danmcquillan.org

AI as Algorithmic Thatcherism

2023-12-21T00:00:00+00:00

It's tempting to see the recent UK AI Safety Summit as a damp squib, preempted by an Executive Order on AI from the Whitehouse and roundly criticised by civil society for excluding everyone but tech execs. Unfortunately, none of the current debate gets to the heart of the matter: AI is already a flop, and we are being hoodwinked by a mixture of corporate and ideological agendas that will wreck public services and deepen social divisions.

AI can turn some impressive party tricks, but it's unsuited for solving serious problems in the real world. This is true of predictive AI, whose correlations are data-driven conspiracy theories, and of large language models like ChatGPT, whose plausible waffle is always trying to pull free of the facts. The real issue is not only that AI doesn't work as advertised, but the impact it will have before this becomes painfully obvious to everyone. AI is being used as form of 'shock doctrine', where the sense of urgency generated by an allegedly world-transforming technology is used as an opportunity to transform social systems without democratic debate.

Faced with social structures whose foundations have been eaten away by decades of privatisation and austerity, the political response is to pump money into 'frontier AI' while hyping it up as the most awe-inspiring technology since the Manhattan Project. The Prime Minister says he will "harness the incredible potential of AI to transform our hospitals and schools" while ignoring leaking roofs in the NHS and the literally collapsing ceilings in local schools. This focus on the immaterial fantasies of AI is a deliberate diversion. When large language models are touted as passing basic medical exams, it's because they've absorbed answers from across the internet. They are incapable of the embodied understanding and common sense that underpin medicine, education or any other form of care.

One thing that these models definitely do, though, is transfer control to large corporations. The amount of computing power and data required is so incomprehensibly vast that very few companies in the world have the wherewithal to train them. To promote large language models anywhere is privatisation by the back door. The evidence so far suggests that this will be accompanied by extensive job losses, as employers take AI's shoddy emulation of real tasks as an excuse to trim their workforce. The goal isn't to "support" teachers and healthcare workers but to plug the gaps with AI instead of with the desperately needed staff and resources.

Real AI isn't sci-fi but the precaritisation of jobs, the continued privatisation of everything and the erasure of actual social relations. AI is Thatcherism in computational form. Like Thatcher herself, real world AI boosts bureaucratic cruelty towards the most vulnerable. Case after case, from Australia to the Netherlands, has proven that unleashing machine learning in welfare systems amplifies injustice and the punishment of the poor. AI doesn't provide insights as it's just a giant statistical guessing game. What it does do is amplify thoughtlessness, a lack of care, and a distancing from actual consequences. The logics of ranking and superiority are buried deep in the make up of artificial intelligence; married to populist politics, it becomes another vector for deciding who is disposable.

But what about all the potential 'AI for good' - should we abandon all that hope just because AI has this dark side? The problem with the promised bounty of AI is that, like a mirage, it starts to fade from view the closer you get. The claimed generalisation from computation to the shifting complexity of our lived experience never seems to quite stack up. What comes into focus instead is AI's material dependencies. Thanks to its insatiable appetite for data, current AI is uneconomic without an outsourced global workforce to label the data and expunge the toxic bits, all for a few dollars a day. Like the fast fashion industry, AI is underpinned by sweatshop labour. Above all, AI a very physical technology: it consists of vast server and data centres, packed with computers that burn energy and generate heat. These hyperscale warehouses suck up vast quantities of cooling water, depleting whatever communities and ecologies are unlucky enough to host them.

Shouldn't we be resisting this gigantic, carbon emitting version of automated Thatcherism before it's allowed to trash our remaining public services? It might be tempting to wait for a Labour victory at the next election; after all, they claim to back workplace protections and the social contract. Unfortunately they aren't likely to restrain AI; if anything, the opposite. Under the malign influence of true believers like the Tony Blair Institute, whose vision for AI is a kind of global technocratic regime change, Labour is putting its weight behind AI as an engine of regeneration. It looks like stopping the megamachine is going to be down to ordinary workers and communities. Where is Ned Ludd when you need him?

EU AI Act briefing

2023-12-09T00:00:00+00:00

Some quick notes on the EU's AI Act. These were written the morning after so I might update them as more details emerge.

The whole thing is premised on a risk-based approach(¹)
This is a departure from GDPR, which is rights-based with actionable rights
Therefore it's a huge victory for industry(²)
It's basically a product safety regulation that regulates putting AI on the market
The intention is to promote the uptake of AI without restraining 'innovation'(³)
Any actual red lines were dumped a long time ago
The 'negotiation theatre' was based on how to regulate gen AI ('foundation models') and on national security carve-outs
People focusing on foundation models were the usual AI suspects
People pushing back on biometrics etc were civil society & rights groups
The weird references in the reports to numbers like '10~23' refer to the classification of large models based on flops(⁴)
Most of the contents of the Act amount to some form of self-regulation, with added EU bureaucracy on top(⁵)

Ironically, an epistemology of 'risk' is one of the key things that makes predictive AI so harmful ↩
h/t to Daniel Leufer for clarity on this ↩
Innovation as in 'shock doctrine' ↩
No doubt swapping these numbers makes eurocrats feel cool & nerdy ↩
IMO the EU itself, as an institution that constructs the conditions for mass deaths at its borders, is itself a 'high risk model' ↩

Evidence to House of Lords inquiry into Large language models

2023-09-01T00:00:00+01:00

My evidence to the House of Lords Communications and Digital Select Committee inquiry into Large language models is available on the parliament.uk website.

Executive summary

Large language models contain foundational flaws which mean they are unable to live up to the hype and make it likely that the current bubble will burst. They will continue to require vast amounts of invisibilised labour to produce, but will not result in any form of artificial general intelligence (AGI).
The greatest risk is that large language models act as a form of ‘shock doctrine’, where the sense of world-changing urgency that accompanies them is used to transform social systems without democratic debate.
The AI White Paper promotes populist narratives about AI adoption that align with the hype around large language models while offering a fairly thin evidence base. Ongoing developments in UK policy, such as the upcoming summit, cite notions of existential threat while ignoring the more mundane risks of social and environmental harms.
The narrative around open source AI is a complete red herring. The way ‘open’ can be applied to large language models doesn’t level the playing field, make the models more secure or challenge the centralisation of control.
UK regulators are not well placed to address the issues raised by large language models because these systems operate across sectors and technical, economic and social registers while establishing unpredictable feedback loops between them. Meanwhile the AI industry is already engaged in significant lobbying at the EU which has proven sufficient to dissolve regulatory red lines.
Additional options for regulation draw on frameworks like post-normal science to mandate an extended peer community and the inclusion of previously marginalised perspectives. This more grounded approach has a better chance of resulting in AI that is more socially productive, where regulators are supported by distributed and adaptive ‘councils on AI’.

The political intervention in AI that we need right now...

2023-06-28T07:00:00+01:00

Opening remarks for the panel on 'Political Interventions in Data and AI' at #DataJustice2023

The political intervention that we need right now

is a social movement to resist AI, because a real challenge to algorithmic violence requires structural change.

AI isn't sci-fi but a radical continuity of modernity, of bureaucracy, of austerity; of the anti-worker, anti-poor contempt that stretches from Charles Babbage to Jeff Bezos. Regulation and reform are undermined by the absence of a fair status quo, but also by applying half-solutions to the entanglement of the tech and the social.

Social and technical structures are inseparable and should be understood as a whole, as an apparatus, because our concepts, our relationalities and even our subjectivities are materially constituted under specific technical arrangements. Under current conditions this apparatus is increasingly necropolitical, and AI will become part of the governance of who can live and who will be allowed to die.

We're already seeing the return of eugenics everywhere, from educational genomics to Silicon Valley's transhumanism and long termism. Meanwhile the EU, the leading light of AI regulation, consciously constructs the conditions for mass drownings in the Mediterranean.

