An ever-growing number of digital and non-digital companies and governments embed forms of artificial intelligence (AI) in their technical infrastructure. The dominant AI technique, Machine Learning (ML), is based on the paradigm that computer systems can emulate humans when provided with enough “training data”. In most cases, this training data is the product of human labour, and the way in which it is collected is problematic. Data collection procedures are opaque; business models fail to account for the value of the labour being contributed by individuals, and consent to collect and use this data is not explicitly requested.

Particularly widespread are systems in which the training data is gathered simply by virtue of users voluntarily engaging with digital platforms and online tools for purposes other than contributing data to a training set used by AI systems. For example, an internet user filling out a reCAPTCHA is actually generating data that is then collected and used for various Google AI applications. As another example, Spotify’s music recommendations are informed by many different types of human input, including user interactions with the platforms (e.g. the music they like or skip, the playlists they create) and music reviews and comments written by music journalists and aficionados on blogs and forum that are scraped by Spotify bots to extract music taste automatically.

Individuals interacting with AI-powered systems are commonly unaware of the ongoing extraction of value as they volunteer their preferences, intelligence, and behaviours to AI owners. They are also commonly unaware of how their information and actions generate corporate profits. Using a Marxian lens, we frame these extractive practices as forms of labour and specifically immaterial labour that has an external value that individuals are steadily but inadvertently producing. Consequently, they cannot use the collective power this affords to make demands of their ‘employers’. The Marxian approach suggests a classification of knowledge class for all individuals whose interactions with AI generate value that is expropriated from them. Framing this issue using the theme of the conference, an interesting world to come would see new political struggles of the knowledge class whose work is exploited by digital capitalism and new ways to break the circuits of surplus value/surplus data as the engine of this type of capitalism.