Skip to content

When AI Is on the Label but Wage Dumping Is on the Inside: How the AI Hype Devalues Human Labour

Not everywhere where AI is on the label, AI also does (all) the work: Under the name Amazon Go and Amazon Fresh, the e-commerce giant operates cashierless supermarkets that, according to their own description, use “computer vision deep learning” and “sensor fusion” to determine the products that customers take from the shelves and, on this basis, create fully automated digital receipts as soon as the customer leaves the shop.1Cf. video “Introducing Amazon Go and the world’s most advanced shopping technology”, URL:, accessed 29 October 2023. In late March, Emmanuel Maggiori pointed out the observation that the time interval between leaving the store and receiving the receipt varies significantly – from a few hours to just over two days. If the process were “fully automated” as Amazon claims, such fluctuations could not be explained. Maggiori therefore suspects that Amazon is using a “fake-it-till-you-make-it approach” here and that people are manually evaluating the video footage of the purchases.2Cf. Maggiori, Emmanuel, Amazon’s cashierless stores: artificial intelligence or major deception? URL:, accessed 29 October 2023.

Kaedim: AI or human artists?

At the beginning of September, accusations were made against the AI start-up Kaedim, which go in a similar direction: Kaedim claimed to use a self-developed machine learning algorithm to automatically convert customers’ two-dimensional illustrations into 3D models.3Cf. Cox, Joseph; Koebler, Jason, Buzzy AI Startup for Generating 3D Models Used Cheap Human Labor, URL:, accessed 30 October 2023. For a monthly subscription price of USD 150, 10 models can be generated in this way. However, insider sources told 404 Media that the output of the AI model was of such poor quality that instead of a tree, for example, only an unrecognisable blob was generated. In a job advertisement, the start-up therefore advertised worldwide for freelancers who are able to “create a 3D model from a 2D drawing within 15 minutes of a request”. When asked about this practice, Kaedim founder Konstantina Psoma spoke of people being used for “quality control” at Kaedim. 404 Media, however, cites sources that humans did the actual work for a payout of between 1 and 4 USD per model.4Cf. ibid. Meanwhile, Kaedim has adapted its website and writes that machine learning and a team of in-house artists work together to produce assets within minutes.5Cf., accessed 30 October 2023.

The AI hype and its pitfalls

Given the AI hype, I dare to predict that in the coming months we will see a number of other examples of start-ups and companies claiming to use AI when in fact humans are doing the work. Already at the time of the blockchain hype, the share price of a beverage company went through the roof after it changed its name from “Long Island Iced Tea Corp.” to “Long Blockchain”.6Cf. Kosoff, Maya, Iced Tea Company Changes Name to “Long Blockchain,” Stock Immediately Skyrockets, URL:, accessed 30 October 2023. Long before that, it was the dotcom hype within which the promise of new technologies (many of which were not well understood even by investors) could quickly generate market and venture capital. The temptation to label something as “AI” when there is actually no (functioning) AI working under the bonnet was and probably still is great.

This does not refer to the “Wizard of Oz” method, which is quite legitimate in development: here, people are used in the context of prototypes for processes that are to be automated in the future, in order to test UX designs, for example.7Cf., retrieved on 30 October 2023.

Social and ethical consequences

It is only problematic if the hoped-for automation is never implemented or cannot be implemented and thus scaling can only be achieved through cheap human labour, i.e. wage dumping. This is problematic from several points of view:

  • First, users of this system unknowingly support wage dumping: in the case of Kaedim, a 3D artist presumably received between 1 and 4 USD per model, while the client paid around 15 dollars per model.
  • If the client actually considers human labour to be a computer-generated product, human labour and art is no longer recognised as such and is thus devalued. An artist whose work is sold as AI-generated is deprived of recognition as an artist.
  • Furthermore, this practice creates a distorted image of the actual possibilities and limits of AI, which ultimately makes the use of AI for tasks for which it may not even be suitable more likely in other contexts.

The reverse trend: Humans replace AI

Therefore, the broadest possible social discussion of AI is necessary: Anyone who informs himself about AI within the scope of his possibilities and tries out AI tools will gain a better understanding of the capabilities, but also of the (partly inherent) limitations of AI. In this context, this understanding would not only help users, but also founders and investors interested in sustainable business models instead of the next flash in the pan.

In the much-discussed debate that AI will replace human labour and jobs, it is ultimately a remarkable, albeit cynical, plot twist when humans actually do the work of the supposed AI, but remain invisible themselves.

Leave a Reply

Your email address will not be published. Required fields are marked *