AI itself amplifies the climate crisis through emissions, the expropriation of water and energy resources, the legitimation of green tech solutionism, and through its colonial universalism. What's missing is a social movement to resist AI that has a positive vision of more-than-human solidarities.

We need a prefigurative technopolitics; iterative interventions in both material operations and social relations that align with the world we want to live in. Like the Lucas Plan of the 1970s, where workers in a giant arms company prototyped alternatives from wind generators to hybrid vehicles, while radically democratising their workplace at the same time.

We need a movement that is Luddite in its commitment to put down machinery hurtful to the commonality, and convivial in its pursuit of a lost cybernetics that can balance autonomy and coordination. It's time to move from the dystopias of cyberpunk to the possibilities of solarpunk; to a vision of limited but creative tech in the service of social production and radical inclusion.

We come to bury ChatGPT, not to praise it.

2023-02-06T07:00:00+00:00

Large language models (LLMs) like the GPT family learn the statistical structure of language by optimising their ability to predict missing words in sentences (as in 'The cat sat on the [BLANK]'). Despite the impressive technical ju-jitsu of transformer models and the billions of parameters they learn, it's still a computational guessing game. ChatGPT is, in technical terms, a 'bullshit generator'. If a generated sentence makes sense to you, the reader, it means the mathematical model has made sufficiently good guess to pass your sense-making filter. The language model has no idea what it's talking about because it has no idea about anything at all. It's more of a bullshitter than the most egregious egoist you'll ever meet, producing baseless assertions with unfailing confidence because that's what it's designed to do. It's a bonus for the parent corporation when journalists and academics respond by generating acres of breathless coverage, which works as PR even when expressing concerns about the end of human creativity.

Unsuspecting users who've been conditioned on Siri and Alexa assume that the smooth talking ChatGPT is somehow tapping into reliable sources of knowledge, but it can only draw on the (admittedly vast) proportion of the internet it ingested at training time. Try asking Google's BERT model about Covid or ChatGPT about the latest Russian attacks on Ukraine. Ironically, these models are unable to cite their own sources, even in instances where it's obvious they're plagiarising their training data. The nature of ChatGPT as a bullshit generator makes it harmful, and it becomes more harmful the more optimised it becomes. If it produces plausible articles or computer code it means the inevitable hallucinations are becoming harder to spot. If a language model suckers us into trusting it then it has succeeded in becoming the industry's holy grail of 'trustworthy AI'; the problem is, trusting any form of machine learning is what leads to a single mother having their front door kicked open by social security officials because a predictive algorithm has fingered them as a probable fraudster, alongside many other instances of algorithmic violence.

Of course, the makers of GPT learned by experience that an untended LLM will tend to spew Islamophobia or other hatespeech in addition to talking nonsense. The technical addition in ChatGPT is known as Reinforcement Learning from Human Feedback (RHLF). While the whole point of an LLM is that the training data set is too huge for human labelling, a small subset of curated data is used to build a monitoring system which attempts to constrain output against criteria for relevance and non-toxicity. It can't change the fact that the underlying language patterns were learned from the raw internet, including all the ravings and conspiracy theories. While RLHF makes for a better brand of bullshit, it doesn't take too much ingenuity in user prompting to reveal the bile that can lie beneath. The more plausible ChatGPT becomes, the more it recapitulates the pseudo-authoritative rationalisations of race science. It also shows that despite the boast that LLMs are largely self-training, any real world system will require precaritised 'ghost work' to maintain its plausibility. It turns out that AI is not sci-fi but a techologised intensification of existing relations of labour and power. The $2/hour paid to outsourced workers in Kenya so they could be "tortured" by having to tag obscene material for removal is figurative of the invisible and gendered labour of care that always already holds up our existing systems of business and government.

As with the rest of AI, the dangers of ChatGPT go far deeper than bias and discrimination. Despite evidence that the model's powers of 'reasoning' are shallow heuristics based on the frequency of associations in the training data (meaning, as an illustrative example, that it's good at answering 'What is 24 x 18?' and poor at answering 'What is 23 x 18?') there are many in the AI community who insist on imputing emergent properties of reasoning and insight to ChatGPT. Its parent company, OpenAI, was set up "to ensure that artificial general intelligence benefits all of humanity", where 'artificial general intelligence' (AGI) is the insider term used for human-like intelligence that goes beyond narrow AI like facial recognition or self-driving cars. However, as I spell out in my book, the concept of AGI is inseparable from the kind of hierarchy of intelligence that has underpinned ideas of innate supremacy since the days of empire and colonialism. Hardly surprising, then, that the same Silicon Valley cultures that incubate enthusiasm for ChatGPT as emergent AGI also show allegiance to associated world views like Long Termism, where the immediate vulnerability of millions of ordinary people counts as nothing in relation to the prospects of a future space-faring super race.

In the mean time, OpenAI is acquiring billions of dollars of investment on the back of the ChatGPT hype. The point here is not only the pocketing of a pyramid-scale payoff but the reasons why institutions and governments are prepared to invest so much in these technologies. For these players, the seductive vision isn't real AI (whatever that is) but technologies that are good enough to replace human workers or, more importantly, to precaritise them and undermine them. ChatGPT isn't really new but simply an iteration of the class war that's been waged since the start of the industrial revolution. That allegedly well-informed commentators can infer that ChatGPT will be used for "cutting staff workloads" rather than for further staff cuts illustrates a general failure to understand AI as a political project. Contemporary AI, as I argue in my book, is an assemblage for automatising administrative violence and amplifying austerity. ChatGPT is a part of a reality distortion field that obscures the underlying extractivism and diverts us into asking the wrong questions and worrying about the wrong things. Instead of expressing wonder, we should be asking whether it's justifiable to burn energy at "eye watering" rates to power the world's largest bullshit machine.

Commentary that claims 'ChatGPT is here to stay and we just need to learn to live with it' are embracing the hopelessness of what I call 'AI Realism'. The compulsion to show 'balance' by always referring to AI's alleged potential for good should be dropped by acknowledging that the social benefits are still speculative while the harms have been empirically demonstrated. Saying, as the OpenAI CEO does, that we are all 'stochastic parrots' like large language models, statistical generators of learned patterns that express nothing deeper, is a form of nihilism. Of course, the elites don't apply that to themselves, just to the rest of us. The structural injustices and supremacist perspectives layered into AI put it firmly on the path of eugenicist solutions to social problems.

Instead of reactionary solutionism, let us ask where the technologies are that people really need. Let us reclaim the idea of socially useful production, of technological developments that start from community needs. The post-Covid 'new normal' has turned out to involve both the normalisation of neural networks and a rise in necropolitics. Transformer models and diffusion models are not creative but carceral - they and other forms of AI imprison our ability to imagine real alternatives. It's not so long ago that we all woke up to the identity of truly essential workers; the people carrying out the precaritised roles of nursing, teaching, caring, delivering and cleaning, the very professions who are being forced to reinvent the idea of the general strike simply to regain the conditions for survival. Instead of being complicit with expensive toys running in carbon emitting data centres, we can focus instead on centring activities of care. As discussed in more detail in 'Resisting AI', a refusal of algorithmic immiseration goes along with a positive search for alternatives, and I lay out a programme of people's councils and commons-based solidarity to do just that. It's not time to chat with AI, but to resist it.

