Neural Interpellation

In 2016, when I co-founded the “Artists + Machine Intelligence” programme at Google AI, it had been 50 years since 9 Evenings, a seminal art event produced by Bell Labs’ EAT (Experiments in Art and Technology) programme. In that series of nine performances mid-century artists like Robert Rauschenberg, Merce Cunningham and John Cage collaborated with engineers from Bell Labs. The 9 Evenings programme was controversial. There were technical problems; some called it a failure. The pairing of big tech and the avant-garde art world begged for critique and – in light of bizarre AI phenomena like 2015’s DeepDream – another attempt. Could we also inject poetry into a corporate behemoth? Why not try again in the 21st century?

As I researched the EAT event, I came across the work of Fred Turner, a Stanford historian whose subterranean narrative links Silicon Valley to the acid tests of the 1960s and further back to a 1939 propaganda programme devised at the highest levels of the American state. The Committee for National Morale was convened by the Roosevelt administration so that anthropologists, educators, psychologists, journalists, art historians, doctors, social scientists, radio and motion picture industry, foreign affairs experts and others could work together to build wartime morale.

The commission was tasked with determining how to create subjects that would resist the hypnotic effects of single channel broadcast media. Having observed that populations could be induced to entrain to a single totalitarian message via radio transmissions, the American government sought to counteract fascism by designing media that appealed to and cultivated specific types of subjects, namely democratic, free-thinking, individualistic subjects.

The genius of the commission lay in their recognition that the structures of media produce subjectivity. Commission participants Margaret Mead and Gregory Bateson would go on to embody this theory in films and installations for MoMA. The formal structures they used would later be adopted by the hippie counterculture which gave rise to Silicon Valley, multimedia and the internet, which inherited, consciously or not, many of the commission’s priorities. Turner calls this democratic media form “surround media”, placing it in a separate category from single channel broadcast media like film and television.

In the philosophical framework of Louis Althusser, the concept of interpellation, often referred to as hailing, is described as a general and subconscious mechanism through which ideological systems present individuals with specific identities, which they are encouraged to embrace. The illustrative example involves a police officer hailing a person on the street. The officer calls out, “Hey, you there!” In responding to the call, one accepts the identity offered: that of a criminal. But this isn’t always so direct. Interpellation is performed by every ideological apparatus, such as the family, church etc. including technical structures. To think critically about artistic and media driven experiences means recognising interpellation while it is happening and becoming aware of the range of identities offered by a specific piece of technology, media or art.

Turner’s distinction between broadcast and surround media is useful for thinking about interpellation. But the arrival of AI and generative media expands available media types and the kinds of subjectivity they offer for interpellation. If one accepts that AI (which I include here under a larger category of “neural media”) is a new media type, and that we are now in the process of collectively negotiating the character of that media type, then one must extrapolate that we are also negotiating options for interpellation. Put another way, new kinds of subjectivity will be produced in us as we interact with the structures of AI. So it’s very important that we begin thinking now about what AI does and will do to our consciousness and subjectivity. What will AI bring out of us? Who will we become with it?

To make this easier, I have devised a taxonomy of media types that builds on Turner’s broadcast/surround, fascist/democratic shorthand. Within each category I have identified three properties: space, content and identity. For each media category these are unique, though the categories build on each other and their properties emerge from each other. Below, I describe each category and its relation to its predecessors and historical context.

As the chart above makes clear, broadcast media produce a centralised space with programmed content for demographic identities.

Despite their ephemeral nature, electronic media organise physical space, even before they organise sensory or mental space. In the case of radio and television, families and communities gather in private and “third space” environments like living rooms, bars, etc. to pay attention to a single message transmitted throughout these spaces. All participants experience the same broadcast. Even at the municipal, regional and national scales, content and narrative are centralised in private studios and broadcast companies. For example, for most of the 20th century, American television consisted of the NBC, ABC and CBS networks. The space produced by broadcast media is centralised, convergent and uniform.

