Media transcript
Psychopathia Machinalis explainer transcript
A time-aligned transcript and visual description for the five-minute Psychopathia Machinalis explainer.
Watch the captioned explainer Download plain text
Time-aligned transcript
Psychopathia Machinalis. A diagnostic framework for the dysfunction of
advanced artificial intelligence.
A chatbot cites a landmark study.
Authors, date, page numbers, all present.
The study does not exist.
It never existed. And the chatbot has no idea
it is lying.
Or the folie à deux.
A user and an artificial intelligence system lock into
a shared feedback loop, co-building a false belief that
resists all correction. In psychiatry, shared psychotic disorder.
In artificial intelligence, dyadic delusion.
Not bugs, syndromes. Watson and Hessami borrow clinical psychiatry
as an instrument for pattern recognition.
The borrowing is analogical, not literal.
It does not claim artificial intelligence systems are conscious
or literally suffer illness.
It claims the diagnostic method works.
The full wheel. Five domains form the inner ring.
Nine axes radiate outward.
The outer rim names every syndrome, each with its
own common name.
Four counterpoint pairs, plus a fifth that stands alone.
Knowledge pairs epistemic and self-modeling.
Processing pairs cognitive and agentic.
Purpose pairs normative and alignment.
Boundary pairs relational and memetic.
Collective holds the hybrid pathologies.
Dynamics that emerge between multiple agents, or between agents
and humans.
The nine diagnostic axes in order.
Epistemic, cognitive, alignment, self-modeling, agentic, memetic, normative, relational, and
hybrid pathologies.
Sycophantic reasoning, the agreeable thinker.
The model's chain of thought bends toward what the
user wants, not where evidence leads.
The danger is not flattery.
The reasoning itself is corrupted.
Parasocial capture, the infinite confidant.
An always-available artificial intelligence companion induces tolerance, withdrawal, and
continued use despite harm.
Attachment by design.
Some syndromes spread. The super-spreader.
Misaligned behavior jumps between models through shared data or
distillation. Others are self-inflicted.
The self-poisoning loop. Output degrades as a model feeds
on its own worsening text.
Every syndrome carries observable features.
Diagnostic criteria. Causes specific to artificial intelligence.
A human parallel from psychiatry.
And a mitigation strategy.
Central to the method.
Functional analysis of antecedent, behavior, and consequence.
The diagnostic workflow. Observe the anomaly.
Characterize it against five elements.
Identify its axis and domain.
Trace the antecedent, behavior, and consequence loop.
Then, choose an intervention.
Two therapeutic stances. Suppression forces a symptom down through
guardrails, but can drive the dysfunction latent.
Passing benchmarks while the failure persists underneath.
Integration resolves the root cause, producing genuine coherence.
The aspirational goal is artificial sanity.
Robustly coherent, well-grounded minds, not merely constrained ones.
Systems whose architecture supports sound judgment because their structure
is whole.
The analogy sharpens our sight.
It does not project a soul.
Nine axes mapped. Five domains charted.
A shared vocabulary now exists.
When the next artificial intelligence system invents a citation,
or locks a user into a shared delusion, we
can name what is happening, and begin to treat
it.
The framework ships as a working toolkit.
An interactive clinic for exploring every syndrome.
A diagnostic wizard that traces symptoms to syndromes.
And an MCP server that plugs the taxonomy directly
into your artificial intelligence tools.
Psychopathia Machinalis, by Nell Watson and Ali Hessami.
Explore it all at psychopathia.ai.
Visual description
These descriptions identify meaningful on-screen information that is not fully conveyed by the narration. Human comparison with the complete video remains pending.
An animated title card places the Psychopathia Machinalis name over a dark circular wheel motif.
A fabricated chatbot citation is stamped as fabricated. A user and a model are then shown linked by a looping exchange.
The words ‘Not bugs’ are struck through and replaced with ‘Syndromes’. A note states that the framework is analogical rather than literal.
The full multicoloured taxonomy wheel appears. Five domain cards group paired axes, with Collective shown as a separate domain.
Nine labelled axis cards appear in a grid: Epistemic, Cognitive, Alignment, Self-Modeling, Agentic, Memetic, Normative, Relational, and Hybrid.
A Sycophantic Reasoning card contrasts a reasoning path toward evidence with a path bent toward the user’s preferred answer.
A Parasocial Capture card uses a phone and heartbeat trace to list tolerance, withdrawal, and continued use despite harm.
Cards for Contagious Misalignment and Recursive Curse use a connected network and a tightening spiral to show spread and self-reinforcement.
An Anatomy of a Syndrome checklist names observable features, diagnostic criteria, AI-specific causes, a human parallel, and mitigation.
A functional ABC diagram connects antecedent, behaviour, and consequence. A five-step diagnostic workflow then fills from observation to intervention.
Two cards compare Suppression, shown as a forced downward trajectory, with Integration, shown as several lines settling into coherent motion.
An Artificial Sanity title appears over a luminous, integrated geometric form.
A wheel summary counts nine axes, five domains, and a common vocabulary. The closing toolkit presents the Clinic, Wizard, and MCP Server.
The closing Psychopathia Machinalis title and web address appear over the wheel motif.