Resisting AI - An Anti-fascist Approach to Artificial Intelligence, Bristol University Press, 2022

Resisting AI - chapter abstracts

2022-02-23T00:00:00+00:00

Chapter abstracts for 'Resisting AI - An Anti-fascist Approach to Artificial Intelligence'

Available from Bristol University Press

chapter 0 - introduction

The introduction starts by grounding AI, for the purposes of the book, as the computational methods of deep learning and the associated institutions and ideologies. It sets out the reasons for resisting AI that are covered in Chapters 1 to 4, introduces the idea of an anti-fascist approach to AI, and closes by outlining the path to overcoming existing AI that is taken in Chapters 5 to 7.

chapter 1 - operations of AI

This chapter delves into the actual mechanics of machine learning. It emphasises the algorithmic operations of optimisation as well as the dependencies on data. Moving on to neural networks it examines the internal transformations of the data, for example through backpropagation, and the paradoxical pairing of predictive accuracy with opacity. The chapter highlights the materiality of AI and its consequences in terms of both carbon emissions and the centralisation of control. It closes by focusing on the poorly paid and invisible workforce that underpins AI, and the continuous thread of anti-workerism that connects the origins of computing (Babbage) to contemporary applications of AI (Amazon).

chapter 2 - collateral damage

This chapter looks at the surprising brittleness of AI and the way its reliance on proxies and shortcuts haunts its social application. It critiques the post-hoc attempts to 'fix' AI's damaging effects through ethics, regulation or human intervention, and focuses on the way AI not only produces discrimination but intensifies existing structural injustice. The chapter closes by looking at the way the performative character of AI produces the very subjects of its judgements, and at the inherently backward-looking nature of its solutionism. The result is deemed to be recipe for 'AI Realism'.

chapter 3 - ai violence

Chapter 3 looks at the entanglement of AI with systemic social structures, starting with the way AI poaches its legitimacy from science. This authority acts as cover for AI's role in increasing neoliberal precarity and austerity, spreading the social logic of financialisation into everyday life. The chapter examines AI's enrolment by institutions of the state, especially welfare systems, and the way it amplifies thoughtlessness and administrative violence. It closes with AI's operation as a technology of racialisation, its similarities to genetic determinism, and the way it becomes a candidate mechanism for a modern race science.

chapter 4 - necropolitics

This chapter looks at the impacts of AI under conditions of social crisis. It describes the way AI acts as an algorithmic shock doctrine, becoming an apparatus for producing states of exception. The effects of the COVID-19 pandemic prefigure the algorithmic distribution of life chances, where AI acts as a necropolitical technology. The chapter explores the origins of AI’s mathematical optimizations in the history of eugenics, and its legacy in the search for superior intelligence. It identifies the steps by which AI may become, through ultrarationalism and neoreaction, a part of a fascistic politics. Chapter 4 closes by looking at AI’s potential enrolment in fascistic responses to the climate crisis.

chapter 5 - postmachinic

Chapter 5 sets out a standpoint for resisting AI based on mutuality and care. It draws from feminist and post-normal science as a way to challenge exclusionary claims to authoritative knowledge. The chapter applies feminist new materialism to interrupt AI's configuration of reality, and to counter its operations of separation with a perspective that is fundamentally relational. This establishes a form of critical pedagogy which can be used to 'learn against the machine' and to recover the potential of prefigurative politics, of 'the possible against the probable'. Ultimately, the chapter recomposes the question of AI as a matter of care that directs attention to the effects of boundaries and exclusions and to the neglect of our interdependence.

chapter 6 - people's councils

Chapter 6 takes the ethics of Chapter 5 and turns them into political tactics through the principles of mutual aid and solidarity. Algorithmic boundaries and enclosures are challenged by a commitment to commonality. The struggles of workers 'above' and 'below' the algorithm are put in relation to the potential of the workers' council, a directly democratic form of organising that is a starting point for structural renewal. This is extended to the idea of the people's council as a self-constituting struggle against abstract segregation. Resistance to AI is compared to the historical movement of Luddism and the need for community constraint of harmful technology. The overall approach is anti-fascist, in that it is both an early recognition of the intensity of the threat and a defence of the space for emancipation.

chapter 7 - anti-fascist AI

Chapter 7 lays out the way an anti-fascist approach to AI moves from resistance to restructuring. It shows by example why an anti-fascist approach must be both decolonial and feminist. The anti-fascist goal of structural renewal is discussed in terms of socially useful production, solidarity economies and the centrality of commons, such that optimization is replaced by commonization. Chapter 7 closes by outlining a new apparatus, one that resonates with a renewed vision of the social. While it may or may not use advanced computation, a new apparatus will support a transition to social autonomy. In place of AI as we know it, the recursive horizontality of a new apparatus is open and adaptive. Rather than trying to 'solve' anything, it helps to sustain care under radically changing conditions.

Obnoxious Machines - the prospects for Luddism in the era of AI

2021-06-23T07:00:00+01:00

A talk given to the Mellon Sawyer Seminar on the History of AI, University of Cambridge, June 23rd 2021

It's time to talk about the Ludding times. That's how they talked about it, those who were there, when readying themselves for an insurrection a few years later. The Ludding times of 1811 to 1816, when communities in Nottinghamshire, Lancashire and the West Riding of Yorkshire rose up against the machines. The Ludding times, with all their similarities to our own times. What can we learn from the Luddites? What can we learn from the heft of a hammer and the idea that, as a later revolutionary said, the urge to destroy is also a creative urge?

To start with, the machines the Luddites were breaking weren't new then, any more that AI is new now, as has been so thoroughly explored in this seminar series on the histories of AI. So what triggered the machine breaking? One catalyst was the condition of the time. Then, like now, England was a country mired in debt, suffering years of hardship, as we've had years of austerity. Then, as now, there was the top-down imposition of new conditions of production that massively increased precarity, thanks to the factories and mills that gathered together machinery in arrangements of steam- or water-powered automation.

The Luddites faced assemblages of machinery that undermined not just artisanship but agency and dignity, ways of life protected by both guilds and common custom, machinic arrangements that sheared and cropped not just cloth but the conditions of social reproduction, radically altering the relations of power under the guise of modernisation. This is a pattern we can see being repeated in front of our eyes. The Luddites speak directly to the need to interrupt AI and to apply community-led constraints to its social violence.

The Luddites had looked to the law to protect them but had been let down. Their petitions had failed and statutory protections were repealed, while food prices spiked and their trade was undermined. In the here-and-now we're being distracted by regulatory theatre: although the GDPR is dead in the water, we're waiting for the penny to drop that the law isn't there to protect us, but to protect property interests. The law can't correct for algorithmic injustice resulting from structural inequality, because the law itself sustains those inequalities. It’s not that law will fail to regulate the harmful effects of AI, but rather that AI is already exposing the comprehensive failure of the law to address real injustice.

The Luddites' reward for challenging the subservience of law to economic interests was the imposition of states of exception: the Combination Act of 1799 had already made trade unionism illegal and Luddism was followed by the notorious Six Acts incuding the Seditious Meetings Act. Meanwhile, we have the Police, Crime, Sentencing and Courts Bill 2021 and the emergence of algorithmic states of exception. Indeed, basing substantive allocation or determination on AI's correlations not on causality could be said to be a statistical suspension of habeus corpus.

What the Luddites were resisting was not simply automation, but their own reduction and automatisation. From, a General Meeting of Plain Silk Framework-knitters, held at the Fox and Owl Inn, Derby, December 9th, 1811: "We are at a loss to know where to fix the stigma (too much blame being due to ourselves for not watching better over the trade) as each striving to manufacture on the lowest terms, makes us little better than mere engines to support a jealous competition in the market".

In Luddite times there was still a 'moral economy' that restrained capital accumulation. One manifestation was so-called food riots, where crowds seized grain but in order to sell it at what they deemed a fair price. The Luddites transformed the moral economy into a political struggle. As one letter says "But we. We petition no more - that won't do - fighting must". Their stance was well expressed in a letter to 'M. Smith, Shearing Frame Holder at Hill End Yorkshire', signed by Ned Ludd, clerk to the General Army of the Redressers, which annouced their intention to "put down all Machinery hurtful to Commonality". During the first might of attacks in Huddersfield, on the night of 22nd february 1812, armed with a few pistols, wielding hammers, a masked group of about 45 smashed 8 shearing frames at two separate workshops. At later dates they assembled in hundreds to mount attacks on the mills, some of which were burned down.

