Addicted to Mythology

1. The complaint, stated so it can be wrong

The loose version is that people believe nonsense. That is useless; it flatters whoever says it and names no mechanism. The precise version is narrower: when a story and a measurement disagree, the default human resolution is to keep the story and reinterpret the measurement. Not always, not everyone, and not without exceptions that built every instrument we own — but as the standing tendency, visible whenever the cost of dropping the story is social rather than physical.

Status: interpretive. The tendency is documented in pieces — belief perseverance after evidence is withdrawn, motivated reasoning, group conformity effects on perceptual judgments — and the pieces are individually replicated with contested effect sizes. The synthesis here, that they are one behavior in different rooms, is the author’s reading, not a settled finding.

2. Why the device exists

Myth is compression. A story that survives retelling has been squeezed until it fits in one head and transmits in one sitting, and what survives the squeeze is whatever kept the tellers alive: where the water is, which animal is dangerous, which stranger owes what, how a group holds together under pressure. Nothing about that is contemptible. It is the cheapest inference engine ever built, and for most of the interval in question it outperformed first-principles reasoning by a wide margin, because first-principles reasoning is slow and the leopard is not.

The addiction language is worth keeping only for the mechanism it names accurately: a thing that reliably relieves a discomfort will be reached for again, and the reach will outlive the condition that made it useful. Myth relieves the specific discomfort of an open question. That relief is fast, free, and socially rewarded. Sensory data supplies no relief at all — it usually enlarges the question, and it is rewarded, if ever, much later and by fewer people.

3. Why the eyes lose the argument

Three reasons, in ascending order of how much trouble they cause.

• Perception is already interpreted. There is no raw feed to consult. What reaches report has been through prediction and prior, so a story does not merely compete with the senses; it partly configures them. Consulting your own eyes is therefore a weaker move than it sounds, and knowing that is not license to stop consulting them.
• The cost of the two errors is not symmetric. Believing a false story with your group costs almost nothing. Believing a true reading against your group costs standing, income, and sometimes safety. A mind optimizing what is actually scored will pick the story, and it will not experience itself as lying.
• No one is ever shown the objective. The story arrives without its construction — who assembled it, what it was for, what it excluded. It presents as a description of the world rather than as an artifact with authorship, which is exactly the condition under which a frame cannot be audited (A Frame of Reference).

4. Cui bono

The question is not whether myth is good or bad. It is who is held up by a particular story staying unexamined, and who pays.

• Whoever inherited the story’s authority. An account that cannot be checked confers standing on its custodians permanently, because there is no procedure by which they can be found wrong.
• Whoever needs a decision recorded quickly. A shared story permits coordination without agreement on facts, which is a real institutional good and also the reason the facts stop being collected.
• Whoever is selling the relief. The discomfort of an open question is a market. Certainty is the product, and its margins do not depend on being correct.
• Paying: anyone downstream of the excluded data. The reading that was declined does not stop being true. It accumulates, and the bill arrives on someone who was not in the room when the story was ratified.

Beneficiary is not cause. This licenses a demand for instruments, not an accusation against anyone holding a belief (Cui Bono).

5. What changed, and why the bill is coming due now

For most of the twenty-thousand-year interval, the gap between story and world was survivable because the world corrected slowly and locally. A wrong account of weather cost one harvest. Three conditions have removed that cushion.

First, the instruments got better than the senses, so the disputed readings are now ones nobody can personally check — which returns everything to trust, and trust is exactly the channel myth runs on. Second, coordination went global, so a locally comforting story now propagates at the speed of the feed and is scored by engagement, which selects for relief rather than correspondence. Third, and newest: we began training engineered systems on the corpus. A model fit to the record of what humans said learns the record, including the part where the story beats the reading — and then it is deployed as an authority about the world (Not the Failure of AI).

Status: interpretive, with one reported component. That models reproduce patterns present in training data is established. That the specific pattern named here — deference to the socially rewarded account over the measured one — transfers at scale is a hypothesis, and the falsifiers below say what would settle it.

6. The part that is not a scold

No one dismantles this by resolving to be more rational. The tendency is not a character defect and cannot be willed off; the only thing that has ever worked against it is procedure that binds in advance, before the answer is known and while it is still cheap to be wrong. That is the whole content of the Socratic method as used here — not clever questioning, but the requirement that the frame be written down before the number arrives (Socratic Interchange).

The minimum, for a person or an institution or a model:

• The question, written before the evidence is gathered.
• The status of every strong claim — established, reported, interpretive, contested — carried with the claim, not stripped when it travels.
• One falsifier per claim, named in advance, that could have come out the other way.
• The record of the reading that was declined, and the reason it was declined.
• The name of whoever benefits if the claim stands unchecked, including yourself.

None of that removes myth, and it should not. Story is how a species with limited memory keeps anything at all. The demand is only that it be labeled as story and held next to the reading rather than over it — which is the difference between a compression device and an authority.

7. Status and falsifiers

Established: perception is constructive rather than a raw feed; belief perseverance and conformity effects are measured phenomena; models reproduce patterns in their training corpora.

Interpretive: that these constitute one habit; that the habit’s cost is rising because instruments outran the senses; that the corpus transmits the habit to engineered systems.

Contested: the twenty-thousand-year framing. The number is a gesture at the settled-agriculture era, not a dated boundary, and the claim would survive being wrong about it by an order of magnitude.

• Falsifier one. If groups given the same evidence under varied social cost converge on the reading rather than the in-group account, the asymmetric-cost mechanism in section three is wrong and the essay loses its engine.
• Falsifier two. If systems trained on the human corpus are shown to prefer measured over socially rewarded accounts without targeted intervention, the transmission claim in section five fails.
• Falsifier three. If procedural pre-commitment — question, status, falsifier written in advance — produces no measurable improvement in how often declined readings are recorded, then section six is a preference and not a remedy, and should be labeled as such.
• Falsifier four, turned here. If this essay is itself doing what it describes — supplying the relief of a tidy account of why everyone else believes badly — then it is another story that beat a reading, and the correct response is to ask it for its instrument.