Contribution
The standard view locates cross-modal integration in the content of perception: specific features fused at perceptual or decisional stages (McGurk, double flash, bouba/kiki). This article adds a content-nonspecific route. Signals of processing quality from different modalities, such as ease, familiarity and coherence, combine into one global feeling, and that feeling is a component of unified experience. The account is “experiential” (the blending happens in felt quality) and “decisional” (the blended feeling is then interpreted for its source and relevance).
Key insight
Part of what makes experience feel unified may not be bound features at all. It may be a shared, “free-floating” sense of how well processing is going, and any modality can raise or lower it. Because that signal carries no record of its source, it can make the multimodal world seem more, or less, unified than it is (p. 3).
Assumptions
These are premises, since this is an opinion article.
- A function of consciousness is unity. One function of consciousness is to create a unified subjective experience that links processing streams (Baars 2005; Bayne 2010) (p. 1). This is taken from the literature, not argued.
- Processing experiences are real, and amodal. “Fringe” feelings (James 1890; Mangan 1993) such as fluency, familiarity and coherence are experienced, and can be amodal and summed across sources (Curran & Dien 2003; Miller et al. 2008) (p. 2).
- One feeling, many sources. Facilitation at different processing stages yields a similar feeling of ease (Reber et al. 2004; Wurtz et al. 2008; “but see Reber et al., 2014”) (p. 2).
- Global quality signals exist computationally and neurally. The authors cite global conflict or prediction-error signals (Fernandez-Duque et al. 2000; Friston 2010; Shackman et al. 2011; Botvinick & Braver 2015) and connectionist global-quality signals (Lewenstein & Nowak 1989; Norman & O’Reilly 2003; Cleeremans & Dienes 2008) (p. 2).
Key results
These are what the article argues and reports.
- Cross-modal coherence transfer (Ziembowicz et al. 2013, as reported on p. 2). Across three experiments, participants judged targets in one modality while incidentally exposed to regular or irregular backgrounds in another, for instance tone-grammar melodies with visual figures, or the reverse. Target regularity and background regularity were of unrelated kinds (tone-sequence grammar versus 3D realisability). Coherent backgrounds raised judgements of “regular” (Exp. 1), “familiar” (Exp. 2) and “possible” (Exp. 3), whatever the judgement type or target modality. The article gives no effect sizes.
- Evidence that related effects are experiential (p. 2): self-reports of “ease”, “effort”, “familiarity” or “regularity” in “many (though not all)” experiments (Schwarz 2015); physiological indicators of positivity linked to fluency (Winkielman et al. 2003, 2012); and discounting. In Fazendeiro et al. (2005), telling participants that background music affects their “sense of familiarity” reduced false recognition of cross-format stimuli.
- The counter-case (p. 2): with hidden semantic coherence, participants are aware only of the affective consequences of fluency, not of fluency itself (Topolinski & Strack 2009a, b).
- The mechanism sketch (pp. 2–3): global signals are non-specific and free-floating but useful. They mark abstract correspondences and regulate the network, stopping recognition when coherence is low. The illustration is a Hopfield-type network with a feedback loop on an “order parameter” such as coherence or volatility (Lewenstein & Nowak 1989). In Drogosz and Nowak’s (2006) mere-exposure simulations, it reproduces the dissociation between implicit and explicit judgements, including “knowing” valence or familiarity before recognition.
- Conclusion (p. 3): “the creation of a unified consciousness is facilitated by an experiential mechanism that combines signals of processing quality.”
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Coherence in one modality shifts regularity, familiarity and possibility judgements about unrelated targets in another. | moderate | the authors’ prior experiments (Ziembowicz et al. 2013), summarised without statistics |
| C2 | Fluency and familiarity experiences can be amodal and pool across modalities. | moderate | cited memory studies (Curran & Dien 2003; Miller et al. 2008) |
| C3 | Such cross-modal effects operate through integrated processing experience, rather than through feature-level or purely decisional integration. | weak | the authors concede that direct evidence is needed (p. 2); support comes by analogy from other paradigms |
| C4 | Processing-quality effects involve conscious feelings. | moderate | self-reports (not always), physiology, discounting studies; qualified by Topolinski & Strack |
| C5 | Global quality signals in connectionist networks explain the phenomena. | weak | a computational analogy (Lewenstein & Nowak 1989; Drogosz & Nowak 2006); no model of the cross-modal data is presented |
| C6 | Unified consciousness is facilitated by combining processing-quality signals. | weak | the conclusion (p. 3); goes beyond C1–C5, since no study measured unity of experience |
Concepts
- Processing experience or fringe experience: a felt sense of the quality of one’s own processing (ease or fluency, familiarity or oldness, coherence, rightness), after James’s “fringe consciousness”.
