AI will advance to the point where its output is accepted as authoritative even if trivial. This doesn't assume that the AI will never be wrong, that it will be infallible, but that the rate of error will be low enough to be tolerable, much as bureaucratic mistakes are.

The future of AI may turn out to be restricted language models rather than the largest possible. Much of what is described as "AI training" is actually human intervention to limit the interpretative scope of the software: to rein it in. This will help address the contamination issue, essentially through brute force quality control, but it will also allow the AI to operate within a narrower semantic field where the epistemological rules are rigorously observed. In other words, just like a bureaucracy. This is intelligence in a very narrow, dry and unimaginative form. And that points to a rather depressing future. While there may be exciting applications of the technology in sexy areas like medical scanning and diagnosis, the big returns have always been anticipated in administrative and service functions, hence the predictions for "lost" jobs tend to focus on accountancy, customer support and the like. AI will probably thrive best in areas where rigidity of thought and a strictly bounded intelligence, even an unyielding monomania, is prized. 

Technological development reflects above all the appetite and capability of the socio-economic environment to exploit new techniques (or old ones rediscovered). And that in turn may be determined by the longevity and ubiquity of previous technologies. Famously, the Chinese writing system - tens of thousands of morphemes - led to the dominance of woodblock printing and the relative underutilisation of movable type until the mid-19th century. But the latter technology produced a revolution when combined with the Latin alphabet in 15th century Europe. To give a contemporary example, modern software is riddled with skeuomorphs, from calendar apps that mimic desk diaries to the shutter-click sound of a smartphone camera app. The visual and aural prompts are intuitive only to the extent that we have been trained to recognise their forbears. If camera apps mimiced the "poof" of a flashlight powder explosion, rather than a click, we'd still understand it perfectly well, despite few of us ever directly experiencing a magnesium flare.

We may not be able to create a genuine artificial intelligence - i.e. artificial in the sense that it convincingly mimics the human sort - because we cannot escape the constraints that we place on human intelligence. The great myth, shared by liberals and libertarians (though not echt conservatives), is that human genius is unbounded. In reality, it is inescapably situated in history and society. In theory, globalisation and modern mass media should mean that new ideas spread rapidly and pervasively, but you'd have to be naive to imagine that there are no technologies underappreciated or lying dormant in the modern world. AI hasn't come to the fore because it is our shining hope (though it's worth noting that it has quickly acquired the near horizon of expectation characteristic of fusion power), but because it seems already familiar, all too familiar, in its combination of impressive authority and crass stupidity. AI will advance largely through the realms of business administration and public services, and so it will inevitably inherit the cultures (i.e. the vocabulary and semantics) of those realms. AI will not be a Culture Mind, of the sort imagined by Iain M. Banks, but a faceless version of the DHSS circa 1983.

AI, Comparative Advantage and Natality

Beyond the banality of stochastic parrots and their "hallucinations", AI exists as a socio-political thought-exercise: a what-if. As has been the way since the emergence of sociology and its creative cousin Science Fiction, this encompasses both dystopian hell and utopian heaven. Perhaps the most obvious combination of the two is the idea that AI will take all the jobs, delivering either something akin to The Matrix, where humanity is reduced to its utility as fuel, or to a world of leisure and ease in which we can all pursue our talents and interests: "to hunt in the morning, fish in the afternoon, rear cattle in the evening, criticise after dinner, just as I have a mind, without ever becoming hunter, fisherman, herdsman or critic." We have now reached the stage where official projections for growth routinely factor in AI as a magic ingredient. But beyond the glorified press releases of business consultancies predicting that AI will "impact" half of all jobs (consider what an accurate prediction of the impact of electricity would have been), there hasn't been much discussion of the mechanisms. In other words, how in practice will AI spread though the economy and how will employment patterns respond?

