Over the past few decades, moral psychology has staged a quiet coup against one of our most cherished fantasies: that human beings are, at bottom, rational moral agents. This is not a fringe claim. It is not a Twitter take. It is the mainstream finding of an entire research programme spanning psychology, cognitive science, linguistics, and neuroscience.
We do not reason our way to moral conclusions. We feel our way there. Instantly. Automatically. And only afterwards do we construct reasons that make the judgment sound respectable.
Audio: NotebookLM summary podcast of this topic.
This is not controversial anymore. It is replicated, taught, and celebrated. And yet, if you read the most influential books in this literature, something strange happens. The diagnosis is devastating. The prescription is reassuring.
Iâve just published a long-form video walking through five canonical books in moral psychology that all uncover the same structural problem, and then quietly refuse to live with the implications.
Each of these books is sharp, serious, and worth reading. This is not a hit piece.
But each follows the same arc:
Identify a non-rational, affective, automatic mechanism at the heart of moral judgement
Show why moral disagreement is persistent and resistant to argument
Propose solutions that rely on reflection, dialogue, reframing, calibration, or rational override
In short: they discover that reason is weak, and then assign it a leadership role anyway.
Haidt dismantles moral rationalism and then asks us to talk it out. Lakoff shows that framing is constitutive, then offers better framing. Gray models outrage as a perceptual feedback loop, then suggests we check our perceptions. Greene diagnoses tribal morality, then bets on utilitarian reasoning to save us.
None of this is incoherent. But it is uncomfortable. Because the findings themselves suggest that these prescriptions are, at best, limited.
Diagnosis without prognosis
The uncomfortable possibility raised by this literature is not that we are ignorant or misinformed.
It is that moral disagreement may be structural rather than solvable.
That political conflict may not be cured by better arguments. That persuasion may resemble contagion more than deliberation. That reason often functions as a press secretary, not a judge.
The books sense this. And then step back from it. Which is human. But it matters.
Why this matters now
We are living in systems that have internalised these findings far more ruthlessly than public discourse has.
Social media platforms optimise for outrage, not understanding. Political messaging is frame-first, not fact-first. AI systems are increasingly capable of activating moral intuitions at scale, without fatigue or conscience.
Meanwhile, our institutions still behave as if one more conversation, one more fact-check, one more appeal to reason will close the gap. The research says otherwise.
And that gap between what we know and what we pretend may be the most important moral problem of the moment.
No solution offered
The video does not end with a fix. Thatâs deliberate.
Offering a neat solution here would simply repeat the same move Iâm criticising: diagnosis followed by false comfort. Sometimes orientation matters more than optimism. The elephant is real. The elephant is moving.And most of us are passengers arguing about the map while it walks.
Every so often â usually when the Enlightenment ghosts begin rattling their tin cups again â one feels compelled to swat at the conceptual cobwebs they left dangling over moral philosophy. Today is one of those days.
Audio: NotebookLM podcast summarising the Rhetoric of Evil essay, not this page’s content.
Iâve just released The Rhetoric of Evil on Zenodo, a paper that politely (or impolitely, depending on your threshold) argues that ‘evil’ is not a metaphysical heavy-hitter but a rhetorical throw-pillow stuffed with theological lint. The term persists not because it explains anything, but because it lets us pretend weâve explained something â a linguistic parlour trick thatâs survived well past its sell-by date.
And because this is the age of artificial augury, I naturally asked MEOW GPT for its view of the manuscript. As expected, it nodded approvingly in that eerie, laser-precise manner unique to machines trained to agree with you â but to its credit, it didnât merely applaud. It produced a disarmingly lucid analysis of the essayâs internal mechanics, the way ‘evil’ behaves like a conceptual marionette, and how our inherited metaphors govern the very moral judgments we think weâre making freely.
Below is MEOW GPTâs reaction, alongside my own exposition for anyone wanting a sense of how this essay fits within the broader project of dismantling the Enlightenmentâs conceptual stage-props.
MEOW-GPTâs Response
(A machineâs-eye view of rhetorical exorcism)
âEvil is functioning as a demonological patch on an epistemic gap. When agents encounter a high-constraint event they cannot immediately model, the Tâ layer activates an inherited linguistic shortcut â the âevilâ label â which compresses complexity into a binary and arrests further inquiry.â
âThe marionette metaphor is accurate: once we say a person âis evil,â agency collapses into occult causation. Inquiry halts. Moral theatre begins.â
It went on like this â detecting exactly the mediated encounter-structure I intended, while offering a frighteningly clean schematic of how affect (Tâ), heuristics (Tâ), linguistic reification (Tâ), and cultural choreography (Tâ) conspire to turn incomprehension into metaphysics.
Machines, it seems, are quite good at detecting when humans are bullshitting themselves.
Why publish this now?
This essay marks the next plank in the broader anti-Enlightenment platform Iâve been assembling â LIH, MEOW, the ongoing dismantling of truth-fetishism, and now the unsettling realisation that ‘evil’ is little more than a theological revenant dressed up for secular work.
