NOTE-998: Hohwy & Paton — Explaining away the body

Contribution

A version of the rubber hand illusion in which the foreign arm is seen, through a head-mounted display, in the very place the participant’s own arm lies, so that there is no proprioceptive discrepancy to overcome. In that set-up the authors push the illusion with input that should break it: a tapping finger lifted off the seen arm, and the seen arm replaced by a cardboard box. Ratings stayed higher under synchrony than asynchrony in both, and people volunteered supernatural explanations of the hovering finger. A box shown without prior onset gave no synchrony difference. The experiments were run by a philosopher to support a probabilistic account in which the body image is “explained away”.

Key insight

Once the system has settled on the seen arm as the place where touch is happening, new conflicting input is folded into that solution rather than used to overturn it: touch is felt to be caused across a gap, or caused on a box. On the authors’ reading, the prior that one’s body looks like a body is not a fixed constraint on multisensory integration but a hypothesis that loses probability as integration proceeds. Whether a non-body object is accepted depends on what came before, not only on what the object is.

Assumptions

• A single rating measures the illusion. Each experiment analyses one question per condition, scored −3 to +3 (Table 1). None asks whether the seen arm feels like one’s own. Experiment 1’s item is about where touch is produced and by what; experiment 2’s are about what caused the touch “even though there was a visible gap” and whether touch was caused “on the box”.
• Skin temperature marks the illusion. Temperature of the real hand, averaged over two sensors, is taken “as an independent measure of the general validity of this design”, following Moseley et al. (2008) (Methods, Temperature measurements). Drift could not be used, since the arms are aligned.
• Asynchrony is the comparison. Each participant had both synchronous and asynchronous tapping (about 1 tap/s; asynchrony 500 ms to 1 s), counterbalanced. “Participants were aware of the set up.” Expectations were not measured.
• Two control items guard against bias. “Did the touch you could feel become painful?” and “Did it feel as if you had two bodies?” were asked in experiments 2 and 3; their values are not reported.
• Open-ended reports are evidence of a genuine illusion. The authors argue that if touch were not genuinely felt on the rubber arm, people “would not … elaborate lack of physical contact in supernatural terms” (Discussion).

Key results

Sections are the paper’s. Sample sizes are from Methods, Participants.

• Set-up (Methods, Fig. 1). The display showed experimenter B’s right forearm (experiment 1), a rubber hand (experiments 2–3) or a white cardboard box, “positioned to appear in the head mounted display to be spatially coincidental with the participant’s own right arm”. Experimenter A tapped the real forearm and the seen one near the wrist.
• Experiment 1, continuous touch (n = 13). Ratings after 10–20 s and after 30 s of tapping, given orally “to ensure illusion onset”. Synchronous M = 1.04 (SD 2.46), asynchronous M = −2.50 (SD 1.14); main effect of tapping F(1, 17) = 72.90, p < .01; no effect of time. The error degrees of freedom (17) do not fit 13 participants in a within-subject design, which would give 12; the paper does not explain this.
• Experiment 1, intermittent touch (same 13). After onset, pauses of 10, 30 and 60 s with no tapping, each ended by one tap. Synchronous M = 1.81 (SD 1.02), asynchronous M = −2.19 (SD 1.10); F(1, 17) = 135.06, p < .01; no effect of time. “Continuous stimulation is not necessary to maintain the illusion over the periods tested.”
• Temperature (Results, Experiment 1; Fig. 4). Over the last 60 s of the intermittent condition the real hand was 0.13 °C cooler under synchrony (M 28.92, SD 2.27, against 29.05, SD 2.31), t(12) = −1.84, p = .04. In the continuous condition, over 30 s, t(12) = −1.36, p > .09. By my computation these are one-tailed values (two-tailed .09 and .20), and the Bonferroni correction the Methods announce for these t-tests would not leave the first significant. In experiments 2 and 3 “There was no significant difference in temperature in any of these experiments, all ps>.05.”
• Experiment 2 (n = 11 elevated, n = 10 box; between groups). Elevated touch: 60 s of tapping on the rubber arm, then the moving finger lifted “approx. 5 cm” off it for 3 more minutes while the real arm was still tapped. Box: 30 s on the rubber arm, then the camera switched to the box for 2 min 30 s. Ratings in writing after each condition, “to avoid possible bias”. Main effect of tapping F(1, 19) = 51.71, p < .01; synchronous M = 1.03 (SD 1.74), asynchronous M = −2.06 (SD 1.58); no effect of condition and no interaction. Per-condition means are not given.
• Open-ended reports (Table 2), elevated touch. “There are opposed magnets on the finger and the skin”, “It is witchy”, “It is black magic”, “There is an invisible extension on the finger”, “It’s ESP”, “It’s telekinetic”, “A magician makes my muscles contract”. How many participants gave such reports, and under which condition, is not said.
• Experiment 3 (n = 9). Three minutes of tapping on the real hand and a seen box, with no rubber arm first. “A paired samples t-test revealed no significant differences between participant scores in synchronous touch versus asynchronous touch (p>.05).” No means, t or effect size are reported. (The Methods point to Fig. 3c, the elevated finger, for this condition; Fig. 3d, the box, is meant.)
• Controls. No synchrony difference on the pain and two-bodies items in experiments 2 and 3 (p’s > .05); the values are not reported, and the items were then “excluded from further analyses”.
• Reconciling earlier results (Discussion). Tsakiris & Haggard’s failure to get touch referred to a non-hand object “could be due to not introducing this subsequent to onset of the basic RHI”, and Armel & Ramachandran’s table-top success to having first exposed people to the rubber hand. “Our experiment 3 support this analysis.”
• Explaining away (Discussion). “The internal model that represents visual body image and proprioceptive body schema decreases its probability in the face of ongoing solutions to visuotactile conflict. Thus, increasingly odd somatic, causal and tactile experiences can occur as the prior body representation is ‘explained away’.” An alternative for the box is offered: the hand felt “inside the box”, touched through it.
• Temperature, read (Discussion). The cooling “may arise because the foreign arm is somehow processed as non-self”, though “there is as yet no well-established explanation of this phenomenon”.
• Delusion (Discussion, Wider significance). Participants “do not truly believe” the supernatural explanations, perhaps because they know they could test them. Healthy people’s readiness to offer them “gives some support for one-deficit theories of delusions”; “a second deficit of rationality … may not be needed for these rather bizarre experiences to arise”. The authors mark this as speculation.