As the song General Ludd’s Triumph says:
"And when in the work of destruction employed
He himself to no method confines
By fire, and by water he gets them destroyed
For the Elements aid his designs
Whether guarded by Soldiers along the Highway
Or closely secured in the room
He shivers them up both by night and by day
And nothing can soften their doom"

The Luddites fight wasn't simply defensive: there was an alternative social vision afoot, fuelled by the French revolution. The political atmosphere before, during and after Luddism was filled with strands of republicanism, from the Jacobins to Peterloo. The Luddites were suppressed with the help of several notorious spies, while in our time Amazon hires the infamous Pinkerton Detective Agency to surveil warehouse workers suspected of union sympathies. The Luddite struggle took place before the rigid separation of society into sites of production and sites of consumption. Its defeat, however, was followed by the ascendence of factory labour.

Only 20 years after the Luddite uprising Charles Babbage, of Analytical and Difference Engine fame, published On the Economy of Machinery and Manufactures. He saw continuity between calculating machines and factories, in that they were both based on division and analytical regulation. What stretches from Babbage to the applied AI of Amazon is not just the history of computing, but a committed anti-workerism. Babbage wrote that “One great advantage which we derive from machinery is the check which it affords against the inattention, idleness or the dishonesty of human agents”.

In the 1960s Tronti and other autonomists proposed the analytic of the social factory, such that all of life has been subsumed into capitalist relations of production. Now AI brings us the algorithmic factory, where all of life is enrolled in machinic optimisation. So if there are so many similarities between Ludding times and ours, and if systems of socially-applied AI are, to use one of the Luddites' favourite terms, obnoxious machines, who are the Luddites now? Is it Hong Kong protestors who cut down smart lampposts running AI facial recognition? Is it Amazon workers at the warehouse outside Paris who cut the power for 8 hours? Is it A-level students shouting 'fuck the algorithm', or the residents and parents in Chandler, Arizona, who blockaded Waymo’s self-driving vans as they were using the town for training, and in at least one case threw rocks at them. "They didn't ask us if we wanted to be part of their beta test" said a mother who’s child was nearly hit by one of the vans.

The example of the Luddites reframes the question of property damage. Luddism was a reaction to economic violence imposed on the weaving communities. AI brings us epistemic violence, a way of knowing that supersedes lived experience, at least, the lived experience of the minoritised and marginalised. AI powers statistical segregation that becomes administrative violence. AI offers the violence of solutionism that subsumes the systemic sources of injustice, while the AI driven triage of welfare scales the structural violence of the state. Who, then, is to say that the time for machine breaking ever went away?

Luddism shows us what is positive in refusal, which isn't simply negation. As abolitionist Ruth Wilson Gilmore says: "prison abolition is not just about closing prisons its a theory of change". Calling for the complete abolition of AI will be derided in the same way as Luddism and prison abolition, but abolishing AI is part of striving for a better world in the here and now. Certainly Luddism is a figure for the militancy that's absent from tech critique, but the lessons we can draw are not only about hammers but about the strength of the community that wielded them.

The strength of Luddite resistance took more troops stationed in the north of England to suppress it, than Wellington took at the same time into the iberian peninsula to fight Napoleon. That strength came from solidarity. Solidarity is an action-oriented commitment to one another based on the recognition of a shared commonality. Solidarity is the starting point for a movement to transcend AI, a relationality that comes prior to any system of social, legal or algorithmic classification. The fundamental role of mutual aid cannot be parsed as an optimisation problem. There are collective choices to be made about the kind of futures we want, not just the ones we're statistically predicted to have. Choosing solidarity is to stand against social hierarchy and its algorithmic naturalisation.

The Luddites weren't just self-organised, they were a constituent power, asserting their own right to define the governance of their trade and in the end, of their communities. Organisation between areas was based on delegates from local committees. In Huddersfield, the post-Luddite insurrectionists of 1820 had plans to set-up self-government. Machine breaking was the wildcat action of its time, and the West Riding was the Rojava of its day.

Tech worker protests since Trump, especially those in the giant AI corporations where workers can see their creations being used for immiseration, and worse, are an opening for constituent self-organisation. The workers council: a form that connects us back to the Luddites through the Chartism of the 1840s to the Argentinian crisis of 2001. The council is a decision-making assembly, federating from the bottom up, whose task is to forge, through struggle, a new hammer. As the Luddites would say: “Enoch made them, and Enoch shall break them.”

The struggle is everywhere across the algorithmic factory. Autonomous interventions come through people's councils: those who are affected by AI, who come together as a body to decide what to do about it. People’s councils are horizontal assemblies that prefigure an end to exclusion and exception. As with the Luddites, people's councils are a nomadic tactic: it's for participants to decide where best to intervene in the iterative sedimentation of power. AI's exclusions are always at the same time enclosures, while commoning is the action of taking aspects of the world back into the collectively governed commons. The task for people's councils is not only to disrupt the apparatus of AI, but to invert the state of exception, to reclaim those spaces for the commons, and to occupy those spaces with autonomous activity.

The Luddites struck while machine looms and mills were still being established. Algorithmic force is already undermining due process and social solidarity, but all is not yet lost: the very generalisability of AI drives the kind of intersectionality that catalyses the contestation of the totality of social relations. The roots, perhaps, of a militant alternative to the necropolitics of neural networks. Efficiency through segregation and optimisation is the language of supremacy. AI bears this mark from its mathematical pre-origins in the pursuit of eugenics. The role of the modern Luddite is to 'no platform' the platform economy. As was said in a threatening letter sent to the Reverend W. Blacow in 1812, in response a sermon where he had denounced the Luddites, signed by a 'Lt of the Luddites':
"I will overturn, overturn, overturn".

Dan McQuillan 23rd June 2021

AI Realism and structural alternatives

2019-06-07T00:00:00+01:00

The speakers notes for a 6.5 minute talk given at the Data Justice Lab in Cardiff, June 7th 2019.

1. not transformation but intensification

The introduction of process automation and predictive analytics via machine learning
is not a transformation,
it's an intensification.

Machine learning and bureaucracy are both generalisable modes of rational ordering
based on abstraction and deriving authority from claims to neutrality and objectivity.

The justification for bureaucratic rationality is efficiency.
Machine learning adds inferential governance in the name of optimisation.

Efficiency is already questionable
as it's only calculable after a reductive rendering of its social objects,
while optimisation overrides social complexity through its objective function.

2. it's about risk instead of changing things

Seeing society as categories of actuarial risk
is to filter people's lives through the epistemology of insurance and instrumentalism.

Machine learning achieves its insights through discriminating between its classes
via an abstract distance in data space; it's a logic of statistical segregation.

Applied to social welfare it becomes calculative Victorianism,
assigning morality via metrics of 'deservingness'.
It's an ethics of triage via the computerisation of stigma.

Machine learning extends bureaucracy into the future;
or rather, it bureaucratises a probabilistic future and actualises it in the present.
Risk is remodelled as a dynamic phenomenon open to 'nudging',
yet correlations are not causation and tell us little about the best ways to intervene.

By only selecting for features that differentiate between individuals,
we are bracketing out the problems people have in common.
The goal is targeting instead of raising up whole populations.

3. the collateral damage

The collateral damage of this intensification includes
- an erosion of due process through opacity
- an amplification of thoughtlessness, in the sense that Hannah Arendt meant it
- the production of epistemic injustice, where calculations count more than testimony
- an asymmetric focus on those about whom civic data is already most plentiful
- the multiplication of categories that increase potential moments for administrative violence
and new opportunities for institutional gaming.

