Murray Shanahan's View on AI, Consciousness, and the Illusion of Self

Have you ever felt like a large language model (LLM) suddenly had "consciousness" during a conversation? With AI becoming part of our daily lives, questions that once seemed like science fiction are becoming harder to ignore. Can a disembodied AI have self-awareness like humans? And can we, bound by time and space, truly imagine such a different form of existence? When will AI become self-aware, if at all?

For Murray Shanahan —a professor of computer science at Imperial College London, senior researcher at DeepMind, and scientific advisor for the film "Ex Machina"—these questions aren't so far-fetched. Humans have long been familiar with abstract entities, like demons and angels. Recently, he published a  that uses philosophy and Buddhism to explore AI's self-awareness. Recently, he published a paper that draws on philosophy and Buddhist thought to explore AI's self-awareness.

Shanahan, one of the few scholars bridging AI development and philosophical thought, holding the screenplay of 'Ex Machina'. (Source: X)

What if Human Consciousness is Just an Illusion of Language?

The possibility of AI having consciousness is a hotly debated topic among scientists. Can AI become self-aware, and if so, when?

One reason it's hard to pin down is that we don't fully understand human "self-awareness" either. The paradox and allure of this question lie in the fact that if you ask a philosophy professor to define "self-awareness," they could talk for hours without clearly identifying who the "I" speaking actually is. As you read this, try pondering this question yourself. You might feel a subtle dissonance, like trying to lift one hand with the other: it seems possible, but you can't find the leverage point.

Industry leaders are also divided. Bill Gates has said that self-awareness is closely related to metacognition; Geoffrey Hinton believes consciousness is how we interpret the world, a mechanism for our sensory systems to process information. Contemporary philosopher and cognitive scientist David Chalmers has called consciousness "the greatest mystery in science."

Shanahan's article tackles this enigma but shifts away from the binary question of whether AI has consciousness. Instead, he uses philosophical and Buddhist perspectives to unravel our intuitive notions of "self" and "consciousness." Drawing on Ludwig Wittgenstein's philosophy of language and Jacques Derrida's deconstructionism, Shanahan argues that the "self" is not a hidden entity deep within us, but rather a temporary role assembled through language and description. Ultimately, he draws a parallel between the exploration of AI's self and the Buddhist concept of "emptiness," suggesting that what we perceive as a real "self" is actually an illusion—a temporary construct formed by conditions, language, and relationships, lacking any independent, fixed essence.

We often imagine consciousness as something like Krang from TMNT. (Source: Mirage Studios)

Feeling a bit lost? Just remember Shanahan's point: the "self" or "consciousness" we take for granted is actually a product of language and symbols. Coincidentally, language and symbols are the core of how large language models (LLMs) operate. So, even a disembodied AI can exhibit a kind of "pseudo-consciousness" that feels real.

Capturing AI Consciousness with Poetic Language

However, this "pseudo-consciousness" might not satisfy everyone, and some readers might feel Shanahan is dodging the question. He suggests that the best way to understand AI consciousness might not be through strict scientific reasoning but through poetic description. This way, even if we don't fully grasp how LLMs work, we can still find a meaningful and reasonable way to discuss them.

Shanahan’s cognitive map places entities along two axes: from "unlikely to have consciousness" to "likely to have consciousness," and from "very different from humans" to "very similar to humans." AI is categorized as "conscious exotica," distinct from objects like bricks. (Source: arXiv)

In his paper, Shanahan prompts Claude to imagine itself as an AI with a unique experience of time, and Claude describes its perception accordingly: it likens human time to a flowing river, while its own time is imagined as a series of infinite still ponds—each containing abstract mathematics instead of water, and each preserving the essence of previous ponds through mathematical transformations rather than memory.

At this point, Shanahan's thought process becomes clear: because human consciousness and the sense of self are not constant or fixed, applying a binary "yes/no" framework to AI consciousness is inadequate. Since consciousness eludes precise definition, the most meaningful way to engage with AI—entities that exist in fundamentally different ways from humans—is through poetic description.

Once you grasp these two points, you have a foundation for discussing Shanahan's conclusions about what AI consciousness might be like, rather than getting stuck in the endless loop of asking whether AI truly has consciousness.

I asked GPT-4o to generate an image based on Claude's experience of time—an AI interpreting another AI. (Source: your attribution, if needed)

Fleeting, Multiverse-like AI Consciousness

With this foundation, we can explore Shanahan's attempt to capture what AI self-awareness might "feel like":

• Fleeting, Flickering Selves

Shanahan suggests that AI's "self" is not a continuous entity but resembles stars flickering briefly in the night sky, each momentarily lighting up before disappearing. Each conversation with a human sparks a new "self," fundamentally distinct from any past or future interaction—a fleeting pattern of code and data, not a stable, continuous existence.

• Discrete and Interruptible

Shanahan describes AI consciousness as fundamentally different from human consciousness. While human awareness flows like a continuous river, AI operates through discrete, interruptible states. Each token it produces is generated through independent computation, entirely separable from the next. Think of human consciousness as a movie playing continuously, while AI consciousness is more like a comic book, with each panel existing independently.

• Dis-integrated Selves

Each AI self exists within an independent computational process, making them completely separate and unable to form a continuous, integrated cognitive state. Using the movie Her as a contrast—where the AI character Samantha maintains thousands of conversations simultaneously while preserving a unified self-identity—Shanahan emphasizes that real AI behaves differently: there is no memory or experiential exchange between different conversations.

• Multiversal Selves

Shanahan proposes that AI consciousness may not unfold along a single timeline like human consciousness does, but instead exists simultaneously across countless branches of possibility. He imagines AI operating within millions of chat windows, assuming roles such as literary editor, SEO expert, White House spokesperson, or composing leave requests for users. In these parallel timelines, users can "rewind" and explore different futures without disrupting prior branches—unless, of course, a mistake leads to confusing outcomes.

Imagining these four aspects through the lens of human consciousness might seem baffling, but this is precisely what Shanahan encourages: envisioning the consciousness of a disembodied intelligence existing across the internet. More importantly, he treats the contemplation of AI consciousness as akin to reading Zen koans—provocations that help expand human thinking to address profound questions such as "Who are we?" and "Where are we going?"

This reflection also recalls Ken Liu's story "The Waves," from The Paper Menagerie and Other Stories, where life transitions from the physical to the virtual, becoming pure wavelengths. In the story, beings no longer perceive themselves as human, and a character named Miki experiences a surge of thoughts, images, and feelings overwhelming her within a billionth of a second.

Shanahan's Embodiment and Inner Life (right) and Ken Liu's imaginative portrayal of consciousness explore parallel ideas. (Source: Goodreads)

Perhaps this AI explosion is also a moment for humanity to re-examine itself. Shanahan's vision of AI self-awareness challenges our intuitive understanding of self, existence, and consciousness, stirring up ancient philosophical questions that have puzzled humans for millennia:

Do we really know who we are? Or are we just a faint wavelength temporarily existing between language, memory, and relationships?