This uniformity extends to the type of content broadcast over these media. Television and radio structured content in “programmes”. These were not executable packets of computer code but units of entertainment structured around formulas and patterns, like the half-hour family sitcom, the late night show and the evening news broadcast. Each programme came with its own subject-forms: the nuclear family with mother and father, the wise-cracking tuxedoed comic and his backing band, or the authoritative and beloved news anchor, a father figure making reality. Thus the physical, technical, business and content structures of broadcast media all exhibit the property of programmatic centralisation.

The centralised and low resolution identities offered by broadcast programmes are reflected in the ways that viewers are understood by the managers of these platforms, through market research. The Nielsen company’s market research and performance surveys were first begun in 1923. In 1936, the company acquired the rights to a device called an Audimeter that sensed and recorded viewer habits, allowing Nielsen to add a Television Index to its Consumer, Liquor and Pharmaceutical Index reports. The Audimeter’s precise sampling of viewer habits made ratings systems possible and enabled the advertising strategies that gave rise to modern adtech and consumer surveillance.

These systems were quite crude when it came to modeling user identity. For example, in the 1960s, Neilsen raters were broken into segments organised by age, gender, socioeconomic status and location. This specific matrix of identities may skew more consumerist than fascist, yet it reflects the limited range of subjectivities offered by the interpellating mechanism of broadcast media. In this sense, the identities offered by broadcast media are demographic.

Scheduled broadcasting for mass audiences began with radio station KDKA in Pittsburgh, Pennsylvania, in 1920, who broadcast the results of the U.S. presidential election. The UK’s BBC initiated regular television broadcasts in London in 1936. By the 1950’s radio and TV dominated the mediasphere and a new media form was ascendant, what Turner calls surround and I call immersive media. Note that this thirty year cycle of media type conception, maturation and mutation will appear in each media category.

The Committee for National Morale proposed an alternative to single channel media and its fascistic tendencies. Their new model embodied a democratic ethos of individualistic meaning-production within a spatialised media environment. By navigating a 360-degree media field, each viewer would be forced to interpret the images on offer. This individualistic narration of images was meant to increase democratic morale by allowing and even enforcing free choice. This was understood by the committee as a necessary mitigation of the anti-authoritarian tendencies of American citizens. By creating at least an illusion of choice, nationalistic messages could bypass American skepticism, producing a more subtle influence distributed throughout the media and social environment. In this sense, immersive media produce a distributed space providing an experience for constructing identity.

Almost a decade before the convening of the committee, Bauhaus’s Herbert Bayer had theorised a new perspective based on a 360-degree field of vision. According to Bayer, in contrast to fixed Renaissance perspective, the new perspective was characterised by dynamism, as the eye was compelled to move through space, changing its angle and position over time. His Airways to Peace exhibition at MoMA in 1943 celebrated air travel as a form of military dominance, and featured an inverted globe in which one was surrounded by the Earth and its flight paths. The wall text proclaimed: “Peace must be planned on a world basis. Continents and oceans are plainly only parts of a whole seen from the air. And it is inescapable that there can be no peace for any part of the world unless the foundations of peace are made secure throughout all parts of the world. Our thinking in the future must be world-wide.”

This global, distributed eye was repeated in Edward Steichen’s Family of Man exhibition at MoMA twelve years later. Hundreds of images by photographers working around the world composed “a forthright declaration of global solidarity... celebrating the universal aspects of the human experience.” The 1959 American National Exhibition in Moscow also took place beneath a dome and in a field of images. Charles and Ray Eames’ Glimpses of the USA portrayed America’s vastness, diversity and automotive infrastructure through a multi-panel film presentation.

The immersive mode was taken up by later countercultural figures raised on these exhibitions, including Allan Kaprow’s happenings, such as Yard from 1961, wherein visitors were invited to make what they would of a gallery full of used tires. The Merry Pranksters, originators of the famous “acid tests” added psychedelics to the immersive medium. And in a more technocratic mode, Stan Van der Beek (a Bell Labs alum) created Movie-drome in 1963, in which many films were projected simultaneously on the inner surface of a dome. All of these environments distributed sound and image around viewers, who became participants engaged in individualistic construction of meaning and subjectivity within the provided image set.

