In late 2018, a painting produced with machine-learning algorithms was sold by Christies for 432,500 dollars.¹ Portrait of Edmond de Belamy (2018) heralds a questionable trend, since the work by Paris collective Obvious does not move beyond an aesthetic-conceptual level. It uses algorithms as a mere medium, although the progressive development of artificial intelligence brings about profound changes that call for reflection. The question of novel human-machine relationships, for example, seems extremely relevant. Numerous art projects are currently addressing this interface – set between a sense of being at the mercy of others and scope to influence the changing design and perception of reality as a result of AI.
Pierre Huyghe's installation UUmwelt (2018) is grounded partly in the speculative notion that AI will soon be able to read our minds. Mysterious pictorial mutations can be seen on large-format monitor walls, seemingly referring to coloured objects or to animals that are never completely tangible. The views stagnate and at the same time endlessly keep on forming and re-forming. In a laboratory in Kyoto, Huyghe showed a person images and invited them to imagine further images, all the while recording their brain activity using functional magnetic resonance imaging. Subsequently, artificial intelligence was deployed to analyse the brain waves and searched image databases for similar motifs in order to determine which images the person might have envisaged. The process of machine learning, the modelling of visual approximations, becomes visible on the screens.
The distorted images were generated by artificial neural networks using a Generative Adversarial Network (GAN), algorithms that learn without human oversight. Since the technology was launched in 2014, it has become a popular tool for artistic projects, including the aforementioned painting. More critical GAN-based works include Constant Dullaart's DullDream (2015-ongoing)², in which portraits are stripped of their individual characteristics, while also taking a stand vis-à-vis Google's pattern recognition software DeepDream; Trevor Paglen's series Adversarially Evolved Hallucinations (2017) alludes to the capitalist monstrosity of technology with specially developed taxonomies; and Janek Simon's Synthetic Poles (2019) creates almost photorealistic images of people who do not exist. In an expanded sense, the technology is also used in Marta Revuelta's AI Facial Profiling, Levels of Paranoia (2018). It picks up on the worrying tendency towards use of machine-learning algorithms to infer character traits from facial features. The work consists of a facial recognition system that draws conclusions about an individual's ability to use weapons on the basis of their appearance and thus makes statements about the potential threat that an individual may pose.
In Pierre Huyghe's UUmwelt, the machine images form part of the exhibition's communication system. External factors such as sound, light, temperature, humidity, flies and visitors are detected using sensors and influence the speed of the continuing mechanical hallucinations. It is about an ecosystem that interconnects humans, animals and technology through feedback loops, functioning as an open system in which humans do not necessarily have the upper hand.
In a similar direction, Se ti sabir (2019), a short essay film by James Bridle, points to intelligence as a process that reaches beyond the human being. "Intelligence [...] is itself possibly a kind of network that exists as connections between things, bridges and communications across them, rather than purely reflex motions within individual brains."³ The word "sabir" in the title comes from the lingua franca spoken in the Mediterranean from the Middle Ages to the 19th century, a Romance-based pidgin language that emerged from contact with speakers of non-Romance languages and was commonly used for trade and communication. "Sabir" means "to know." Over time, the word became a greeting that alluded to the question of whether it was possible to communicate with each other.
Bridle's interest in the lingua franca began when he reflected on how nowadays AI, particularly as an application for language translation, contributes to people becoming less and less dependent on their own understanding. The machine mediates, and we rely on its competence. Although deploying technology like this promises advantages, our own abilities are successively replaced; we do not understand the processes unfolding in the machine, nor do we learn anything new. However, Bridle does not want to reduce AI solely to this idea. Instead, he looks for a way to understand machine intelligence not so much as an advocate, servant, tool, or weapon, but more as a new voice that joins the cacophony of non-human voices and intelligences that surround us. In this, he picks up on approaches from technical philosophy, such as Bruno Latour's actor-network theory or Gilbert Simondon's open machine associations. Indirectly, Bridle also ponders what form might be assumed by more intelligent collaborations with AIs, approaches not shaped predominantly by ideologies, market interests, surveillance and opacity, exploring these phenomena in detail in his book New Dark Age: Technology and the End of the Future (2018).