This is odd insofar as we have no shortage of historical data on the way previous technologies were deployed and how they reconfigured society. We all understand how the combustion engine substituted for horses and how that resulted in grooms being replaced by mechanics, and more recently we have seen how the technology of logistics has allowed employment in developed economies to shift from the primary and secondary sectors to the tertiary. The optimistic take on AI is that we'll see something similar: old jobs being replaced by new ones and aggregate wealth increasing, which translates to higher wages and greater purchasing power. If there is a fly in this ointment, it relates to the relative narrowing of wages between different parts of the world as industry reconfigures optimally. The middle class expands in the Far East while its peers in the American Mid-West stagnate, but at a global level there is aggregate growth. AI probably won't have a differential impact around geography, as raw material extraction and manufacturing does, but it will have an impact around cognitive differentials, which is why the more pessimistic prediction is for a decline in whitecollar employment.

Noah Smith is to be found on the optimistic side of the debate (as usual), but while his just-so stories of neoliberal progress can grind your teeth, he has made a useful contribution by trying to explain how the mechanism might work during the transition. He starts by outlining the negative view: "humans will have nothing left to do, and we will become obsolete like horses. Human wages will drop below subsistence level, and the only way they’ll survive is on welfare, paid by the rich people who own all the AIs that do all the valuable work. But even long before we get to that final dystopia, this line of thinking predicts that human wages will drop quite a lot, since AI will squeeze human workers into a rapidly shrinking set of useful tasks." In answering this, Smith's core point is that AI is not limitless. It will be constrained by computing power and energy - i.e. material resources. This will raise the opportunity cost of using it for low-value tasks that could be done by humans. Relative opportunity cost means that it will still make sense for humans to do jobs and be well-paid for them while AI concentrates on the really important stuff.

One of the features of economics (which proves that it is a social science rather than a hard science) is that many of its theories cannot be proven. This is not simply about the crisis of replicability, which typically affects microeconomics, but the difficulty of conducting real-world empirical trials at the macroeconomic level (hence the delusion that micro-foundations can be used to extrapolate macro policy). However, there are a few theories that are demonstrably true because the global economy provides a reliable test environment. One obvious example is the gravity theory of trade, which posits that it is cheaper to trade with near neighbours than far-off countries due to relative transportation costs. Despite the best efforts of Brexiteers to claim that technology has abolished distance, this clearly still holds true. The relevant theory for Smith's intuition about AI is comparative advantage, which is also demonstrably true for reasons to do with differential endowments - e.g. it makes more sense to grow bananas in the Caribbean and oats in Scotland than vice versa, and likewise someone with a high IQ and someone with lots of muscle power will gravitate to different jobs best suited to their abilities. So AI can concentrate on curing cancer rather than trying to win the Nobel Prize for Literature. 

One issue with Smith's model is that AI will produce greater rates of growth in those areas that it addresses. This is partly due to the compounding effect of the technology itself, but also because the movement of humanity into services that cannot be cheaply automated will necessarily intensify the unbalanced growth between the two sectors. William Baumol noted the effect by which low-productivity sector wages rose because of the competition for labour by high-productivity sectors. But if the AI sector isn't competing with the human sector for labour, there is no reason to think that wages will be bidded up. In other words, AI may not make humans redundant but it may lead to further wage stagnation because of that unbalanced growth. The best argument against this is that some of the fruits of AI must be recycled into wages simply to keep the system from collapsing (Smith envisages this in the perjorative terms of "welfare", i.e. UBI, though this is functionally no different to in-work benefits or a minimum wage), but that then emphasises the issue of distribution which comparative advantage does not circumvent. That gal with the high IQ is likely earning more than the guy with the muscles because there are more people with the latter than the former so basic supply and demand leads to different wages.