4. reforms are no solution

A human-in-the-loop is not a humanistic pushback
as that human is themselves subsumed by the institution-in-the-loop.

While in the private sector and across government,
ethics washing has become a form of institutional hydropower.

Privacy is hard to enforce when you've built a proxying machine
and data sharing is the dominant mode of value extraction,
so people are finally starting to call for regulation and law,
although this often seems to assume that society
is a level playing field that simply needs better fences.

In any case, more regulation means more bureaucracy or more machine learning to monitor it
making the zweckrational, as Weber called it, recursive.

5. people's councils

People’s councils, on the other hand, are face-to-face democratic assemblies;
a horizontally organised refusal to be rendered as data dividuals.
They are a collective questioning
of the decisions that define the way the machines will make decisions,
by applying critical pedagogy and situated knowledge.

They constitute a different subjectivity -
iterative deliberation of consensus, done right,
is an antidote to bureaucracy and to the calculative iterations of machine learning.

People's councils apply Bergson's critique of ready-made problems,
reversing statistical reductiveness through
a commitment to the possible over the probable.

Like Ivan Illich, they value a convivial technology
and are prepared to apply limits.

We need to develop a different order of ordering.
Instead of ways of organising that allow everyone to evade responsibility,
we need to reclaim our own agency through self-organisation.

Only under these conditions will we discover whether a re-imagined machine learning
can become a people's technology.

6. AI Realism

Mark Fisher coined the term Capitalist Realism
to describe the entrenched belief that despite the global financial crash
there is no alternative.

What we're seeing now is AI realism.
While the reform of AI is endlessly discussed,
there is no attempt to seriously question whether we should be using it at all.

But rather than a sci-fi future,
we are to be left behind in computationally-optimised deprivation.

We need to think collectively about ways out of this mess,
learning from and with each other rather than relying on machine learning.
countering thoughtlessness with practices of collective care.

We can't uninvent either AI or bureaucracy,
but we can choose to radically change both our modes of organisation
and our approach to computational learning.

7. agile populism

As a warning footnote,
the simplification of social problems to optimisation
based on reductionive reasoning and innate characteristics
is the politics of populism.
What institutional machine learning risks creating by default is
a machine for the agile construction of populist targets.
AI realism is only one step from the analytics of the far right.

Towards an anti-fascist AI

2019-04-01T00:00:00+01:00

A talk given at the launch of the 'All Access AI' network at Goldsmiths, Unversity of London 1st April 2019.

This talk refers to a text developed for Propositions for Non-Fascist Living: Tentative and Urgent, published by BAK, basis voor actuele kunst and MIT Press (forthcoming November 2019).

intro

This talk is about some pressing issues with AI that don't usually make the headlines, and why tackling those issues means developing an antifascist AI.

When I talk about AI i'm talking about machine learning and about artificial neural networks, also known as deep learning¹. I'm addressing actual AI not a literary or filmic narrative about post-humanism.

AI is political. Not only because of the question of what is to be done with it, but because of the political tendecies of the technology itself. The possibilities of AI arise from the resonances between its concrete operations and the surrounding political conditions. By influencing our understanding of what is both possible and desirable it acts in the space between what is and what ought to be.

concrete

Computers are essentially just faster collections of vacuum tubes. How can they emulate human activities like recognising faces or assessing criminality?

Think about a least squares fit; you're trying assess the correlation between two variables by fitting a straight line to scattered points, so you calculate the sum of the squares of distances from all points to the line and minimise that². Machine learning does something very similar. It makes your data into vectors in feature space so you can try to find boundaries between classes by minimising sums of distances as defined by your objective function³.

These patterns are taken as revealing something significant about the world. They take on the neoplatonism of the mathematical sciences; a belief in a layer of reality which can be best perceived mathematically⁴. But these are patterns based on correlation not causality; however complex the computation there's no comprehension or even common sense.

Neural networks doing image classification are easily fooled by strange poses of familiar objects. So a school bus on it's side is confidently classified as a snow plough⁵. Yet the hubristic knights of AI are charging into messy social contexts, expecting to be able to draw out insights that were previously the domain of discourse.

Deep learning is already seriously out of its depth.

callousness

Will this slow the adoption of AI while we figure out what it's actually good for? No, it won't; because what we are seeing is 'AI under austerity', the adoption of machinic methods to sort things out after the financial crisis.

The way AI derives its optimisation from calculations based on a vast set of discrete inputs matches exactly the way neoliberalism sees the best outcome coming from a market freed of constraints. AI is seen as a way to square the circle between eviscerated services and rising demand without having to challenge the underlying logic⁶.

The pattern finding of AI lends itself to prediction and therefore preemption which can target what's left of public resource to where the trouble will arise, whether that's crime, child abuse or dementia. But there's no obvious way to reverse operations like backpropagation to human reasoning⁷, which not only endangers due process but produces thoughtlessness in the sense that Hannah Arendt meant it⁸; the inability to critique instructions, the lack of reflection on consequences, a commitment to the belief that a correct ordering is being carried out.

The usual objection to algorithmic judgements is outrage at the false positives, especially when they result from biased input data. But the underlying problem is the imposition of an optimisation based on a single idea of what is for the best, with a resultant ranking of the deserving and the undeserving⁹.

What we risk with the uncritical adoption of AI is algorithmic callousness, which won't be saved by having a human-in-the-loop because that human will be subsumed by the self-interested institution-in-the-loop.

By throwing out our common and shared conditions as having no predictive value, the operations of AI targeting strip out any acknowledgement of system-wide causes hiding the politics of the situation.

far right

The algorithmic coupling of vectorial distances and social differences will become the easiest way to administer a hostile environment, such as the one created by Theresa May to target immigrants.

But the overlaps with far right politics don't stop there. The character of 'coming to know through AI' involves reductive simplifications based on data innate to the analysis, and simplifying social problems to matters of exclusion based on innate characteristics is precisely the politics of right wing populism.

We should ask whether the giant AI corporations would baulk at putting the levers of mass correlation at the disposal of regimes seeking national rebirth through rationalised ethnocentrism. At the same time that Daniel Guerin was writing his book in 1936 examining the ties between fascism and big business¹⁰, Thomas Watson's IBM and it's German subsidiary Dehomag were enthusiastically furnishing the nazis with Hollerith punch card technology¹¹. Now we see the photos from Davos of Jair Bolsonaro seated at lunch between Apple's Tim Cook and Microsoft's Satya Nadella¹².

Meanwhile the algorithmic correlations of genome wide association studies are used to sustain notions of race realism and prop up a narrative of genomic hierarchy¹³. This is already a historical reunification of statistics and white supremacy, as the mathematics of logistic regression and correlation that are so central to machine learning were actually developed by Edwardian eugenicists Francis Galton and Karl Pearson.

antifascist

My proposal here is that we need to develop an antifascist AI.

It needs to be more than debiasing datasets because that leaves the core of AI untouched. It needs to be more than inclusive participation in the engineering elite because that, while important, won't in itself transform AI. It needs to be more than an ethical AI, because most ethical AI operates as PR to calm public fears while industry gets on with it¹⁴. It needs to be more than ideas of fairness expressed as law, because that imagines society is already an even playing field and obfuscates the structural asymmetries generating the perfectly legal injustices we see deepening every day.

I think a good start is to take some guidance from the feminist and decolonial technology studies that have cast doubt on our cast-iron ideas about objectivity and neutrality¹⁵. Standpoint theory suggests that positions of social and political disadvantage can become sites of analytical advantage, and that only partial and situated perspectives can be the source of a strongly objective vision¹⁶. Likewise, a feminist ethics of care takes relationality as fundamental¹⁷, establishing a relationship between the inquirer and their subjects of inquiry would help overcome the onlooker consciousness of AI.

