1. What the thought experiment is for

The argument from illusion is the engine of the entire inner-theater tradition: the single inference that, run once and generalized, delivers sense-data, indirect realism, and the sealed Cartesian room the whole book exists to take apart. Chapter 3 names it “the engine” and runs it at full strength on purpose, naming the escape routes (Austin, disjunctivism) without walking any of them to victory. The reader is left productively stuck, because the decisive dismantling is deferred forward to Chapter 5, where the vehicle/object distinction lands as the designed figure/ground flip. That deferral is deliberate. This entry carries the apparatus the chapter deliberately withholds: the argument’s exact logical skeleton, the premise that actually does the damage, and the reply with its open dependence on a positive theory the chapter has not yet earned.

The entry’s contribution. Beyond reconstructing the argument, the entry advances one claim of its own: the rejection of the Phenomenal Principle is now overdetermined, and the argument from illusion is therefore better treated not as a puzzle to be defused but as a selection argument among the views that defuse it. Three independent routes converge on refusing the Principle — Austin’s ordinary-language demolition, the transparency datum, and the vehicle/content account of misrepresentation. But refusing the Principle is now common ground that the book shares even with its rivals: the contemporary naive realist refuses it too. The live question is no longer whether the sense-datum inference fails but which of the views that reject it pays the lowest price elsewhere — and that question, this entry argues, is adjudicated at the hallucination case — on comparative grounds, and in the book’s favor.

2. The canonical scenario

A straight stick, half-submerged in water, looks bent.¹ Now press the awkward question: when the stick looks bent, what are you directly aware of? Not a bent stick in the world (the stick is straight, and you can lift it out and confirm as much). Yet something bent figures in your awareness. Call it what the tradition has called it — a sense-datum, an appearance, an idea, an inner image. It is bent, you are aware of it, and it is not the physical stick. So the immediate object of awareness, in this case, is something inner.²

That is the local conclusion. The argument’s reach comes from a second move, the generalizing step. There is no felt discontinuity between the bent-looking stick and a normally seen one; lift the stick slowly and the bend smoothly vanishes, with no point at which awareness switches from an inner object to an outer one. So if the immediate object is a sense-datum in the illusory case, parsimony and continuity demand it be a sense-datum in the veridical case too. We never directly perceive physical objects; we perceive inner items, and infer the world behind them.

3. The formal structure

1. In illusion, something looks F (bent) although the perceived object is not F. [datum]
2. The Phenomenal Principle: if something looks F to a subject, then there is something of which the subject is aware that is F.³ [the premise that does the damage]
3. From (1) and (2): there is something bent the subject is aware of, and, by the indiscernibility of identicals (since the stick is straight), it is not the physical stick. That establishes nonidentity, not mentality.
4. The location premise: the immediate bent bearer is neither the straight stick nor another external object occupying the perceived place; it is a mind-dependent item available only in the subject’s awareness — a sense-datum.
5. The generalizing step: illusory and veridical perception differ in no qualitative kind and form a continuum, so whatever the immediate object is in the one case it is in the other.
6. Therefore in all perception the immediate object of awareness is an inner sense-datum; physical objects are perceived only mediately, by inference. [indirect realism]

Three premises carry the weight: the Phenomenal Principle (2), the location premise (4), and the generalizing step (5). The location premise makes explicit what the traditional argument often leaves tacit: nonidentity with the stick does not yet place the bent bearer inside the mind. The generalizing step draws the more frequent attack; the Phenomenal Principle does the deeper damage, and the book’s reply targets it.

4. The pressure point

The Phenomenal Principle is where an innocent-sounding sentence smuggles in the whole inner theater. “There is something bent of which you are aware” reads as a near-tautology — surely if it looks bent, something bent is going on. But the Principle does more than register the look: it converts an adverbial or relational fact (“the stick looks bent to you,” “you are visually presented with the stick as bent”) into the existence of an object that bears the property bent and stands as the thing you are aware of. That conversion, from representing-bent to being-aware-of-a-bent-thing, is the entire argument. Grant it, and the sense-datum is already in the room before any talk of hallucination or continuity. Deny it and the engine never turns over. Everything therefore turns on whether “looks F” requires an F-bearing object of awareness.

5. The book’s reply

The reply has a negative half (Austin, available to the chapter) and a positive half (the representational account, deferred to Chapter 5). Only the two together are decisive.