So perhaps the dystopia of AI is not that we all lose our jobs but that the economy becomes even more unequal in its outcomes. The jobs that AI won't do will include both grunt work and highly-valued work, but at the aggregate level of the economy we may end with many more of the former relative to the latter than was previously the case. Another way of thinking about this is to note that wages (returns to labour) have declined while asset wealth (returns to capital) has increased since the 1970s without the input of AI. We've had plenty of wage stagnation, particularly since 2008. If there is a more fundamental and powerful force at work (let's call it neoliberalism) the question then becomes how will that force accommodate AI? It's a reasonable assumption that it will reinforce or even exacerbate the existing trend towards wealth inequality and wage stagnation. In theory AI could disrupt this trend, but then any number of technological breakthroughs since the 70s that were characterised as disruptive turned out to actually reinforce neoliberal political economy (that is in the real world version, rather than the textbook fantasy, e.g. the growth of monopoly), so there's little reason to think this episode will be any different. 

The error is to assume that AI will carry all before it, reconfiguring society in its own image, hence the hysterical colouring of much commentary, but technologies are moulded by society as much as they do the moulding. Marx noted that "The hand-mill gives you society with the feudal lord; the steam-mill society with the industrial capitalist", but the reality, as Karl Polanyi countered, is that society also gave us the Factory Acts. If AI really were disruptive of the fundamentals of the economic system, the counter-movement would focus on the protection of the key factors of production, such as land, labour and capital, but to date the focus has been on the enforcement of propriety around generative deepfakes and the threat to the traditional media's interpretation of truth. If this carries an echo of the counter-movement of the nineteenth century, it is more of Christian revivalism than social progressivism. What this suggests is that AI won't be anywhere near as impactful as either the optimists or pessimists predict, but it also suggests that its greatest impact may be to exacerbate and accelerate existing trends.

One of those trends, which can be directly linked to the way that neoliberalism has expanded the logic of the market into the sphere of the family, is falling birth rates. As Steve Randy Waldman notes, natality is one area where comparative advantage has to cede to human emotion. Though much of child-rearing is handed over to specialists, most obviously teachers, it remains essentially a form of "cottage production" that struggles under the neoliberal logic of competition: "The relationship between wealth and natality is nuanced. When wealth is certain, its increase is likely pronatal, as people can bank on greater resources to cover the burdens of childrearing. But when wealth is uncertain, when it is delivered via tournaments that deliver outsize rewards to winners, then increases in 'expected' (meaning average) wealth likely translate to decreases in natality. The bigger the prize, the greater the cost of anything that will reduce your chance of winning." There should be no surprise that the current iteration of the Californian Ideology emphasises both the dramatic potential of AI and the necessity of pronatalism.

My guess is that AI won't become a general purpose technology (GPT) on a par with electricity or the Internet. This is because it will be too expensive, and that in turn is because there is no upper limit to its ambition, even if there are hard limits in the form of silicon and energy. You only need so much power and bandwith for most tasks, so GPTs like electricity and datacoms only need to be good enough. We already have good-enough AI, but it's fruits are underwhelming. Consequently, AI resources will increasingly be focused in specialist areas where its potential is greatest, e.g. more medical imaging rather than chasing the dream of fully autonomous vehicles. Noah Smith is right in his emphasis on comparative advantage, but as a neoliberal he assumes that the invisible hand of the market will optimally decide on how to allocate those scarce AI resources, rather than it being a political decision that will be taken to reinforce neoliberalism itself. Wealth inequality will continue to grow, even if AI companies replace oil corporations and technology manufacturers on the stock exchange, and the symptoms of that inequality, notably falling birth rates, will continue to elicit angst and ineffective amelioration.

Beyond the Horizon

Most English Premier League football clubs have now introduced digital passes for ticket-holders. There have been teething problems, but the technology is clearly here to stay: a bit like VAR. There are a number of reasons for this, from clamping down on ticket-touting (though the touts are still to be seen - I have no idea how they do it now) to better crowd safety (the long legacy of Hillsborough and the Taylor Report), but another driver for clubs' investment in ticketing technology since the 1980s was fraud at the margin - i.e. gate operators letting people under or over the turnstile, so the counter ratchet didn't click, and pocketing the money. To an extent this had always been tolerated as the cost of prevention was high and the loss to the club marginal, but it became a much bigger issue as entrance prices went up and clubs became more dependent on their matchday income. While wealthy foreign owners have driven inflation in the game since the millennium, financial fair play rules have simply reinforced the need to squeeze every last penny from the gate.