Claims

id	claim	strength	support
C1	With the seen arm aligned with the real one, synchronous tapping raises ratings that touch is produced on the seen arm, against asynchrony	moderate	n = 13, F(1, 17) = 72.90 and 135.06; one leading item, oral ratings in front of the experimenters; df inconsistent with n
C2	The rating persists through pauses of up to 60 s with no touch	weak to moderate	same 13; no time effect, a null with no equivalence test
C3	After onset, synchrony raises ratings that a hovering finger causes the touch, and that touch is caused on a box	moderate as a report difference	n = 21 in two groups, F(1, 19) = 51.71; per-condition means not given; items worded to the effect
C4	The box effect requires prior onset of the basic illusion	weak	a non-significant paired test in 9 people, no values; no direct comparison with experiment 2’s box group, which differed in sample and duration
C5	The real hand cools during this illusion, replicating Moseley et al.	weak	0.13 °C in one comparison of four, significant only one-tailed and uncorrected; absent at 30 s and in experiments 2–3
C6	The visual body image is “explained away” by ongoing multisensory inference	weak	an interpretation of C3–C4; no model is fitted and the prior is not measured
C7	Unusual experience suffices for supernatural explanation without a second deficit, supporting one-deficit theories of delusion	weak	open-ended reports of unstated number; the participants did not believe them

Concepts

• Explaining away — used from Pearl’s Causality: once one hypothesis explains the input, a rival hypothesis, here the model of one’s visual body image, loses probability. The authors use it informally.
• Basic illusion — touch felt as “produced by the finger you can see and on the arm you can see”, with the seen arm aligned in personal space. It is referred touch and seen causation. The paper does not use the word ownership for it.
• Supernaturally caused (elevated) touch — touch felt as caused by a seen finger that does not contact the seen arm.
• Psychologically induced temperature change — Moseley et al.’s finding, as reported here, of cooling of the real arm “during synchronous but not asynchronous touch”, stronger with stronger illusion ratings.
• One-deficit theory of delusion — the view, credited to Maher, that delusions are normal responses to unusual experiences, against two-factor views (Coltheart) that add a deficit of belief evaluation.

Connections

The set-up is a single-limb version of Ehrsson’s 2007 full-body illusion, in which seen and felt touch coincide in personal space, and the authors place their account beside Schwabe & Blanke’s computational model of out-of-body experience and Metzinger’s self-model. The non-hand question is the dispute between Armel & Ramachandran (2003), who got touch referred to a table top, and Tsakiris & Haggard (2005) and Tsakiris et al. (2009), who did not get it for a non-hand object. For the box they also offer a peripersonal-space account (Makin, Holmes & Ehrsson 2008; Graziano), and predict that elevated touch should fail beyond about 30 cm from the hand. The delusion section joins Hohwy & Rosenberg (2005) and Fletcher & Frith (2009). None of these was read here.