Many of these participants went on to join institutions like the Whole Earth Catalog and the Homebrew Computer Club, both of which gave rise to home computing and the early internet. Following the thirty year cycle of media conception, maturation and mutation, in the 1980s immersive media gave rise to not only home computing but also the internet and the next emergent media type, network media.

For 20th century viewers who had yet to experience the internet’s planetary image deluge, the content of immersive media were exotic and expansive. Yet when read from within a multipolar post-internet 21st century, these contents are clearly revealed to be pro-American, pro-democracy propaganda. The surround environment and its repeated dome structure becomes a form of symbolic enclosure that limits possibility, presaging the global order to come, in which airways bring peace not through military dominance but through trade agreements and cultural hegemony. The boomer countercultural logic of identity construction through appropriation of Eastern or Indigenous artifacts and practices follows naturally from the Family of Man exhibition. Just as the backlash to appropriation follows naturally from the network media that followed immersive media and the multipolar postcolonial world.

Media types do not fully replace each other. Even now, television persists in aggregating attention in programmes and demographics. Similarly, despite the emergence of network media, contemporary immersive forms like VR and AR, MSG’s Las Vegas Sphere, Burning Man or Van Gogh Exhibit: The Immersive Experience all provide experiences for subject production in a distributed field of media. The media feeds we scroll inherit the spatial distribution and globalising ideology embodied in the immersive exhibitions above, yet take them further via the properties of network media. They rely on immersive media’s narrative of individualisation through media consumption, but bring it to its logical conclusion.

The domes and projections of immersive media brought the globe to America and America to the globe. Like television, these spaces were centralised yet within them images were distributed. The network media that follow decentralise the axes of image and information exchange. Home computers, laptops and smartphones connected by networks turn the image dome into a hypersphere that covers every surface of life. Each device has its own form factor and interaction paradigm, what is common among them is their role as nodes in a circulatory network of information distribution.

Visualisations of the internet’s structure have an organic quality reminiscent of veins, lungs and even neurons. Yet despite the neuron-like structure, network media act more like circulatory organs. To distribute content via email, message boards, blogs and social media users must act as nodes in the network, filtering feeds and pushing their own and others’ content into the network. They are the circulatory force that moves content around the network. Because of this, networks favor viral and memetic media. In this sense, network media produce a circulatory space for memetic content and fractal identities.

The term “meme" was coined by the British evolutionary biologist Richard Dawkins in his 1976 book “The Selfish Gene.” Dawkins used the term to describe an idea, behavior, or style that spreads from person to person within a culture. He derived the word “meme” from the Greek word “mimeme,” which means “that which is imitated,” and intended it to be a concept analogous to the biological gene, but in the realm of cultural evolution. Memes, according to Dawkins, are units of cultural information that replicate and evolve in a manner similar to the natural selection of genes in biological evolution. Within network media, this evolution is driven by neurochemistry via slot machine-like variable rewards UX patterns, by platform incentives like revenue-sharing and influencer campaigns, and by quantitative transparency via follower and impression counts on posts and profile pages.

The easy availability of subcultural artifacts online has led many to remark on the death of the underground and counterculture. Yet, the internet and social media have produced numerous memeplexes associated with various subcultures. In fact, network media produce identity fractally in recursive relations of hyperindividuation. These identities slip and slide between demographic and constructed identity forms, creating a molecular foam of nanoidentities.

Perhaps the most obvious expression of this is the political compass meme. The political compass meme is a popular internet meme format that humorously categorises various ideologies and political figures using the framework of a 2D map or compass. One axis represents the economic spectrum (left for more socialist or communal policies and right for more capitalist or individualist policies) and the other axis represents the social spectrum (authoritarian at the top and libertarian at the bottom). In the meme, each quadrant of the compass represents a different political ideology: Authoritarian Left, Authoritarian Right, Libertarian Left and Libertarian Right.

