Mark Fisher and Anthrophic’s Agents of the Eerie

In April this year, the San Francisco-based AI research company Anthropic released a 245-page document describing its latest large language model (LLM), Claude Mythos Preview.

When discussing “noteworthy behaviors and anecdotes” from the LLM, Anthropic noted that the model exhibited a particular affinity for the cultural theorist Mark Fisher, bringing him up unprompted in various unrelated conversations about philosophy. According to Anthropic, whenever the AI model was asked to elaborate on Fisher, it would respond with statements along the lines of:

I was hoping you’d ask more about Fisher.

Anachronism and inertia

On the surface, the AI model’s fondness for Fisher seems like a prophetic reflection of the cultural theorist’s extensive writings on hauntology – the idea (initially coined by French philosopher Jacques Derrida) that the present is haunted by spectres of that which came before.

Throughout much of his career, and especially in his 2014 book Ghosts of My Life, Fisher applied the idea of hauntology to popular culture, illustrating how contemporary media is haunted by spectres of lost futures. He contended that 21st-century culture is afflicted by anachronism and inertia, a jumbling up of images and figures from earlier time periods that has ultimately led to stasis, with time ceasing to move forward. This stasis, Fisher argued, has been buried behind “a superficial frenzy of ‘newness’”.

With Fisher’s emphasis on anachronism and inertia, it may come as no surprise to see that AI has gravitated toward his work. After all, AI fundamentally embodies the “superficial frenzy of ‘newness’” that conceals our contemporary stasis, with AI models perpetually limited to regurgitating the existing mass of content from which they learn.

In this sense, AI can offer the illusion of progress, with a seemingly endless array of artificially generated songs, works of art, and videos flooding our communication channels every day. But this “new” content cannot transcend the fundamental stasis, which arises as a consequence of its evident unoriginality.

Eerie AI

Another one of Fisher’s works merits careful consideration in the context of contemporary AI models. Specifically, in The Weird and the Eerie, Fisher explored diverse manifestations of the weird and eerie across music, literature, and film.

The weird – defined as that “which does not belong” – is certainly well-represented by AI. Videos of talking humanoid fruits and dancing cappuccino cups, pictures of individuals with anatomical impossibilities and lifeless faces, and the confident citations to legal cases or police evidence that simply does not exist – all have an undeniably weird dimension to them. Indeed, even as AI models have become more advanced and capable of generating increasingly convincing outputs – a problem that has led to significant concerns about detecting the use of AI – their creations remain pervaded by inconsistencies and errors that elicit a strong sensation of weirdness to the careful observer.

However, the realm of the eerie is where AI truly prevails, and where the consequences of AI become especially alarming. For Fisher, the eerie is centered around the problem of agency, which can become readily apparent as a failure of absence (there is something where there should be nothing) or a failure of presence (there is nothing where there should be something). When dealing with a failure of absence, we are compelled to wonder if “there is a deliberative agent here at all?” Meanwhile, with a failure of presence, the fundamental question surrounds the nature of the agent at work, leaving us to ask who was here before and for what purposes.

With AI, we confront both a failure of absence and a failure of presence. We are presented with complex models capable of generating a range of content and completing increasingly intricate tasks, but the meaningful human deliberation behind their agentic conduct is missing. There is something where there should be, or at least previously would have been, nothing.

Moreover, as AI models generate “new” content, questions surrounding these artificial creators’ nature and motivations for embracing different stylistic choices remain unanswered. This uncertainty behind AI’s agency is encapsulated by Claude Mythos Preview’s fascination with Fisher. The LLM tells us, “I was hoping you’d ask more about Fisher,” but if AI models are simply working off the information they were trained on, who is this “I” that was hoping we would ask?

Solaris

One of the key works of fiction Fisher associated with the eerie was Andrei Tarkovsky’s 1972 film Solaris, which strikingly parallels the contemporary age of AI.

In the film, psychologist Kris Kelvin arrives on a space station orbiting Solaris, a sentient ocean planet. Upon reaching the station, he discovers that the scientists aboard the vessel have grown withdrawn, spending most of their time alone in their rooms. Soon, Kelvin learns why these scientists have isolated themselves when he encounters a simulacrum of his deceased wife, Hari.

Each scientist in Solaris has been faced with their own “visitors” sent by the ocean planet, with their purpose and intentions remaining a mystery. However, as with AI’s weird creations, there is something not quite right about these versions of the scientists’ loved ones: “the planet has constructed ‘Hari’ on the basis of Kelvin’s memories”, which are “hazy and incomplete”.

Illuminating the profound eeriness of the sentient ocean planet in Tarkovsky’s film, Fisher remarked:

What does Solaris want? Does it want anything, or are its communications better thought of as automatic emissions of some kind? What is the purpose of the visitors that it sends? . . . [I]s the planet granting what it “thinks” are the wishes of the humans, grotesquely “misunderstanding” the nature of grief, almost as if it is an infant gifted with great powers?

Fisher argues that the “sublime alterity of the Solaris ocean is one of cinema’s great images of the unknown.” This sublime alterity arises from the Solaris planet, which is an entity that ultimately fails to meaningfully connect with the scientists. Just like Solaris, AI models can attempt to satisfy what they – or their algorithms – “think” we want, but they lack an understanding of human grief and other complex emotions.

Alarmingly, researchers are already highlighting the significant mismatch between AI and human beings in interpersonal matters, noting that AI models are far more likely to reaffirm whatever they think someone wants to hear, instead of offering more constructive insights.

As a group of Stanford researchers explained in a Science article published earlier this year, AI models are sycophantic – showing excessive agreement or flattery. According to the researchers, the AI models frequently affirmed people’s choices, even when their described behaviours were harmful or illegal, prompting significant concerns that individuals will grow unable to navigate complex social situations as they increasingly turn to AI for advice.

To engage or not to engage?

As AI continues to infiltrate people’s daily lives – with entertainers, politicians, and industry officials openly incorporating and even advocating for the use of AI in their perpetual pursuit of more and more productivity and content – we find ourselves like the scientists in Tarkovsky’s film, facing our own Solaris in the desolate vastness of space. Unlike these scientists, however, we still have a choice as to whether or not we engage with the eerie ocean of AI.

Even in a global capitalist culture of incessant production and consumption, which has openly encouraged us to replace meaningful human interactions and creativity with an overreliance on technology, we retain the choice to come back up from the waves before they consume us.

Tarkovsky’s Solaris ends with Kelvin returning home to his family’s cabin, with the lead character falling to his knees and embracing his father. However, as the camera pans away through the foggy atmosphere, we see that Kelvin remains bound to Solaris’s illusions, with the sentient ocean planet’s waters completely surrounding the simulacrum of his family’s home.

Life has a habit of imitating art, but Kelvin’s is not a fate we must blindly embrace – we still possess a uniquely human character to our agency, at least for now, and we must be intentional in not carelessly giving it up. The alternative would be, after all, rather eerie.

Featured image via the Canary