Bearing on the record

• THEORY-769, the “real part loses ownership” clause. Apps & Tsakiris (LIT-1262, read in NOTE-991) cite this paper with Moseley et al. for the real hand cooling as the rubber hand gains ownership. What the paper shows is thinner: 0.13 °C in one comparison of four, significant only by what reads as an unstated one-tailed test, absent in the 30 s comparison and in all of experiments 2 and 3, with an experimenter’s real arm as the foreign arm, and with no ownership item to show the arm was owned. The authors themselves say “there is as yet no well-established explanation” of the cooling. Abdulkarim & Ehrsson (NOTE-992) report that temperature changes “appear to be quite difficult to reproduce”. The THEORY’s promote_when already asks for a second result on a measure other than skin temperature, so this reading does not change its status; it shows that the temperature support the source leans on is weaker than “cooling of the real hand” suggests.
• THEORY-769, the prior. Its refutation condition is that objects given the same correlations are owned to the same degree “whatever their prior plausibility as body parts”. The box result is the closest the record has to a test of the prior, and it neither meets nor refutes the condition. It measures caused touch, not ownership. And the authors’ reading, that the body-image prior falls as the illusion runs, is exactly the move THEORY-769’s “What this does not say” warns of: a prior that can be lowered after the fact explains acceptance of the box after onset and its rejection before. Experiment 3 is the one part that could have come out otherwise, and it is a null in nine people with no values. A test of the prior would need body-likeness and prior exposure varied together, with ownership measured.
• THEORY-768 and Lenggenhager et al.’s body-likeness constraint. Lenggenhager et al. (LIT-1279, NOTE-999) found that a box got touch felt on it but no self-identification and a weaker drift, and read that as a cognitive constraint, “rejecting a pure Bayesian account”. This paper, at the level of a hand, offers a way round the constraint (prior onset) and a Bayesian reading of it. The two do not conflict: both find touch referred to a box, and neither finds it owned, because this paper never asks. Its “two bodies” control item gave no synchrony difference, with values unreported. That is a whole-body item unchanged during a hand illusion, which is the shape of the first half of THEORY-768’s promotion test, but it does not meet it: it was a control, not a measure of self-location or self-identification, hand ownership was not measured either, and expectations were not.
• THEORY-766. Four more synchrony differences on reports, here on referred touch and seen causation, consistent with that THEORY’s referred-touch clause. They have only an asynchronous condition as control, which that THEORY’s promote_when says “cannot settle it”, and nothing on ownership. Expectation is unmeasured. The items are leading: the elevated-touch question supplies the reading (“even though there was a visible gap”), experiment 1’s ratings were given aloud during tapping, and each participant had both conditions. The supernatural reports, which the authors take as proof the illusion is genuine, are what imaginative compliance with a leading question would also produce. That is the record’s reading against the suggestion debate (Lush 2020, LIT-1276; Lush & Seth 2022, LIT-1267), not the authors’. One result sits less easily with a pure expectation account: experiment 3, where a guessable synchrony effect on the box item did not appear without prior onset. With nine people and no values reported, it cannot carry weight. The paper does not bear on the drift clause: drift could not be measured, and the set-up removes the proprioceptive gap that drift measures. That referred touch arose with no gap fits the THEORY’s separation of the report from drift without testing it.
• THEORY-767. No bearing. Temperature is an autonomic measure, which that THEORY leaves out, and nothing interoceptive was varied.
• THEORY-458. No bearing beyond the general point that the illusion is produced in a passive arm.
• NOTE-372, claim C6, and THEORY-064. The authors align the “explaining away” of the body image with Metzinger’s self-model. The paper shows the representation of where touch happens can be moved far, not that a self-model is all there is.
• QUESTION-048. A case of report and belief coming apart: people rate a supernatural cause and describe it vividly, but “do not truly believe” it. The paper treats the ratings as experience and the disbelief as belief. That a rating of “as if” reports an experience, and not an entertained description, is assumed rather than shown.
• No instruction for machine-learning practice.

Limitations

• Samples of 13, 11, 10 and 9, one laboratory. No effect sizes; the experiment 1 degrees of freedom do not match its sample; experiment 3 and the controls report no values.
• No ownership item anywhere. The illusion is referred touch and seen causation, so the paper says nothing direct about ownership of the arm, despite being cited for it.
• No expectation measure. The predicted items are worded toward the effect, the controls (pain, two bodies) are implausible enough to be rejected under either condition, and experiment 1’s ratings were given orally during tapping.
• The temperature result is one comparison of four, one-tailed by my computation, and not corrected as the Methods say. The foreign arm in that experiment was a real human arm, not a rubber one.
• The claim that the box needs prior onset compares two different groups with different durations (2.5 against 3 minutes of box tapping) without a test between them, on a null in nine people.
• The open-ended reports are not counted or tied to conditions.
• “Explaining away” is named, not modelled. The delusion argument is marked speculative by the authors.

Open questions

• Does the box (or elevated touch) gain ownership as well as referred touch, and does prior onset matter for that? An ownership item and an expectation measure in a design crossing object type with prior onset would answer it.
• Is there any real-hand cooling in this design with two-tailed, corrected tests and a larger sample? A failure would leave THEORY-769’s “loses ownership” clause with less physiological support than its source claims.
• Does elevated touch fail beyond about 30 cm from the hand, as the authors’ peripersonal-space reading predicts?
• Do participants who give supernatural descriptions score higher on trait suggestibility or imaginative absorption?