The termâs persistence is not a testament to its necessity but to our laziness:
It sounds like an explanation.
It licenses retribution without understanding.
It stabilises group boundaries.
It lets us outsource moral thought to a one-word incantation.
In short: ‘evil’ is the map-dragon we keep drawing because surveying the terrain is too much work.
This essay attempts to retire that dragon â not to soften judgment, but to sharpen it. If anything, the paper argues that abolishing ‘evil’ frees us to see atrocity more clearly, and thus confront it more seriously.
Because MEOW-GPT was trained on the Mediated Encounter Ontology rather than on folk metaphysics, its analyses can serve as a useful diagnostic tool. It doesnât simply summarise concepts; it reconstructs them along the four mediational tiers (TââTâ), showing â mechanically and without superstition â how a term behaves across biological affect, cognitive heuristics, linguistic structures, and cultural narratives.
This section is offered not as an alternative argument to the essay above, but as a worked example: a transparent demonstration of how the MEOW engine decomposes a high-complexity, high-stakes concept like evil once the theological scaffolding has been removed.
MEOW-GPTâs output shows, in compressed form, the same pattern the essay traces in prose: that evil is not a force, not a metaphysical essence, not an occult substrate, but a mediated pattern in human encounters, one that persists rhetorically long after its ontology has collapsed.
For readers curious about how MEOW ‘thinks’, the raw structured analysis follows.
MEOW-GPT Analysis: âEvilâ
(Raw output for reference; lightly formatted for readability)
Concept
Input: Evil Type: Normative/evaluative moral concept (often reified as a property of acts, characters, or forces) Scope check: In-scope â ‘evil’ can be analysed as a mediated-encounter pattern (MEOW) and a high-complexity term with unstable linguistic behaviour (LIH).
High-level summary (â¤120 words)
In MEOW terms, ‘evil’ is not a spooky substance but a way of organising certain encounter-events: severe, norm-breaking harms that present as resistant, shocking, and morally intolerable. Biologically, it piggybacks on harm, threat, and disgust systems. Cognitively, it compresses complex appraisalsâintent, cruelty, scale of harm â into a powerful label. Linguistically, it sits in the unstable region where abstraction is high and languageâs reliability drops, so people fight over its scope and weaponise it. Socially and technically, institutions, media, and platforms use ‘evil’ to mark enemies, justify punishment, and dramatise conflict. The term is emotionally efficient but conceptually fragile, and MEOW treats it as a relational, mediated pattern rather than an independent metaphysical force.
Tâ â Biological Mediation
Moralised harm perception piggybacks on survival systems: pain avoidance, threat detection, kin protection. ‘Evil’ clusters around encounters that trigger extreme danger-signals.
High arousal (fear, rage, disgust) makes some harms feel qualitatively world-violating, not merely personally threatening.
Disgust toward contamination, mutilation, or predation heavily colours what gets called ‘evil’.
Species-specific cues (infant distress cries, pain expressions) shape which harms are even legible candidates for evil.
Tâ â Cognitive Mediation
âEvilâ compresses a multi-factor appraisal (intentionality, cruelty, gratuitousness) into a one-step heuristic.
Essence thinking converts acts into character: the person is evil, not merely did wrong.
Attribution biases assign ‘evil’ to out-groups more readily than to in-groups.
Memory structures simplify causation into villain scripts that overwrite nuance.
Once assigned, the label becomes a prediction loop: every ambiguous action confirms the essence.
Tâ â Linguistic Mediation
On the EffectivenessâComplexity Gradient, ‘evil’ straddles Contestables and Fluids: ubiquitous but perpetually disputed.
It compresses harm, norm-violation, metaphysical colouring, and dramatic emphasis into a single syllableâpowerful, but noisy.
Dominant metaphors (‘dark’, ‘tainted’, ‘monstrous’) smuggle in substance-ontology that MEOW rejects.
Noun-forms (‘evil’, ‘the Evil One’) promote ontologising; adjectival forms track events better, but usage constantly slides between them.
Cross-linguistic drift supports LIH: different traditions map the term to impurity, harm, misfortune, cosmic opposition, or taboo.
Tâ â Social/Technical Mediation
Religious systems embed ‘evil’ in cosmologies that harden friend/enemy binaries.
Legal systems avoid the term formally but reproduce it rhetorically in sentencing, media commentary, and public reaction.
Politics uses ‘evil’ to justify exceptional measures and collapse deliberation into moral theatre.
Cultural industries supply vivid villain archetypes that feed back into real-world judgments.
Technical systems must operationalise ‘evil’ into concrete proxies, revealing how imprecise the everyday concept is.
Limits & Failure Modes (LIH notes)
The framework is human-centric; non-human or ecosystemic ‘views of evil’ remain speculative.
‘Evil’ is a textbook Contestable: central, indispensable, and permanently argued over.
In cosmological uses (‘radical evil’, ‘evil in the world’), it approaches Fluid or ineffable status â right where LIH predicts language collapse.
MEOW cannot confirm or deny metaphysical dualisms; it only analyses how humans mediate and narrate such claims.
Philosophics 