The Labour Vanishes: AI, machine learning and automation

We live, we are told, in a golden age of AI, with major tech companies like Google regularly and loudly announcing that they are getting close to “Artificial General Intelligence” (AGI) and large language model based chatbots like ChatGPT being regularly subject to dubious claims about their capabilities. But what is AGI and why is AI so important right now?

AGI is a computer system capable of performing any intellectual task that a human being can. Researchers Timnit Gebru and Émile P. Torres argue in a recent conference presentation on AGI that the drive to create AGI is part of an ideological cluster that Torres defines as “the TESCREAL bundle”. The TESCREAL (Transhumanism, Excropianism bundle is, essentially, a set of beliefs common to many rich and powerful figures in the tech world that humanity is on the verge of making such major breakthroughs in technology that the imaginings of sci-fi authors about Clarke’s “sufficiently advanced technology[…] indistinguishable from magic” will so be a reality. Gebru and Torres show that the ideas of the TESCREAL bundle are closely associated with eugenics, racism and neoreactionary ideology.

In one sense, AGI is important to the tech industry because of the role it plays in these millenarian fantasies; we must create the good AGI that will be a benevolent machine god to save us from the bad AGI that will destroy us all. However, I don’t think this is entirely sufficient to explain the recent explosion of resources and publicity around AGI. The TESCEREAL bundle features many other odd obsessions, such as achieving immortality through niche, pseudoscientific life extension techniques, which have not seen anywhere the explosion of mainstream interest that AGI has. In order to understand what makes AGI special of all the TESCEREAL bundle’s obsessions, I have to talk about an 18th century magic trick.

In 1770, the Hungarian inventor Wolfgang von Kempelen created the “Mechanical Turk”, a machine that, according to von Kempelen, could autonomously play chess against an human opponent. The machine is famous as a fraud, it was operated by a human being hiding within, but in another sense, it was a kind of magic trick. The illusion of something impossible was created. In 2005, Amazon launched its own Mechanical Turk, a service where employees could request tasks be carried out via a web service and those tasks would then be acted on by an obfuscated group of “crowdworkers,” once again an illusion is created of a task being automated, albeit with a degree of honesty. At first, it appears that, like a stage magician, Amazon perform the illusion with the understanding that the audience knows that they’re watching a trick.

One popular use of Amazon’s Mechanical Turk service is labelling machine learning data. The statistical models most often referred to when we talk about existing AI require very large numbers of training examples (as well as a large number of test examples, if you want to do things properly). A company seeking to make a computer vision model, for example, might use Mechanical Turk to label a huge library of example images at relatively low cost. The service being sold here is the labour power of individual people, which is then crystallised in the parameters of a statistical model, fitted to produce similar outputs in response to inputs to those of the training data.

Amazon is not the only company to imitate the Mechanical Turk. Last year the shopping startup Nate was caught out when it transpired that what they’d claimed to be AI filling out forms for shoppers was in fact workers in a data centre in the Philippines. While Nate was excoriated for misleading investors (a far more heinous sin than exploiting labour in the global South), it isn’t the only tech company running services often assumed to be AI that are in fact run on highly exploited labour. Content moderation on a number of major websites, including Facebook, TikTok and ChatGPT, is often assumed to be wholly automated in the first instance, in fact much of the labour is performed by workers in the Global South on extremely low wages, who hit headlines recently as they unionised, demanding better working conditions and support in dealing with the effects of constant exposure to the worst of the internet.

Meanwhile, the ability of automated tools to emulate human tasks to some degree provides an opportunity for bosses looking to deskill workforces. While tools like ChatGPT are clearly not there yet when it comes to some tasks, that hasn’t stopped companies from wanting to be allowed in principal to use them for tasks from writing advertising copy to entire screenplays, with the understanding that deficiencies in the model’s writing can be made up for by human labour after the fact. This threatens to deskill entire workforces, with the creative and potentially fulfilling aspects of many jobs taken out of the hands of workers and the ability of those workers to collectively bargain for better wages and conditions severely undermined.

Whether it’s truly automated or merely colonial exploitation mimicking the form of automation, “AI” is not simply a technology in popular discourse; it is an ideological tool in which human labour is obfuscated behind layers of technology. In this way, the tech industry can create the illusion that it is moving towards the Star Trek future promised by the TESCREAL bundle while making its money the way capitalist industries always have; extracting value from labour. Like other forms of automation before it, the technology of AI in the hands of capital is simply a means of extracting ever more value from human labour at less and less cost (and therefore, with fewer stable jobs and less payment for human labourers).

If you’ve enjoyed reading my thoughts on artificial intelligence here, you might also enjoy my new podcast with Rhi Red of the YouTube channel Spoopy Spoonie; Mad Social Science, a podcast where the two of us talk social science, science and politics. Our first episode will be on AI, eugenics and automation, releasing some time in the next month or so, follow us on PodBean or Twitter to listen to it when it drops.

Sources

• Amazon Mechanical Turk, AWS introduces a new way to label data for Machine Learning with MTurk
• Eliot Higgins on The Gray Zone’s use of “AI” for writing, Twitter
• Maya Kosoff, Peter Thiel Wants to Inject Himself With Young People’s Blood
• Brian Merchant, The writers’ strike is only the beginning. A rebellion against AI is underway, LA Times
• MMitchel_AI on claims of outlandish translation ability from Google Bard, Twitter
• Malique Morris, ‘AI’ Shopping Startup Exaggerated Tech Capabilities to Potential Investors, The Information
• Billy Perrigo, Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic, Time Magazine
• Billy Perrigo, 150 African Workers for ChatGPT, TikTok and Facebook Vote to Unionize at Landmark Nairobi Meeting, Time Magazine
• Timnit Gebru & Emile P. Torres, Eugenics and the promise of utopia through AGI, presented at IEEE SaTML 2023
• Wikipedia, Artificial General Intelligence
• Wikipedia, Amazon Mechanical Turk
• Wikipedia, Mechanical Turk