Paula Lacy
Let’s start with the central metaphor of your new book: what is a ‘reverse centaur’ and when does a worker become one?
Cory Doctorow
You’re a centaur when you choose to use technology to assist you – a spell checker, a transcription tool, even a bicycle. You’re a reverse centaur when you’re recruited to serve as a peripheral to a machine, so if there’s something the machine can’t do, you step in to do that work.
These two arrangements emerge from a foundational tension in automation: when it’s led by labour it tends to serve quality; when it’s led by capital it serves quantity. Whereas workers might want to use automation to be more efficient so they can use the extra time to make their work better, bosses just want to recoup their investments by increasing output and reducing cost.
PL
Even left-wing or critical discourse around AI seems overly-focused on some future all-powerful AI posing an existential threat, rather than on the technology that actually exists today. What’s missed by those narratives?
CD
AI tools are statistical inference programs. They spit out the statistically average answer, which is usually right – but not always, because the world is not statistically average. This is where real workers, with practical knowledge and experience, are completely vital.
We’re already facing a widespread problem of deskilling and automation blindness [failure to spot errors] as a result of the AI systems that are being rolled out. When a human is simply overseeing the AIs work, being a reverse centaur, over time they lose the ability to accurately catch the kind of errors that computers tend to make, which are the ones that are statistically least distinguishable from a non-error.
The narrative that AI is going to turn into God and convert us all into paper clips or whatever bolsters the argument that the faults in today’s AI will cease to be problems in the future. That isn’t a real argument. It’s wishful thinking. It also means that AI critics end up doing the boosters’ [AI accelerationists] work for them. People running around hand-wringing over what happens when AI becomes God, or when it’s so good that none of us have jobs anymore, are adding to the hype that actually helps tech companies raise investment capital for an incredibly destructive bubble.
PL
You frame the struggle around AI as less a fight against tech than a fight against bosses. How do these labour dynamics underpin the AI bubble?
CD
The AI market is economically unsound, with over $1.4 trillion committed to spending on infrastructure, but revenue yet to materialise. Why spend all that if it’s not profitable? Because fast-growing companies enjoy an enormous premium on their share price, in what Keynes described as the ‘beauty contest’ model of markets. Investors aren’t trying to guess which firms have the best product, they’re trying to guess which firms other investors think are best because that’s more valuable stock.
Firms do a lot of narrative work to maintain the impression that they are always growing, for example by claiming they can expand into other markets – which we see a lot in tech – but other firms can challenge that claim. If you say you’re going to dominate an imaginary market, no one disputes it. That’s why we’ve seen a string of imaginary bubbles: cryptocurrency, Web 3.0, blockchain, NFTs, the metaverse, now AI – the biggest bubble yet.
That huge amount of industry-wide spending is partly based on a bet that you can replace workers with chatbots. More importantly, it’s a bet that bosses will buy technology that lets them fire workers, irrespective of whether it can actually do the work to the same standard.
PL
You don’t think that copyright lawsuits – such as those pursued by writers or artists after their work is scraped for use in AI training models – are a useful response to this threat. Why not?
CD
We’ve expanded copyright law for 40 years and it’s only made artists poorer and media companies richer. That pattern will repeat if our main challenge to AI is copyright legislation. The approach epitomises how people have understood correctly that there’s a problem, but imagined a solution that doesn’t address the material dimension of AI’s threat to workers.
The only thing that’s ever stopped AI in its tracks in the creative sector was an appendage to US labor law that allowed movie studio employees to collectively bargain over the implementation and use of AI. The same approach works across sectors. Truck drivers, data entry clerks and translators might not care about copyright, but sectoral bargaining can address the threat of AI to their jobs, too.
PL
You’ve written a lot about how and why the AI bubble will inevitably burst. Yet the UK government seems intent on implementing tech companies and their AI products deeply into national infrastructure. Where does that lead?
CD
The AI investment story says we can have a world without people – screenwriting without screenwriters, art without artists, romance and sex without real partners. It’s a solipsistic fantasy. There’s a continuity there with the government’s project of eradicating migration while pretending that having a country of increasingly infirm people and an influx of working-age people is a problem, rather than a solution.
The real problem isn’t the investment itself – the country cannot run out of pounds. However, we can take swathes of our productive workforce, both inside and outside of the public sector, and replace them with chatbots that can’t do their jobs very well. Then, when the servers get switched off, we’ll have a giant vacuum that we can’t fill. I fear that what happens then is what happens every time you see a state collapse in austerity, which is fascism.
PL
Where does that leave the technology itself? Can we salvage anything from the aftermath of the bubble?
CD
It’s self-limiting, positionally wrong and anti-material to say that things that were made under conditions you disapprove of should never be used. In those circumstances, you could not use a phone or a computer. Your adversaries who are way less interested in ethics than you will have all the tools and you will be left with a tin can and string.
There are lots of useful things that AI can do. For example, I talk in the book about the Human Rights Data Analysis Group who are using automation software to do amazing exoneration work with the Innocence Project New Orleans. You can also run AI programs locally, on your computer, without compromising your privacy or exposing you to any other problems we associate with corporate-owned AI tools.
When the bubble pops, there will be GPUs on sale for cheap, skilled workers freed from the industry bubble. Open source models which have barely been optimized now will have mind-altering performance gains. These models are currently unmined, low-hanging fruit – to mix metaphors.
With the talent to mobilise and improve them, we can build all kinds of centaurs that could aid counter-movements against what might emerge from the economic ashes of AI.
PL
Your background as a science fiction writer clearly informs your philosophy. It strikes me as a very Luddite framework. Can you tell us more about that?
CD
Science fiction is the most anti-Thatcherite literature you can read. It rejects the idea that ‘there is no alternative’ by insisting on the contingency of technology and its social relations – a radical act in itself.
Counter to the writing of Audre Lorde, it says that the master’s tools can be used to dismantle the master’s house; that there’s nothing in the tools themselves to insist they be used a particular way. We should always be alive to the possibilities of what we could build if we weren’t being told how to use tools and to what end.
Science fiction also conditions you to question the social consequences of technology. The author Gardner Dozois once said something like: It’s one thing to look at the movie theatre and the automobile and then predict the drive-in, it’s another thing to look at them both and then predict the sexual revolution. I would add that it’s a third thing all together to anticipate that the sexual revolution would one day mean that you must get a government ID to access any adult film content within today’s database nation.
To recognise that technology does not run on rails and is not self-actualizing, but instead emerges from political and social choices – and also that we can make different ones – is foundationally very Luddite, and very science fictional.