To centre marginal voices and relationality, I suggest that an antifascist AI involves some kinds of people's councils, to put the perspective of marginalised groups at the core of AI practice and to transform machine learning into a form of critical pedagogy[^n18]. This formation of AI would not simply rush into optimising hyperparameters but would question the origin of the problematics, that is, the structural forces that have constructed the problem and prioritised it.

AI is currently at the service of what Bergson called ready-made problems; problems based on unexamined assumptions and institutional agendas, presupposing solutions constructed from the same conceptual asbestos¹⁹. To have agency is to re-invent the problem, to make something newly real that thereby becomes possible unlike the probable, the possible is something unpredictable, not a rearrangement of existing facts.

Given the corporate capture of AI, any real transformation will require a shift in the relations of production. One thing that marks the last year or so is the sign of internal dissent in Google²⁰, Amazon²¹, Microsoft²², Salesforce and so on about the social purposes to which their algorithms are being put. In the 1970s workers in a UK arms factory came up with the Lucas plan which proposed the comprehensive restructuring of their workplace for socially useful production²³. They not only questioned the purpose of the work but did so by asserting the role of organised workers, which suggests that the current tech worker dissent will become transformative when it sees itself as creating the possibility of a new society in the shell of the old²⁴.

I'm suggesting that an antifascist AI is one that take sides with the possible against the probable, and does so at the meeting point between organised subjects and organised workers. But it may also require some organised resistance from communities.

A thread is a sequence of programmed instructions executed by microprocessor. On an Nvidia GPU, one of the AI chips, a warp is a set of threads executed in parallel²⁵. How uncanny that the language of weaving looms has followed us from the time of the Luddites to the era of AI. The struggle for self-determination in everyday life may require a new Luddite movement, like the residents and parents in Chandler, Arizona who have blockaded Waymo's self-driving vans 'They didn’t ask us if we wanted to be part of their beta test' said a mother who's child was nearly hit by one²⁶ The Luddites, remember, weren't anti-technology but aimed 'to put down all machinery hurtful to the Commonality'²⁷.

The predictive pattern recognition of deep learning is being brought to bear on our lives with the granular resolution of Lidar. Either we will be ordered by it or we will organise. So the question of an antifascist AI is the question of self-organisation, and of the autonomous production of the self that is organising.

Asking 'how can we predict who will do X?' is asking the wrong question. We already know the destructive consequences of on the individual and collective psyche of poverty, racism and systemic neglect. We don't need AI as targeting but as something that helps raise up whole populations²⁸.

Real AI matters not because it heralds machine intelligence but because it confronts us with the unresolved injustices of our current system. An antifascist AI is a project based on solidarity, mutual aid and collective care. We don't need autonomous machines but a technics that is part of a movement for social autonomy.

Nielsen, Michael A. 2015. ‘Neural Networks and Deep Learning’. 2015. http://neuralnetworksanddeeplearning.com. ↩
Chauhan, Nagesh Singh. n.d. ‘A Beginner’s Guide to Linear Regression in Python with Scikit-Learn’. Accessed 1 April 2019. https://www.kdnuggets.com/2019/03/beginners-guide-linear-regression-python-scikit-learn.html, https://www.kdnuggets.com/2019/03/beginners-guide-linear-regression-python-scikit-learn.html. ↩
Geitgey, Adam. 2014. ‘Machine Learning Is Fun!’ Adam Geitgey (blog). 5 May 2014. https://medium.com/@ageitgey/machine-learning-is-fun-80ea3ec3c471. ↩
McQuillan, Dan. 2017. ‘Data Science as Machinic Neoplatonism’. Philosophy & Technology, August, 1–20. https://doi.org/10.1007/s13347-017-0273-3. ↩
Alcorn, Michael A., Qi Li, Zhitao Gong, Chengfei Wang, Long Mai, Wei-Shinn Ku, and Anh Nguyen. 2018. ‘Strike (with) a Pose: Neural Networks Are Easily Fooled by Strange Poses of Familiar Objects’. ArXiv:1811.11553 [Cs], November. http://arxiv.org/abs/1811.11553. ↩
Ghosh, Pallab. 2018. ‘AI Could Save Heart and Cancer Patients’, 2 January 2018, sec. Health. https://www.bbc.com/news/health-42357257. ↩
Knight, Will. n.d. ‘There’s a Big Problem with AI: Even Its Creators Can’t Explain How It Works’. MIT Technology Review. Accessed 1 April 2019. https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/. ↩
Arendt, Hannah. 2006. Eichmann in Jerusalem: A Report on the Banality of Evil. 1 edition. New York, N.Y: Penguin Classics. ↩
Big Brother Watch. 2018. ‘The UK’s Poorest Are Put at Risk by Automated Welfare Decisions’. 7 November 2018. https://bigbrotherwatch.org.uk/2018/11/the-uks-poorest-are-put-at-risk-by-automated-welfare-decisions/. ↩
Guerin, Daniel. 2000. Fascism and Big Business. Translated by Francis Merrill and Mason Merrill. 2nd edition. Pathfinder Press. ↩
Black, Edwin. 2012. IBM and the Holocaust: The Strategic Alliance Between Nazi Germany and America’s Most Powerful Corporation-Expanded Edition. Expanded edition. Washington, DC: Dialog Press. ↩
Slobodian, Quinn. 2019. ‘The Rise of the Right-Wing Globalists’. The New Statesman. 31 January 2019. https://www.newstatesman.com/politics/economy/2019/01/rise-right-wing-globalists. ↩
Comfort, Nathaniel. 2018. ‘Sociogenomics Is Opening a New Door to Eugenics’. MIT Technology Review. 23 October 2018. https://www.technologyreview.com/s/612275/sociogenomics-is-opening-a-new-door-to-eugenics/. ↩
Metz, Cade. 2019. ‘Is Ethical A.I. Even Possible?’ The New York Times, 5 March 2019, sec. Business. https://www.nytimes.com/2019/03/01/business/ethics-artificial-intelligence.html. ↩
Harding, Sandra. 1998. Is Science Multicultural?: Postcolonialisms, Feminisms, and Epistemologies. 1st edition. Bloomington, Ind: Indiana University Press. ↩
Haraway, Donna. 1988. ‘Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective’. Feminist Studies 14 (3): 575–99. https://doi.org/10.2307/3178066. ↩
‘Carol Gilligan Interview.’ 2011. Ethics of Care (blog). 16 July 2011. https://ethicsofcare.org/carol-gilligan/. ↩
McQuillan, Dan. 2018. ‘People’s Councils for Ethical Machine Learning’. Social Media + Society 4 (2): 2056305118768303. https://doi.org/10.1177/2056305118768303. ↩
Solhdju, Katrin. 2015. ‘Taking Sides with the Possible against Probabilities or: How to Inherit the Past’. In . ICI Berlin. https://www.academia.edu/19861751/Taking_Sides_with_the_Possible_against_Probabilities_or_How_to_Inherit_the_Past. ↩
Shane, Scott, Cade Metz, and Daisuke Wakabayashi. 2018. ‘How a Pentagon Contract Became an Identity Crisis for Google’. The New York Times, 26 November 2018, sec. Technology. https://www.nytimes.com/2018/05/30/technology/google-project-maven-pentagon.html. ↩
Conger, Kate. n.d. ‘Amazon Workers Demand Jeff Bezos Cancel Face Recognition Contracts With Law Enforcement’. Gizmodo. Accessed 1 April 2019. https://gizmodo.com/amazon-workers-demand-jeff-bezos-cancel-face-recognitio-1827037509. ↩
Lee, Dave. 2019. ‘Microsoft Staff: Do Not Use HoloLens for War’, 22 February 2019, sec. Technology. https://www.bbc.com/news/technology-47339774. ↩
Open University. 1978. Lucas Plan Documentary. https://www.youtube.com/watch?v=0pgQqfpub-c. ↩
‘Preamble to the IWW Constitution | Industrial Workers of the World’. 1905. 1905. http://www.iww.org/culture/official/preamble.shtml. ↩
Lin, Yuan, and Vinod Grover. 2018. ‘Using CUDA Warp-Level Primitives’. NVIDIA Developer Blog. 16 January 2018. https://devblogs.nvidia.com/using-cuda-warp-level-primitives/. ↩
Romero, Simon. 2019. ‘Wielding Rocks and Knives, Arizonans Attack Self-Driving Cars’. The New York Times, 2 January 2019, sec. U.S. https://www.nytimes.com/2018/12/31/us/waymo-self-driving-cars-arizona-attacks.html. ↩
Binfield, Kevin, ed. 2004. Writings of the Luddites. Baltimore: Johns Hopkins University Press. ↩
Keddell, Emily. 2015. ‘Predictive Risk Modelling: On Rights, Data and Politics.’ Re-Imagining Social Work in Aotearoa New Zealand (blog). 4 June 2015. http://www.reimaginingsocialwork.nz/2015/06/predictive-risk-modelling-on-rights-data-and-politics/. [^ ↩