Austin: the Principle is a non-sequitur, and the argument equivocates. Austin’s demolition has two prongs.⁴ First, the argument trades on conflating illusion (things look other than they are: the stick looks bent) with delusion (things are conjured that are not there: pink rats). That is the sharpest instance of his broader complaint that the argument herds a heterogeneous family of cases (refraction, mirrors, perspective, after-images) into a single mold. Illusions do not require any non-physical object: when the straight stick looks bent, there is exactly one object present, the stick, which looks a way it is not. Second, and decisively: “looks bent” does not predicate bentness of anything. To say the stick looks bent is to make a relational, intentional report about how it appears. It does not report the inspection of some bent item. The inference that the Phenomenal Principle would later name illicitly reifies the look into a looked-at object. (Austin wrote before Robinson’s coinage; he targets the inference, not the label.) Austin’s complaint is not “merely verbal”: the inference at the argument’s core is invalid, and naming the Phenomenal Principle explicitly (as Robinson, defending it, was candid enough to do) exposes how much less obvious it is than the argument needs it to be.

The vehicle/content key: why the Principle is false. Austin shows the inference unearned. The representational account shows why it fails and what is true instead. The work now passes from the vehicle/object distinction (the seeing versus the thing seen, the chapter’s figure-ground flip) to its companion, the vehicle/content distinction (the state versus its accuracy-conditions), which carries the misrepresentation reply. A perceptual experience has a vehicle (a neural state) and a content (the world as represented — here, a bent stick). The bent-looking stick is a case of misrepresentation: the experience represents the straight stick as bent, and the representation is inaccurate. Misrepresentation requires no inner object that is bent; it requires only a content whose accuracy condition the world fails to meet — just as the sentence “the road bends” represents a bend with nothing bent in the ink.⁵ So the Phenomenal Principle is false at its root: a system can represent-F without there being any F-bearing item it is aware of. The “something bent” the argument insists on is the content “bent stick,” not an object that is bent. Content is not an inner thing seen but the manner in which the outer thing is (mis)presented.

The generalizing step, turned. The continuity premise (5) has a true core: illusory and veridical perception are states of the same fundamental kind. But that kind is representational, not sense-datum-involving. So the generalizing step, read correctly, supports the opposite conclusion: both cases are the world represented (accurately or not), neither is awareness of an inner object. The argument’s own continuity intuition, once the Phenomenal Principle is dropped, becomes an argument for representationalism.

The contemporary fork: who else refuses the Principle, and why the book sides with content. Rejecting the Phenomenal Principle is no longer heterodox; the live debate is how to refuse it, and two camps do. The contemporary naive realist, paradigmatically Bill Brewer, refuses it by denying that perceptual experience has representational content at all. Perception is a direct relation to mind-independent objects. Illusion is explained not by inner items and not by false content but by visually relevant similarities: the half-submerged straight stick, in those viewing conditions, bears salient similarities to paradigm bent objects. That is the whole story of why it looks bent.⁶ The content theorist (Siegel, Tye, the book) refuses the Principle by the misrepresentation route already given: the experience has content, the content is false, and no inner object is required.⁷ Both camps reject the sense-datum; they part on whether experience has content at all. (History records a third refusal — the adverbialism of Ducasse and Chisholm, on which one senses bent-stick-ly, no object required — which traded the inner item for an unanalyzed manner of sensing and could never recover the structure within experience that perception plainly has; its retreat is one reason the live contest now runs two-sided.)

The book sides with content, and the reason is not parochial. The naive realist’s content-free treatment of illusion buys its elegance with a debt that comes due in the companion entry. Having denied any common factor between veridical perception and the bad cases, the relational view must give a disjunctivist account of hallucination, treating perception and hallucination as fundamentally different kinds. It therefore inherits the negative, “merely indiscriminable” characterization of the hallucinatory state that the companion entry shows to be disjunctivism’s weakest joint. The content route refuses the Principle and keeps a single representational kind across illusion, hallucination, and veridical perception, with misrepresentation and empty content as the one unifying mechanism. So the choice between the two ways of refusing the Phenomenal Principle is not settled here. It comes due at hallucination, where the view that handles illusion most cheaply (content-free naive realism) pays most dearly on the harder case. The comparison, argued out there, tilts the book’s way. That asymmetry is the entry’s selection argument in miniature.

6. Stress-test

Objection 1 — Robinson’s defense of the Phenomenal Principle. Robinson treats the Principle as close to self-evident: if something looks bent to you, there must be something that is bent standing in the relevant relation to your awareness, on pain of denying that the look has any object at all.⁸ → Response: This simply is the reification the reply rejects, now asserted rather than argued. The representationalist grants that the look has an intentional object (the stick, represented as bent) and denies that it has a bent object. The disagreement is real and goes deep, but it is not a stand-off the sense-datum theorist wins by stipulation: the representational reading explains the same data (the look, its accuracy or failure) without the ontological cost, and is independently supported by transparency (see the Transparency entry). Verdict: the Principle is refused, not refuted from neutral ground — decisive only once the representational account (Ch5) is in hand.