The reason I raise this is that it provides a useful entry-point for discussing what has come to be known as the Post Office scandal. There are two parts to this: the bug-ridden Horizon IT system and the response of the Post Office management to the unfolding debacle. Both were informed by a suspicion that the organisation was being defrauded at the margin, though it's central to the sorry tale that this started out as a belief that benefit claimants were defrauding the state, not just sub-postmasters and sub-postmistresses defrauding Post Office Limited. It's also worth emphasising at this point that, with the exception of a handful of Crown Post Offices, those sub-postmasters are independent franchisees - i.e. small business people. This is important because it highlights a discrepancy between the positive rhetoric around self-employment and sub-contractors and the harsh reality of asymmetric commercial relations (it's also worth noting in passing that there are a lot of pseudo-SME arrangements in the public sector, e.g. doctors, as well as in the grey areas of public corporations, e.g. the BBC, and state owned commercial entities such as the Post Office).

While the now-famous (and likely to be awards-laden) ITV drama, Mr Bates vs The Post Office, focused on the human interest of the present, and while much of the subsequent outrage has focused on the perversity of the law and the recent underfunding of the courts, necessitating an act of Parliament to unpick the mess, to do the story justice (sic) you really need to look at the longer history. It might appear unnecessary to start with the origins of the General Post Office in 1635, but we should certainly go back as far as 1838 and the introduction of money orders, which is when the opportunities for fraud at the margin start to proliferate. The need for strict financial controls was reinforced by the creation of the Post Office Savings Bank in 1861 and the gradual expansion of state financial services culminating in the payment of old age pensions in 1909. The GPO has been investigating fraud at the margin for over 300 years, but this has been reported in the context of the scandal as evidence that it had become a law unto itself: that it shouldn't have been pursuing private prosecutions outside the purview of the Crown Prosecution Service. 

But this misses the crucial point that the Post Office has a long data series - a corporate memory, if you will - on fraud at the margin. Why did no one notice the statistical anomaly of a significant rise in fraud prosecutions, estimated to be from an average of 5 a year to 55, after the introduction of the Horizon system? The likely answer is that Post Office management suspected the higher level of fraud was always there but that only now, with the benefit of computer-based accounting, were they able to uncover it. It's also worth noting here the downward secular trend in Post Office revenues and the pressure this exerted on management to reduce losses and thus reliance on state subsidies. What's less understandable is why government ministers weren't (as far as I can tell) asking the obvious question, i.e. is this increase in cases statistically credible?, particularly when you consider how wedded both New Labour and the subsequent coalition were to targets and metrics as part of the culture of New Public Management (NPM). The suspicion must be that politicians who spout about "business rigour" and "evidence-based policy" are mostly clueless when it comes to basic data analysis.

The second historical strand worth looking at is the origin and development of ICL. The business was created under the second Wilson government in 1968 and overseen by Tony Benn, the then Minister for Technology. It was an example of an industry "champion" (what would later be derided as "picking winners") formed by the merger of three UK computer fims. The state would have a 10% stake and would provide funding for research and development. The aim was to build a domestic competitor to IBM that could service both the UK public and private sectors and also develop an export market. Unfortunately, it started out with a 6-bit architecture instead of the by then standard 8-bit byte, as used by IBM and others, making its products a technological dead-end. It ended up marketing Fujitsu mainframes and minis, which were IBM clones. That gave the Japanese a route into the UK public sector, which they consolidated by buying a majority stake in ICL in 1990. The point to note here is that unlike IBM, which reinvented itself around software and services, ICL and Fujitsu were always chiefly hardware businesses. 

