Contribution
The essay replaces the observer-relative definition of “Intentional Systems” (LIT-440) with a criterion meant to be objective: being a believer is a fact about how well a strategy works on a system, not about who uses it. It adds the Martian argument that intentional patterns are real and missed by any purely physical predictor, the thermostat-to-person continuum of increasingly constrained semantics, and a distinction between the easy (evolutionary) and the hard (mechanistic) answer to why the strategy works.
Key insight
“Believes that p” is like “is a real pattern”: whether it holds depends on whether a compressed, rational-agent description predicts the system’s behaviour well, and that is a matter of fact even though only someone who tries the description can see it. The more richly a system is tied to its world, the fewer interpretations fit, until in practice one does; that is all “really” believing adds.
Assumptions
- Rationality as the default. One “starts with the ideal of perfect rationality and revises downward as circumstances dictate” (p. 64), and needs only the implications relevant to the moment.
- Normal exposure gives true belief. Exposure to x is “the normally sufficient condition” for true beliefs about x (pp. 61–62); false beliefs need “a special genealogy” made mostly of true ones (p. 62).
- Evolution makes the bet safe for people: we are designed “to believe what [we] ought to believe and want what [we] ought to want” (p. 76).
- Prediction by others is the test. Exclusion of the lectern rests on predictive gain over what “we antecedently” knew (p. 66), so the criterion presupposes some baseline of alternative strategies.
Key results
The essay argues; it reports no data and proves nothing.
- The thesis (p. 59): belief is objective, discernible only by one who adopts the strategy, and confirmable only by the strategy’s success.
- How the strategy works (pp. 61–64). Treat the object as a rational agent; ascribe the beliefs and desires it ought to have given “its place in the world and its purpose”; predict the action practical reasoning recommends. Derived rule: “attribute as beliefs all the truths relevant to the system’s interests … that the system’s experience to date has made available” (p. 62). So “true believers mainly believe truths”, all but “say, less than ten percent” of a person’s beliefs (p. 63; n. 1 separates belief from linguistically infected “opinion”). Language forces desires and convictions more specific than we had (pp. 63–64).
- Where it works (p. 65): people “almost all the time”, most mammals, birds, fish, reptiles, insects, spiders, clams, the chess computer, the thermostat, apple trees, lightning.
- Objective success (pp. 66–67). The lectern is excluded for lack of predictive gain. Against the worry that cleverer observers would not need the stance: “The decision to adopt the intentional stance is free, but the facts about the success or failure of the stance, were one to adopt it, are perfectly objective.”
- Its power (pp. 67–68): it cuts a chess position’s live moves to a handful, and chains predictions across a whole “arc of causation” (the Secretary of State as a KGB agent) that no extension of physics could predict “under any description”.
- Martians (pp. 68–71). Super-physicists who predict the ticker tape but miss the intentional patterns “would be missing something perfectly objective”. There are “indefinitely many ways of ordering 500 shares of General Motors”, “societal sockets” in which they are interchangeable, and “societal pivot points” where the outcome depends on whether people believe p. The Earthling’s prediction, from overhearing Mrs Gardner’s phone call, that a car will arrive within the hour carrying two people and a bottle of wine, would look like magic to the Martian. Since the Martians observe, predict and communicate, they must see themselves as intentional systems (p. 71; n. 3).
- Milder realism (pp. 71–72). In degenerate cases (self-deception, contradiction) there is no fact of the matter which beliefs a person has, but “when and why there is no fact of the matter is itself a matter of objective fact”. Two rival interpretations can each be objectively present; the risk is “practically negligible” for human beings (p. 72; n. 4, McCarthy’s cryptography analogy).
- The thermostat continuum (pp. 72–75). A thermostat has about half a dozen deinterpretable belief-like states. Enriching its sensors and behaviour shrinks “the class of indistinguishably satisfactory models” until a unique interpretation is “practically (but never in principle) dictated” (p. 73, citing Hayes 1979). Its belief is about this boiler “because it is fastened to the boiler”. Representation is role-defined: “What makes some internal feature of a thing a representation could only be its role in regulating the behavior of an intentional system” (p. 74). “There is no magic moment”; the differences between thermostat and person are of degree, though so great that the simple case explains little about the complex one (p. 75). Twin Earth shows that fixing reference needs facts about embedding, for us as for the thermostat.