Objection 2 — the continuity/spreading step stands on its own. Even without the Phenomenal Principle, the smoothness from illusion to veridical perception suggests a common analysis. → Response: Granted, and welcomed. The common analysis is representational content; the step shows both cases are world-representing, not both sense-data. Verdict: conceded and turned.

Objection 3 — Austin is “merely linguistic.” A standing complaint is that Austin catalogs ordinary usage without engaging the metaphysics. → Response: The charge misreads the work. Austin’s decisive point is that the argument’s central inference (the Phenomenal Principle) is invalid — a metaphysical verdict reached by attending to what “looks F” actually claims. What Austin lacks is the positive account of what perception then is; that the representational reply supplies. Verdict: the negative critique is sound but incomplete; it needs the Ch5 supplement.

Objection 4 — the naive realist’s rejoinder: content over-intellectualizes perception. Grant that the content route refuses the Phenomenal Principle. Brewer presses back that it does so at a price of its own: ascribing representational content to perception reinserts an intermediary, a proposition-like item interposed between perceiver and world, and so concedes the very mediation direct realism (and the book) means to deny. Better, says the naive realist, to relate the subject straight to the object and explain illusion by similarity, with nothing propositional in between.⁹ → Response: The rejoinder mistakes content for an intermediary, and the vehicle/content distinction is the answer. Content is not a thing seen but the manner in which the object is presented. What a state says about the world occupies no position between perceiver and world, so no item stands there to play the intermediary’s part. So the content route is not less direct than naive realism; it is direct and equipped for the bad cases without disjunctivism. And the similarity account faces a mirror pressure. “Visually relevant similarity” has finally to be cashed in terms of how the object looks — which is to say, how it is presented or represented — so the content the relational view set out to avoid threatens to reappear inside the notion of similarity itself. Verdict: the rejoinder names a real desideratum (directness), which vehicle/content satisfies without going content-free; the deeper cost-comparison is settled at the hallucination case, not here.

Objection 5 — the reader may convert. Chapter 3 presents the argument undefeated by design; a reader may leave it a convert to sense-data. → Response: This is precisely why the resolution lives in the appendix and in Chapter 5, and why Chapter 3’s prose flags, in voice, that the stuck feeling is deliberate and temporary. Verdict: managed by structure and framing.

7. Secondary literature

The classic modern statements, with sense-data as the inner items, are H. H. Price’s Perception (1932) and A. J. Ayer’s The Foundations of Empirical Knowledge (1940).¹⁰ The generalizing step (the move from the misperceived case to a claim about all perception) is the argument’s most questionable joint, reconstructed and dismantled in J. L. Austin’s Sense and Sensibilia (1962) and defended, via the explicitly named Phenomenal Principle, in Howard Robinson’s Perception (1994).¹¹ A. D. Smith’s The Problem of Perception (2002) is the major modern systematic treatment of both this argument and the argument from hallucination, and the natural anchor for anyone taking this entry’s argument further; Tim Crane’s “Is There a Perceptual Relation?” (2006) isolates the Phenomenal Principle as the premise on which the whole dispute turns.¹² The disjunctivist alternative to the book’s representational reply belongs primarily to the companion entry on the argument from hallucination.

The contemporary debate has moved past the sense-datum theory to a contest between two ways of rejecting it. Bill Brewer’s Perception and Its Objects (2011) is the leading statement of the object (naive-realist) view and its similarity treatment of illusion. Susanna Siegel’s The Contents of Visual Experience (2010) is the leading recent defense of the rival thesis that experience has content; James Genone’s survey work on naive realism maps the terrain and presses the hallucination problem the relational view inherits.¹³ The representational reply sits explicitly inside that contest: refusing the Phenomenal Principle with Brewer and against the sense-datum theorist, while siding with Siegel and against Brewer on whether experience has content. Smith (2002) supplies the systematic reconstruction of the sense-datum inference in its “swift” and “laborious” forms.