The meme often places political figures, pop culture references and stereotypes in relation to these quadrants to create a satirical or insightful commentary on their perceived political positions. It’s used to mock and critique the oversimplification of complex political views but also as a game board on which to collect and organise manifold online political and cultural identities. This fractalisation of identity and subjectivity is reflected in the larger multipolar geopolitical and cultural environment. If immersive media prefigured the rise of the neoliberal world order, network media prefigured its collapse.

What kind of memetic monsters evolve in a primordial stew made of fractalized identities, multipolar geopolitics and viral media? Conspiracy theories, of course. Network media subject production and interpellation are deeply intertwined with bespoke political histories and cosmologies. The hallucinatory tendency of neural networks is anticipated in this emergent and generative phenomenon. If it is the nature of networks to dream, we are caught in the dream of the internet.

Marshall McLuhan famously wrote that “the 'content' of any medium is always another medium.” In the case of neural media, the content is all of network media. This is quite literally the case with large language models and image generators trained on massive corpuses of text and imagery scraped from the web. It is also true in an abstract sense. In contrast to users of network media, users of neural media act on (rather than in) the network. Instead of acting as nodes of a network that circulate content, they interact directly with the totality of the neural network.

It may be fruitful to think of neural media as an emergent property of network media. Consider the AI-based product recommendation systems that power Netflix and Amazon. Trained on the behaviour of internet users (that is, users of network media) these neural systems were slowly integrated into platform backends over the last decade. The introduction of the algorithmic social media timeline in the early and mid-2010s quietly announced the mutation of network media into neural media; the two were from then on inseparable. AI chatbots and image generators are the fruiting body of neural media always embedded in network media.

In recent AI discourse, the term hallucination has come to refer to false information “made up” by neural nets. This comes from analogy to a human brain hallucinating an unreal sensation or experience. Yet in both cases, the term ought to refer to all generative neural activity. Generating neural activity in a machine model is always hallucination. The process is the same, whether or not the output corresponds to known facts. Similarly, human brain activity is largely hallucinatory. We do not experience reality directly but as a neurally mediated subset of sensory input mixed with internally generated elements like attention and memory. The content and output of neural systems is fundamentally hallucinatory.

High-dimensionality and embedding are tightly intertwined in machine learning. (For an explanation of high-dimensionality, see the previous essay in this series). In an AI system, an embedding refers to a method of representing data, typically high-dimensional data like words, images, or complex user profiles, in a lower-dimensional, continuous vector space. This representation is designed to capture essential features and relationships in the data. It is also a form of identity. If broadcast media interpellated subjects through demographic clusters and programmed content, immersive media did so through an experiential mediasphere, and network media did so through fractal circulation, then neural media interpellate subjects through hallucinatory high-dimensional embeddings. The immersive media hypersphere that surrounds us via smartphones and other devices is increasingly animated by AI systems that reflect us back to ourselves as embeddings in their high-dimensional space. In this sense, neural media produce a high-dimensional space with hallucinated content for embedded identities.

Unlike previous media types, neural media sense back. Their hallucinatory or generative nature enables a two-way exchange. AI learn from us and we are interpellated by them. For example, we might converse with a chatbot that models our interests and behavior in a user profile while generating or hallucinating answers to our questions. Like the canonical shopper who discovers she is pregnant through product recommendations, we are interpreted within a neural embedding space and the system adjusts its behavior and content to reflect this interpretation back to us.

Following the thirty year pattern of media types described above, we can estimate that we are in the very early stages of neural media and that it will continue to evolve for a few decades before mutating into something new. I have focused on AI here, but chose the term “neural media” so that this type could be expanded to include brain-computer interfaces and any other media type built on neuronal structures and neural networks.

It is easiest to influence a new medium while it is young, just as it is easiest to determine where a tree grows at the time that it is planted. Neural media inherit many of the contexts and characteristics of the media that precede them, notably, an immersive quality and an interpellating function. If we want to choose how we are interpellated by neural media, that is, if we want to shape ourselves and our subjectivity through the apparatuses of neural media, now is the time to work. Doing so will require that we embrace the unique characteristics of neural media through literacy and experimentation, with a critical sense and a clear understanding of what exactly we want to become.