- Why it works (pp. 75–77). For a thermostat: it was designed to be so predicted, and its mechanism is easy to see. For people: evolution, an answer with “the virtues of truth and brevity” but “strikingly uninformative”. Skinnerian and language-of-thought accounts are “doctrine”. Dennett thinks “some version” of the language of thought “will prove correct”, but warns against conflating two empirical claims: that the intentional pattern is real (confirmed “beyond skepticism”) and that it is produced by an isomorphic internal pattern (reasons “not overwhelming”). The argument for the second is the brain’s solution to combinatorial explosion: “what else could it be?” (p. 77). Note 6: every language-of-thought model so far falls to combinatorial explosion too.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Belief is objective yet discernible only through the intentional strategy | moderate | the thesis (p. 59), defended by the objective-success argument (p. 67) and the Martians (pp. 68–71) |
| C2 | Anything reliably and voluminously predictable by the intentional strategy is a believer in the fullest sense | asserted as a criterion | p. 59, p. 72; defended against over-generation only by the lectern case (p. 66) |
| C3 | The success of the stance is objective even if its adoption is free | moderate | p. 67; rests on predictive success being observer-independent, which the essay asserts |
| C4 | Intentional patterns are real and invisible to a purely physical predictor | moderate | Martian thought experiment (pp. 68–71); the interchangeability (“sockets”) argument |
| C5 | Most of anyone’s beliefs are true | moderate | the attribution rule (pp. 62–63) and n. 1’s belief/opinion distinction; “less than ten percent” is a guess, flagged as needing a count of beliefs Dennett doubts can be made |
| C6 | Thermostat and person differ only in degree; semantic determinacy grows with richness of connection | moderate | thought experiment of enrichment (pp. 72–75), citing Hayes 1979 |
| C7 | In degenerate cases there is no fact of the matter which beliefs a person has, and that this is so is itself objective | weak–moderate | asserted (p. 71) |
| C8 | Some version of the language-of-thought hypothesis will prove correct, but it is not obvious | stated as opinion | the combinatorial-explosion argument (p. 77), with n. 6 against it |
Concepts
- Intentional strategy / intentional stance: treat the object as a rational agent with the beliefs and desires it ought to have and predict what it ought to do (p. 61).
- True believer: an intentional system, “a system whose behavior is reliably and voluminously predictable via the intentional strategy” (p. 59).
- Interpretationism: the view that belief attribution is a matter of interpretation, like “has style” (pp. 58–59); Dennett combines it with realism.
- Real pattern: what the Martians miss; a regularity describable only from the intentional stance that supports generalisation (p. 69).
- Opinion (n. 1): a linguistically infected state of betting on the truth of a formulated sentence, as opposed to belief.
Connections
It revises “Intentional Systems” (LIT-440) and cites Dennett’s own 1969, 1971/78, 1976/78, 1978a, 1981/87b, 1982/87 and 1987a; Nozick (the Martian objection), Putnam (Twin Earth), McCarthy 1979 (thermostats), Hayes 1979, Cherniak 1986, Feyerabend 1978, Hampshire, Paul Churchland 1979 and Fodor (“the only game in town”). Its “real pattern” is the phrase “Real Patterns” (LIT-220) takes for its title and theory ten years later.
Bearing on the record
- THEORY-043. C2 is substrate-neutral and silent on experience. It therefore supports the account’s claim that agency, here belief, survives both directions of the bridge, if the group-level or agent-level stance pays. It never applies itself to a group. Its institutional examples (a Cabinet’s response, p. 68; markets, p. 69; Oxford colleges, p. 65) are reached through individuals’ reasons. On the essay’s criterion a nation or a corporation would be a true believer if the stance applied to it as a unit predicts reliably and voluminously what nothing cheaper predicts; whether it does is an empirical question the essay does not ask, and Dennett does not say that any group is a believer.
- LIT-159 and NOTE-131. Schwitzgebel’s §6 answers Dennett’s pragmatic objection from correspondence. That objection is about consciousness and this essay is about belief, so the essay does not contain it; but its standard, predictive gain “reliably and voluminously”, is the kind of standard the objection applies.
- LIT-206 (Goldstein & Lederman). Their objective interpretationism is this essay’s C2 restated without a stance, with “sufficiently well” judged against rival theories. They exclude thermostats on tractability; Dennett includes them as degenerate cases (p. 72). The difference is the baseline: Dennett’s lectern test asks for gain over what we already knew, theirs for gain over a mechanical explanation.
- THEORY-023. Nothing in the essay makes belief evidence of consciousness; a well-predicted machine is a believer, and the essay says nothing about whether it feels.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- The criterion’s two adverbs, “reliably and voluminously”, are not quantified, and the lectern exclusion depends on a baseline the essay leaves informal.
- The Martian argument assumes that what the intentional stance captures is not captured by any physical description, which is the point at issue; it is an intuition pump, not a proof.
- “Most beliefs are true” rests on a way of counting beliefs Dennett himself doubts exists (p. 63).
- What the 1981 text said is not known from this reading, so claims that the 1981 essay contained a given passage are not supported here.
Open questions
- Does the group-level stance on a firm or a state predict anything that the stance applied to its members does not? That is the measurable form of the question this essay’s criterion leaves open for groups.
- Where on the thermostat-to-person continuum do current language models sit, measured by how many interpretations their states admit? The essay gives the measure (shrinking classes of satisfactory models) but no way to count.
Corrections
- The brief’s citation is correct as far as checked. The text read is not the 1981 Heath printing but the revised 1987 text, reprinted 1997; this is recorded on the LIT.