When it comes to how intelligent our collaborations with AIs are today, especially from an ethical perspective, the question rapidly arises of whether the technology itself should perhaps function differently. Are statistical pattern recognition and automatic conclusions, requiring no in-depth knowledge of the context - the approach that is currently gaining ground under the machine learning label as part of AI and serves as a springboard to deep learning - really the best solution? Some argue that a different kind of AI is needed. Scientist and artist Luc Steels argues for "responsible AI" and advocates a combination of pattern- and knowledge-based approaches in AI research.⁴ Steels coined the term "behavior-based robotics", which is currently not in the foreground in development of AI algorithms. In Sprachagenten (2018) Armin Linke & Giulia Bruno talk to Steels and show archive material about his experiments with robots that develop their own language when interacting with each other. Steels' talking heads experiments between 1999 and 2001 took as their point of departure the way in which language, rather than being a rigid system, is constantly evolving. Categorization of the world and the language of simple robot-like agents are also not pre-programmed; through learning games in an environment that, for example, contains obstacles, the agents construct or learn about these for themselves. Steels is currently working on the Odycceus (2017-21)⁵ project, which investigates how and why social media have recently supported language use aimed at disinformation, divisive discourses, and postfactual politics. What influence does technology have on this significant change in human discourse?
Franz Wanner's installation PUBLIC VOID (2018) deals with the influence of algorithms on the digital public realm. The Internet may offer a huge range of potential choices, yet AI algorithms contribute to shaping our information field. Images, videos and texts are projected onto a large surface suspended in space: a flow of information from social media profiles whose users, contrary to standard practice, have consented to recording of their content. Leaked algorithms from Cambridge Analytica that allegedly contributed to the current US president's election in 2016 by microtargeting social networks were used as a code. Wanner uses the algorithms like found footage and shows that the same code can convey very different stances. Searches were carried out with the following selectors in online posts from Russian social networks: millennium baby, micro target, troll factory, click worker and data cleaning. Contrasting with Cambridge Analytica's manipulative approach, a critical Internet public realm is thus revealed. The work was created during the election year in Russia, and in this context "millennium baby" for example alludes to the generation of first-time voters lured with special election advertising. Unlike the situation in the real world, in the installation it is easy to head to the algorithmic control page. While the filtered data stream is projected onto the semitransparent surface, the code is transferred to its flipside. Both are loosely interlinked. The transparency of the chosen double projection points strikingly to the lack of transparency in the digital public realm.
The often limited influence of users in an online world organized by algorithms is the subject of a second screen with a touch function. A red knob is visible and below it the word "reset" with the instruction to press the knob. "Swiping the display with your finger does not necessarily mean taking part in what is happening", as Wanner aptly puts it.⁶ Activating the knob simply leads to the word changing into Russian. To understand the piece, it is also important to know that the US Secretary of State presented a reset button to her Russian colleague in 2009 as an object symbolising a positive relaunch of diplomatic relations - but it contained an inaccurate translation. The Russian term chosen to translate "reset", "peregruzka", actually means overload.
Hito Steyerl's film installation The City of Broken Windows (2018) refers to the questionable expansion of AI's competencies and fields of activity. Specifically, it is about the use of private security technology in the urban sphere. This development is contrasted with the potential to become visible as human actors. Whereas in one video a security company trains AI to recognize the sound of breaking glass - by having engineers break countless panes of glass -, in a second video a community ensures that broken windows in North American cities are simulated by painted reproductions. These trompe-l'oeils are intended to function in the sense of the "broken windows" theory: although this does not eliminate the causes of crime, it does help stop the buildings from becoming entirely derelict. While some rely on technological progress, which certainly contributes to making cities not just smarter, but also more "gentrified", others combat urban decay in poor areas using grassroots methods. The space between the two videos is pervaded by a ringing sound, an AI's flawed attempt to reproduce the sound of windows shattering.
Kate Crawford and Vladan Joler show in their artistic research Anatomy of an AI System (2018)⁷ the complexity and problematic aspects of the processes underlying the smooth surface of an Amazon Echo, the voice-controlled digital home assistant. The tagline (an anatomical map of human labor, data and planetary resources) makes clear the intention underlying this detailed map: the issue is predatory exploitation of human labour, data and resources, taking Amazon Echo as an example. A diagram system, divided into sections addressing the device's birth, life and death, and an essay shed light on the global ecological and economic interrelations associated with this convenient technology. These include the world's largest lithium reserves in Bolivia and their extraction in Uyuni salt lake; voluntary data work by users that helps train the device's neural networks, fostering all-encompassing quantification; the exploitation of outsourced intellectual work to developing countries, in which data is categorized for machine learning; and informal workers who dispose of toxic waste. "The true costs of these systems - social, environmental, economic and political - will remain hidden for some time to come. [...] And yet you use the system every time you address a simple vocal command to a small cylinder in your living room: 'Alexa, what time is it?'"⁸ Thinking along similar lines, Hito Steyerl recently called for AI perhaps only to be used for art in museums and galleries, keeping it isolated from the world until we know about all its consequences.⁹

 

Translated by Helen Ferguson

 

---