Déjà Vu Makes a Cog in the Machine a Cognizant Machine
Déjà vu is a brief conflict: your brain's familiarity detector misfires and signals recognition for a novel moment, but your metacognitive supervisor catches the error. The sensation feels strange yet harmless because the arbitration system works. You feel the familiarity, but you know it's wrong.
That self-correction mechanism makes all the difference. Without it, false familiarity becomes pathological. Neurologists document cases of chronic déjà vécu ("already lived"), where patients insist novel experiences are genuine memories, unable to flag the conflict their monitoring system should catch. Current AI systems live in that failure mode permanently.
Adjust the parahippocampal familiarity signal and hippocampal recollection evidence to see where recognition is stable, where déjà vu emerges, and when pathological déjà vécu takes over.
Memory hierarchy mismatch
Neural evidence points to the medial temporal lobe. The parahippocampal cortex provides rapid familiarity signals; the hippocampus processes slower, context-rich episodic memories. fMRI studies show activation in regions associated with memory conflict during déjà vu reports, while hippocampal signals correctly flag novelty.
Temporal lobe epilepsy patients report déjà vu as seizure auras. Direct stimulation of the entorhinal cortex reproduces the sensation, demonstrating a cache-memory coherency failure in biological hardware: the fast familiarity pathway fires prematurely while the slow episodic system correctly registers novelty.
Step through intracranial stimulation sessions from Adachi et al. and Bancaud et al. to see electrode sites, currents, and patient reports that triggered déjà vu.
Metacognitive monitoring
Dual-process models frame déjà vu as successful error detection. The familiarity pathway misfires, but metacognition flags the conflict, preserving awareness that the experience is novel. When monitoring fails, as in chronic déjà vécu, patients insist entire episodes are genuine memories.
Computational triggers
Diary and clinical studies identify specific triggers. Novel environments with partial feature overlap activate familiarity without full context match: similar spatial layouts, familiar lighting patterns, or recurring sensory cues. Fatigue, stress, and hyperexcitability degrade temporal synchronization between pathways. The mechanism resembles a cache hit on stale data: the tag matches, but retrieved content doesn't align with current context in the slower, higher-capacity system.
AI memory systems lack hierarchy
LLM hallucinations differ fundamentally. Transformer attention is flat pattern matching; no distinct familiarity versus episodic pathways, no metacognitive supervisor saying "wait, what?" When KV cache contains stale context, models generate plausible nonsense with the confidence of a bone-digging dog. Déjà vécu without the self-awareness. Retrieval-augmented systems surgically add external memory but still lack the dual-process architecture that lets biological systems flag internal conflicts.
Switch between human cognition and LLM behaviour to compare déjà vu, déjà vécu, confabulation, and hallucination across monitoring, conviction, and self-correction dimensions.
Building conflict detection
Transformers lack the layered memory architecture that lets human brains catch their own errors. External validation can flag problems after generation, but it's fundamentally different from built-in conflict detection. Even alternative architectures like Mamba, which rethink how models process sequences, don't address this core gap. Whether any future design can integrate this kind of monitoring remains an open question.