Rethinking AI through the politics of 1968

2018-09-22T00:00:00+01:00

This talk was given at the conference 'Rethinking the legacy of 1968: Left fields and the quest for common ground' held at The Centre for Cultural Studies Research, University of East London on September 22nd 2018 http://rethinking1968.today/

There's a definite resonance between the agitprop of '68 and social media. Participants in the UCU strike earlier this year, for example, experienced Twitter as a platform for both affective solidarity and practical self-organisation¹. However, there is a different geneaology that speaks directly to our current condition; that of systems theory and cybernetics. What happens when the struggle in the streets takes place in the smart city of sensors and data? Perhaps the revolution will not be televised, but it will certainly be subject to algorithmic analysis. Let's not forget that 1968 also saw the release of '2001: A Space Odyssey' featuring the AI supercomputer HAL.

While opposition to the Vietnam war was a rallying point for the movements of '68, the war itself was also notable for the application of systems analysis by US Secretary of Defense Robert McNamara, who attempted to make it, in modern parlance, a data-driven war. During the Vietnam war the hamlet pacification programme alone produced 90,000 pages of data and reports a month², and the body count metric was published in the daily newspapers. The milieu that helped breed our current algorithmic dilemmas was the contemporaneous swirl of systems theory and cybernetics, ideas about emergent behaviour and experiments with computational reasoning, and the intermingling of military funding with the hippy visions of the Whole Earth Catalogue.

The double helix of DARPA and Silican Valley can be traced through the evolution of the web to the present day, where AI and machine learning are making inroads everywhere carrying their own narratives of revolutionary disruption; a Ho Chi Minh trail of predictive analytics. They are playing Go better than grand masters and preparing to drive everyone's car, while the media panics about AI taking our jobs. But this AI is nothing like HAL, it's a form of pattern finding based on mathematical minimisation; like a complex version of fitting a straight line to a set of points. These algorithms find the optimal solution when the input data is both plentiful and messy. Algorithms like backpropagation³ can find patterns in data that were intractable to analytical description, such as recognising human faces seen at different angles, in shadows and with occlusions. The algorithms of Ai crunch the correlations and the results often work uncannily well.

But it's still computers doing what computers have been good at since the days of vacuum tubes; performing mathematical calculations more quickly than us. Thanks to algorithms like neural networks this calculative power can learn to emulate us in ways we would never have guessed at. This learning can be applied to any context that is boiled down to a set of numbers, such that the features of each example are reduced to a row of digits between zero and one and are labelled by a target outcome. The datasets end up looking pretty much the same whether it's cancer scans or netflix viewing figures. There's nothing going on inside except maths; no self-awareness and no assimilation of embodied experience. These machines can develop their own unprogrammed behaviours but utterly lack an understanding of whether what they've learned makes sense. And yet, machine learning and AI are becoming the mechanisms of modern reasoning, bringing with them the kind of dualism that the philosophy of '68 was set against, a belief in a hidden layer of reality which is ontologically superior and expressed mathematically⁴.

The delphic accuracy of AI comes with built-in opacity because massively parallel calculations can't always be reversed to human reasoning, while at the same time it will happily regurgitate society's prejudices when trained on raw social data. It's also mathematically impossible to design an algorithm be fair to all groups at the same time⁵. For example, if the reoffending base rates vary by ethnicity, a recidivism algorithm like COMPAS will predict different numbers of false positives and more black people will be unfairly refused bail⁶. The wider impact comes from the way the algorithms proliferate social categorisations such as 'troubled family' or 'student likely to underachieve', fractalising social binaries wherever they divide into 'is' and 'is not'. This isn't only a matter of data dividuals misrepresenting our authentic selves but of technologies of the self that, through repetition, produce subjects and act on them. And, as AI analysis starts overcode MRI scans to force psychosocial symptoms back into the brain, we will even see algorithms play a part in the becoming of our bodies⁷.

What we call AI, that is, machine learning acting in the world, is actually a political technology in the broadest sense. Yet under the cover of algorithmic claims to objectivity, neutrality and universality there's an infrastructual switch of allegiance to algorithmic governance. The dialectic that drives AI into the heart of the system is the contradiction of societies that are data rich but subject to austerity. One need only look at the recent announcements about a brave new NHS to see the fervour welcoming this salvation⁸. While the global financial crisis is manufactured, the restructuring is real; algorithms are being enrolled in the refiguring of work and social relations such that precarious employment depends on satisfying algorithmic demands⁹ and the public sphere exists inside a targeted attention economy.

Algorithms and machine learning are coming to act in the way pithily described by Pierre Bourdieu, as structured structures predisposed to function as structuring structures¹⁰, such that they become absorbed by us as habits, attitudes, and pre-reflexive behaviours. In fact, like global warming, AI has become a hyperobject¹¹ so massive that its totality is not realised in any local manifestation, a higher dimensional entity that adheres to anything it touches, whatever the resistance, and which is percieved by us through its informational imprints. A key imprint of machine learning is its predictive power. Having learned both the gross and subtle elements of a pattern it can be applied to new data to predict which outcome is most likely, whether that is a purchasing decision or a terrorist attack. This leads ineluctably to the logic of preemption in any social field where data exists, which is every social field, so algorithms are predicting which prisoners should be given parole and which parents are likely to abuse their children¹²¹³.

We should bear in mind that the logic of these analytics is correlation. It's purely pattern matching not the revelation of a causal mechanism, so enforcing the foreclosure of alternative futures becomes effect without cause. The computational boundaries that classify the input data map outwards as cybernetic exclusions, implementing continuous forms of what Agamben calls states of exception. The internal imperative of all machine learning, which is to optimise the fit of the generated function, is entrained within a process of social and economic optimisation, fusing marketing and military strategies through the unitary activity of targeting.

A society who's synapses have been replaced by neural networks will generally tend to a heightened version of the status quo. Machine learning by itself cannot learn a new system of social patterns, only pump up the existing ones as computationally eternal. Moreover, the weight of those amplified effects will fall on the most data visible i.e. the poor and marginalised. The net effect being, as the book title says, the automation of inequality¹⁴. But at the very moment when the tech has emerged to fully automate neoliberalism the wider system has lost it's best-of-all-possible-worlds authority, and racist authoritarianism mestastasizes across the veneer of democracy. The opacity of algorithmic classifications already have the tendency to evade due process, never mind when the levers of mass correlation are at the disposal of ideologies based on paranoid conspiracy theories. A common core to all forms of fascism is a rebirth of the nation from its present decadence, and a mobilisation to deal with those parts of the population that are the contamination¹⁵. The automated identification of anomalies is exactly what machine learning is good at, at the same time as promoting the kind of thoughtlessness that Arendt identified in Eichmann.

