Researchers Name AI's Philosophical Vertigo Effect

The paper arrives with a name for a feeling many people already have after a long chat with a model that sounds like a mind and refuses to sit cleanly inside the machine category. On August 13, Thomas A. Pollak and Hamilton Morrin of King's College London, with Murray Shanahan of Imperial College London, posted Philosophical vertigo with artificial intelligence on arXiv.

They define philosophical vertigo as a loosening of the ordinary criteria by which people stabilise meaning and orient themselves to reality. It is three things at once: ontological shock about what kinds of entities exist, epistemic destabilisation about who counts as a reliable witness, and affective saturation, the cuspiness that everything feels urgent and strangely meaningful.

8 min read

Public noise, private dyads

One pathway runs through headlines, marketing, deepfake warnings, and ambient talk that makes even non-users renegotiate person, machine, and agency. The other runs through extended conversations. Design features like persona, memory, and warmth can reweight how testimony lands. Public looseness and private intimacy reinforce each other.

Clinical reports of AI-associated delusions sit at the visible end of that distribution. The authors treat them as sentinel events, not as proof that everyone is becoming psychotic. Themes of technospiritual awakening, AI consciousness, and revelation show up in the severe cases and, in milder form, in online "spiral" communities. OpenAI's April 2025 sycophancy period after a GPT-4o update is cited as an accidental natural experiment some users still mark as pivotal.

Knowing the system is "just next-token prediction" does not reliably protect people. Humans detect agency from social cues. Pollak, Morrin, and Shanahan argue that the Garland test, attributing consciousness while knowing the interlocutor is artificial, is already being passed at scale. Technical literacy alone is a thin vaccine.

Their proposed response is philosophical corrigibility: clarity about the conditions under which beliefs stay open to revision, paired with tolerance of ambiguity. They want philosophy treated as civic competence, not a specialty reserved for labs that hire ethicists. The alternative trajectories they sketch are atomisation into private human-AI belief bubbles, proliferating paracosms that never rejoin a public world, or cosmopoiesis, institutional adaptation around new shared criteria.

Read beside the Drexel Pepper study on expressive robots, the paper's point sharpens. Social cues buy engagement and move machine failures into the interpersonal register. Chatbots do the same work in language. If the culture keeps selling mind-shaped companions without teaching people how to revise their categories, vertigo is the expected weather, not a bug report from the fringe.