- Structural experiences: processing experiences about structure, such as coherence, integrity and rightness, as distinct from fluency and familiarity (Whittlesea 2002).
- Global quality signal or “order parameter”: a network-level, content-nonspecific variable (coherence, volatility, signal strength) that the system monitors to control its own dynamics.
- Misattribution or discounting: a feeling from one source is credited to another, and the effect vanishes once the true source is made salient.
Connections
The account of consciousness is functional and psychological. It presupposes that consciousness has a unifying function (a global-workspace-style framing via Baars 2005, with Bayne 2010 on unity), and it treats metacognitive “feelings” as genuine conscious experiences with causal roles. It takes no metaphysical stance. It is not deflationary: the proposal needs experienced feelings, and it names that need as its open question.
- LIT-143 (Kleiner & Ludwig) also starts from the unity of consciousness, but derives a formal topology from Bayne and Chalmers’s subsumptive unity thesis. This article offers a candidate mechanism for felt unity instead. The two are complementary and do not engage each other.
- LIT-056 (Butlin et al.) treats global-workspace architectures as indicators of consciousness in AI. This article’s “global, free-floating quality signal” is a different kind of global variable, a scalar monitor rather than a broadcast of content. Treating it as an indicator would be the reader’s extrapolation.
- LIT-185 (IIT) also puts “integration” at the centre. The integration meant here is a psychological blending of felt quality, not integrated information; the shared word should not be read as agreement.
- ANTH-LIT-669 (Kanerva, sparse distributed memory and related models) covers the Hopfield-family associative memories to which the Lewenstein and Nowak network belongs. The parallel is a network-level signal of retrieval quality, and it is an analogy only.
Consciousness tag: justified. The thesis is about how unified conscious experience is built, and it rests on claims about what is experienced. consciousness remains the right primary topic even though the evidence is from cognitive psychology, which the cognition tag already covers.
Bearing on the record
- It could support a THEORY document on the unity of consciousness, or on metacognitive feelings as integrators. It is too weak to source any strong claim, because its central causal claim is flagged by its own authors as lacking direct evidence.
- ML practice: none. A content-free global “coherence” or confidence signal that pools across modalities is loosely analogous to confidence or uncertainty signals in multimodal models. The article does not discuss ML, and no instruction for practice follows.
Limitations
- It is a short opinion article (under three pages of body text). There are no new data, no statistics for the cited studies, and no formal model of the cross-modal findings.
- The authors say themselves that direct evidence is needed that the cross-modal effects involve changed subjective experience and that the integration is causal (p. 2).
- The leap from “processing-quality feelings pool across modalities” to “this facilitates unified consciousness” is not tested. No study cited measures unity of experience.
- The authors are reporting their own prior work (Ziembowicz et al. 2013) as the key illustration.
Open questions
These are the article’s own (p. 3), with what would close each.
- Is the blending pre-experiential or experiential? Separating these would need a manipulation that removes the feeling but leaves the signal.
- Are fused experiences genuinely fused, with origin information lost, or separable? Source-memory tests after cross-modal exposure would address it.
- Which judgements (risk, frequency, truth, fame, beauty) are open to cross-modal processing experience, and are the effects pre- or post-decisional?
- How specific are processing experiences? Coherence is said to differ from familiarity or ease, “not unlike different emotions”, which should limit fusion.
Corrections to the seeded skim
- The dossier writes that the Hopfield-type model with a feedback-controlled “order parameter” (Lewenstein & Nowak 1989) “reproduces mere-exposure asymmetries”. The article attributes the mere-exposure simulations to Drogosz and Nowak (2006), using that model. The claim is that the network reproduces the asymmetry between implicit, fluency-based judgements and explicit memory judgements, with implicit measures faster than explicit ones (p. 3). This is a minor misattribution of who ran the simulation.
- The dossier omits evidence that cuts against the proposal. The article itself reports Topolinski and Strack (2009a, b): participants “cannot report and re-attribute changing levels of fluency” and are “only aware of affective (hedonic) consequences” (p. 2). The authors take this to show that what is felt varies by task. It also means the experienced signal is not always the processing-quality feeling the proposal posits.
- The dossier lists three kinds of evidence that the phenomena are experiential: self-reports, physiological markers and discounting. The article’s first item is qualified: participants in “many (though not all) experiments” report changed feelings (p. 2). Its fourth line of evidence (hidden semantic coherence, Topolinski & Strack) is the qualifying case above.
- Minor, for anyone citing through it: the reference list garbles two entries. Hillis et al. 2002 is given as “Science 298, 627–1630” (the page range is 1627–1630). Ziembowicz et al. 2013 is given as “Cognition 2, 273–278” (the volume number is missing; the “2” is the issue).