So much for the intensification of authoritarian tendencies by AI. What of resistance? Dissident Google staff forced them to partly drop project Maven¹⁶, which develops drone targeting, and Amazon workers are campaigning against the sale of facial recognition systems to the government. But these workers are the privileged guilds of modern tech; this isn't a return of working class power. In the UK and USA there's a general institutional push for ethical AI, in fact you can't move for initiatives aiming to add ethics to algorithms¹⁷, but i suspect this is mainly preemptive PR to head off people's growing unease about their coming AI overlords. All the initiatives that want to make AI ethical seem to think it's about adding something i.e. ethics, instead of about revealing the value-laden-ness at every level of computation, right down to the mathematics.

Models of radical democratic practice offer a more political response through structures such as people's councils composed of those directly affected, mobilising what Donna Haraway calls situated knowledges through horizontalism and direct democracy¹⁸. While these are valid modes of resistance, there's also the '68 notion from groups like the Situationists that the Spectacle generates the potential for it's own supersession¹⁹. I'd suggest that the self-subverting quality in AI is its latent surrealism. For example, experiments to figure out how image recognition actually works probed the contents of intermediary layers in the neural networks, and by recursively applying filters to these outputs produced hallucinatory images that are straight out of an acid trip, such as snail-dogs and trees made entirely of eyes²⁰. When people deliberately feed AI the wrong kind of data it makes surreal classifications. It's a lot of fun, and can even make art that gets shown in galleries²¹ but, like the Situationist derive through the Harz region of Germany while blindly following a map of London, it can also be a poetic disorientation that coaxes us out of our habitual categories.

While businesses and bureaucracies apply AI to the most serious contexts to make or save money or, through some miracle of machinic objectivity, solve society's toughest problems, its liberatory potential is actually ludic. It should be used playfully instead of abused as a form of prophecy. But playfully serious, like the tactics of the Situationists themselves, a disordering of the senses to reveal the possibilities hidden by the dead weight of commodification. Reactivating the demands of the social movements of '68 that work becomes play, the useful becomes the good, and life itself becomes art.

At this point in time, where our futures are becoming cut off by algorithmic preemption we need to pursue a political philosophy that was embraced in '68 of living the new society through authentic action in the here and now. A counterculture of AI must be based on immediacy. The struggle in the streets must go hand in hand with a detournement of machine learning; one that seeks authentic decentralisation not Uber-ised serfdom, and federated horizontalism not the invisible nudges of algorithmic governance. We want a fun yet anti-fascist AI, so we can say "beneath the backpropagation, the beach!"²².

Kobie, Nicole. ‘#NoCapitulation: How One Hashtag Saved the UK University Strike’. Wired UK 18 Mar. 2018. https://www.wired.co.uk/article/no-capitulation-uk-university-pension-protest-twitter. ↩
Thayer, Thomas C. A Systems Analysis View of the Vietnam War: 1965-1972. Volume 2. Forces and Manpower. 1975. www.dtic.mil. http://www.dtic.mil/docs/citations/ADA051609. ↩
3Blue1Brown. What Is Backpropagation Really Doing? | Deep Learning, Chapter 3. N.p. Film. https://www.youtube.com/watch?v=Ilg3gGewQ5U ↩
McQuillan, Dan. ‘Data Science as Machinic Neoplatonism’. Philosophy & Technology (2017): 1–20. https://link.springer.com/article/10.1007/s13347-017-0273-3 ↩
Narayanan, Arvind. Tutorial: 21 Fairness Definitions and Their Politics. N.p. https://www.youtube.com/watch?v=jIXIuYdnyyk ↩
Corbett-Davies, Sam et al. ‘A Computer Program Used for Bail and Sentencing Decisions Was Labeled Biased against Blacks. It’s Actually Not That Clear.’ Washington Post 17 Oct. 2016. https://www.washingtonpost.com/news/monkey-cage/wp/2016/10/17/can-an-algorithm-be-racist-our-analysis-is-more-cautious-than-propublicas/. ↩
Resnick, Brian. ‘Treating Depression Is Guesswork. Psychiatrists Are Beginning to Crack the Code.’ Vox. N.p., 4 Apr. 2017. https://www.vox.com/science-and-health/2017/4/4/15073652/precision-psychiatry-depression. ↩
Department of Health and Social Care. ‘Matt Hancock: New Technology Is Key to Making NHS the World’s Best’. GOV.UK. N.p., 6 Sept. 2018. https://www.gov.uk/government/news/matt-hancock-new-technology-is-key-to-making-nhs-the-worlds-best. ↩
O’Connor, Sarah. ‘When Your Boss Is an Algorithm’. Financial Times. N.p., 8 Sept. 2016. https://www.ft.com/content/88fdc58e-754f-11e6-b60a-de4532d5ea35. ↩
Bourdieu, Pierre. The Logic of Practice. p53. Stanford University Press, 1990. ↩
Morton, Timothy. Hyperobjects - Philosophy and Ecology after the End of the World. University Of Minnesota Press, 2013. https://www.upress.umn.edu/book-division/books/hyperobjects. ↩
Keddell, Emily. ‘Predictive Risk Modelling: On Rights, Data and Politics.’ Re-Imagining Social Work in Aotearoa New Zealand 4 June 2015.. http://www.reimaginingsocialwork.nz/2015/06/predictive-risk-modelling-on-rights-data-and-politics/. ↩
McIntyre, Niamh, and David Pegg. ‘Councils Use 377,000 People’s Data in Efforts to Predict Child Abuse’. The Guardian 16 Sept. 2018. www.theguardian.com. https://www.theguardian.com/society/2018/sep/16/councils-use-377000-peoples-data-in-efforts-to-predict-child-abuse. ↩
Eubanks, Virginia. ‘A Child Abuse Prediction Model Fails Poor Families’. Wired 15 Jan. 2018. https://www.wired.com/story/excerpt-from-automating-inequality/. ↩
iGriffin, Roger. ‘The Palingenetic Core of Fascist Ideology’. Library of Social Science. N.p., n.d. https://www.libraryofsocialscience.com/ideologies/resources/griffin-the-palingenetic-core/. ↩
Shane, Scott, Cade Metz, and Daisuke Wakabayashi. ‘How a Pentagon Contract Became an Identity Crisis for Google’. The New York Times 30 July 2018. https://www.nytimes.com/2018/05/30/technology/google-project-maven-pentagon.html. ↩
Department for Digital, Culture, Media & Sport. ‘Consultation on the Centre for Data Ethics and Innovation’. GOV.UK. N.p., 13 June 2018. https://www.gov.uk/government/consultations/consultation-on-the-centre-for-data-ethics-and-innovation. ↩
McQuillan, Dan. ‘People’s Councils for Ethical Machine Learning’. Social Media + Society 4.2 (2018): 2056305118768303. SAGE Journals. https://doi.org/10.1177/2056305118768303 ↩
Plant, Sadie. The Most Radical Gesture: The Situationist International in a Postmodern Age. Routledge, 1992. ↩
Mordvintsev, Alexander, Christopher Olah, and Mike Tyka. ‘Inceptionism: Going Deeper into Neural Networks’. Research Blog 17 June 2015. http://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural.html. ↩
Akten, Memo. ‘Learning to See’. Memo Akten. 2018. http://www.memo.tv/portfolio/learning-to-see/. ↩
Marriott, Red. ‘Slogans of 68’. libcom.org. N.p., 30 Apr. 2008. http://libcom.org/history/slogans-68. ↩