8. Residuals and limits

• The decisive half of the reply is deferred. Austin gets the Phenomenal Principle declared a non-sequitur; but the positive account of what “looks bent” is — content without an inner object — is the representational thesis argued only in Chapter 5. This entry’s reply is therefore complete only relative to a thesis the chapter has not yet earned, which is the book’s deliberate sequencing, not a gap.
• The Phenomenal Principle clash is a clash of starting points. A determined sense-datum theorist can keep the Principle. The representationalist does not refute it from neutral ground; he shows it unforced, costly, and independently displaced by transparency. The win is comparative, not demonstrative.
• Pairs with hallucination. This argument and its sibling share the Phenomenal Principle and a common-factor structure; the strongest pressure (Martin’s causal/screening-off argument) lives on the hallucination side and is answered there.
• The reply competes with a content-free rival, not only with the sense-datum theory. Refusing the Phenomenal Principle is now common ground with the naive realist; the entry’s real burden is the comparative case that the content route handles illusion and hallucination where the relational route handles illusion only at the cost of hallucination. That comparison is made good with the companion entry, not here — an acknowledged dependency, and the reason the two Part-One entries must be read as a pair.

---

Notes

1. The bent-stick illusion is the textbook entry point; the canonical modern statements are H. H. Price, Perception (London: Methuen, 1932), and A. J. Ayer, The Foundations of Empirical Knowledge (London: Macmillan, 1940). ↩
2. The sense-datum reading of the bent stick — “since the stick is not bent, and since the perceiver is aware of something bent, the perceiver must be aware of a sense-datum, a mental entity that genuinely possesses bentness” — is the inference the entry targets. ↩
3. The label “the Phenomenal Principle” is Howard Robinson’s: Perception (London: Routledge, 1994), 32. Its force, as Robinson concedes by stating it explicitly, is far less obvious once isolated than the argument that relies on it requires. ↩
4. J. L. Austin, Sense and Sensibilia, reconstructed from manuscript notes by G. J. Warnock (Oxford: Oxford University Press, 1962), esp. secs. III and VII. The two prongs: the illusion/delusion conflation, and the demonstration that “looks F” does not predicate F of any object of awareness. ↩
5. The vehicle/content distinction does the work: the experience’s content is a bent stick; the bentness is attributed to the (straight) stick and the attribution is false. No inner item bears the bentness. Cf. Tim Crane, “Is There a Perceptual Relation?,” in Perceptual Experience, ed. Tamar Szabó Gendler and John Hawthorne (Oxford: Oxford University Press, 2006), 126–146, on the Phenomenal Principle as the argument’s true hinge. ↩
6. Bill Brewer, Perception and Its Objects (Oxford: Oxford University Press, 2011); the “visually relevant similarities” treatment of illusion is developed there and in earlier papers (Brewer, “Perception and Content,” European Journal of Philosophy 14, no. 2 [2006]: 165–181, esp. 172–173; “How to Account for Illusion,” in Disjunctivism: Perception, Action, Knowledge, ed. Adrian Haddock and Fiona Macpherson [Oxford: Oxford University Press, 2008]). On Brewer’s account the half-submerged straight stick looks bent because, in those circumstances, it bears visually relevant similarities to paradigm bent objects — no inner item and no false content. ↩
7. Susanna Siegel, The Contents of Visual Experience (Oxford: Oxford University Press, 2010), ch. 2: the leading recent defense of the “Content View” — that perceptual experience has representational content — against the relational/naive-realist denial. The book’s misrepresentation reply is a species of the Content View, here turned specifically against the Phenomenal Principle. ↩
8. Robinson, Perception (1994), 32. The defense leans on the intuition that a look with no F-bearing object is no genuine look at all — precisely the intuition the representational account rejects by distinguishing intentional object from borne property. ↩
9. Brewer, Perception and Its Objects (2011), esp. chs. 4–5. The “over-intellectualization” worry — that ascribing proposition-like content to perception interposes an intermediary between subject and world — is a central motivation for the relational view; the vehicle/content distinction is the book’s answer that content is a mode of access, not an object of access. ↩
10. Price, Perception (1932); Ayer, The Foundations of Empirical Knowledge (1940). ↩
11. Austin, Sense and Sensibilia (1962), esp. secs. VIII–XI; Robinson, Perception (1994), esp. 31–33. ↩
12. A. D. Smith, The Problem of Perception (Cambridge, MA: Harvard University Press, 2002); Crane, “Is There a Perceptual Relation?” (2006, pp. 126–31). Smith is the systematic modern treatment and the recommended spine for a paper-length development of this entry. ↩
13. Brewer (2011); Siegel (2010); James Genone, “Appearance and Illusion,” Mind 123, no. 490 (2014): 339–376, and “Recent Work on Naïve Realism,” American Philosophical Quarterly 53, no. 1 (2016): 1–25, esp. secs. 2 and 5, which map the relational view and the hallucination problem it inherits; A. D. Smith, The Problem of Perception (2002), on the “swift” and “laborious” forms of the sense-datum inference. ↩