The primary political driver for the creation of the Horizon system was the plan, developed under the Major administration, to introduce magnetic swipe-cards for the payment of benefits, which it was thought would reduce fraud. The conversion of other Post Office counter business from a largely paper-based system to an electronic point of sale (EPOS) system was a secondary political consideration, though clearly from the perspective of the Post Office itself this was the commercial priority. The project was an early example of a private finance initiative (PFI), with the developer to recoup their costs and profit via transaction fees. ICL Pathway Limited was set up with the explicit purpose of winning the bid for Fujitsu, which it did in 1996. By 1999, with Labour now in government, the project was already a mess, not least because of doubts about the long-term viability of swipe-cards (chip-and-pin, common in Europe in the 1990s, arrived in the UK in 2003). As a result, the benefits element was ditched and the project scope reduced to the counter business.

Instead of going back to the drawing board, ICL Pathway decided to simply build on top of the already flaky prototype. It is clear from evidence already presented in public inquiries (here, here and here) that Horizon was a badly-run project, with a poor architectural design, that tried to integrate with new third-party technology as it went (e.g. Windows, Oracle, SAP and wide-area networking). There appears to have been no consistent development methodology, no automated test framework, and poor bug management and version control. The smell it gives off is of a second-rate mainframe-centric business circa the mid-80s, which isn't surprising. What this in turn highlights is that there was inadequate IT competence at a senior level within the Post Office, which could have independently assessed ICL Pathway, and equally little technical nous within govenment, despite the public technophilia of the likes of Tony Blair. What is also clear is that the then Trade and Industry Secretary, Peter Mandelson, was pushing ICL Pathway to meet deadlines when the correct course of action would have been to abort the project, re-specify the requirements, and start a fresh tender process on a fixed-cost plus margin basis.

There have been multiple TV dramas about Hillsborough and Grenfell, and yet justice has not been delivered in either case and the prospect of special acts of parliament to deliver it are negligible (though it should be said that the proposed Post Office act may never see light of day given the issues of principle it raises). The difference between those tragedies and the Post Office scandal is that they highlighted the negligence of the state towards ordinary people: the contempt of South Yorkshire police towards football fans in the one, and the disdain of a Conservative borough council towards poorer residents in the other. At Hillsborough, there was no appetite in government to criticise a force that had been central to the defeat of the miners five years earlier. As Thatcher herself said in response to the Taylor report, "The broad thrust is devastating criticism of the police. Is that for us to welcome?" Likewise, in the shadow of Grenfell Tower there was little desire to criticise a borough for cheeseparing when that had been the order of the day in local government for decades and when building safety had been presented as a contraint on enterprise ripe for deregulation. 

Ironically, had the Post Office remained fully in the public sector - i.e. as a public corporation rather than as a state-owned private company - the scandal could have dragged on even longer as there might have been greater political ramifications, but equally it might not have happened in the first place as the imperatives - that combination of a bad technology choice and government pressure to be a PFI success - could have been lacking. I don't imagine post offices would still be run on paper records, but a more thoughtful project might have decided to simply adopt a proven, off-the-shelf EPOS system with standard back-end ERP integration (post offices are just retail outlets, after all). The root problem was that initial demand to handle benefit payments, which in turn arose from the delusion that fraud at the margin could be reduced by a large enough figure to satisfy the critics of welfare. The compounding factor was the insistence of ministers that Post Office Limited should operate at arms-length from the state while being put under pressure by its owner (those same ministers) to be a commercial success.

The foot-dragging by the Post Office after the initial evidence that sub-postmasters were being wrongly convicted was not just the usual arse-covering or desire to minimise costs, it also reflected the history of petty fraud within the business and the fear that a general amnesty and blanket compensation might benefit the guilty as well as the innocent. Over-and-above bad management and poor IT practice, the scandal highlights the gulf that exists between the theory of the small business (incentivised, hard-working, responsible) and the reality (muddling through, sometimes incompetent, dodgy at the margins). What's depressing about the otherwise laudable TV drama and the public response to it is the idea that these people must have been innocent precisely because they were small business operators: that cuddly Toby Jones. In other words, they have been given the benefit of the doubt (albeit there are limits) in a way that the victims of Hillsborough and Grenfell never were.