Source
Pernu (2018), LIT-147 (Synthese 195(12), 5159–5174), read in full in NOTE-091.
What was actually shown
The framework. C causes E iff C □→ E and ∼C □→ ∼E (p. 5161). Under multiple realisation, the neural realiser N fails the second condition and the mental property M meets it, so M and not N is the difference-maker (p. 5162). That is the List and Menzies result this account presupposes (THEORY-133).
The argument it targets (§4, pp. 5164–5165). Woodward (2008a) and others read brain–computer interface work as showing an intention realised by different neural patterns on different occasions, so citing any one pattern is “overly specific”.
The diagnosis (§§4–5). The many patterns are single-neuron realisers of one cortical-level value N, “the average firing rates of neurons in the relevant ensembles” (pp. 5167–5168). They do not realise the intention directly (Fig. 2a). The argument needs an intention realised by distinct cortical states N and Z (Fig. 2b), and neuroprosthetics does not present that. The redundancy is micro over macro within the physical, like microstates under a thermodynamic macrostate.
The consequence (pp. 5168–5171). With N read as the cortical variable, its counterfactual ∼N □→ ∼B holds whenever the mental one does, so the contrast the argument needs collapses. The mental is “not an epiphenomenon”, but only “because the mental has now been identified with the physical” (p. 5168). Preferring coarse over fine causes is compatible with physicalism, since “physicalism is not microphysicalism” (p. 5171).
What could have come out otherwise. The decoding evidence could have shown one intention realised by two distinct ensemble-level states. On Pernu’s reading it does not, but that reading is asserted, not shown.
What this does not say
- That intentions are causally idle. The conclusion is that they are causes because they are identical to cortical states: the deflationary defence of mental causation, not the substantial one (§6).
- That the empirical premise is established. That one intention equals one ensemble-average value rests on the research agenda and a quotation from Musallam et al. (2004). Decoders are not simple averages, and the mapping is not checked across sessions, electrodes or subjects (NOTE-091, C3: weak).
- That interventionism in general fails. Footnote 1 confines the argument to difference-making with proportionality. Replies that do without proportionality “are not affected”.
- That cross-species multiple realisability fails. That monkey-to-human transfer tells against it is one sentence (p. 5170).
- That higher-level causes are never autonomous. It concerns this evidence for this case. THEORY-133’s criterion stays system by system.
Connections
- Presupposes THEORY-133. It takes List and Menzies’s criterion as given and argues that the brain–computer interface case fails it. THEORY-133’s summary already notes that Pernu disputes its main example.
- QUESTION-033 asks whether “who drives?” has an answer at a given level. This constrains what can count as an answer: many-to-one realisation shows that some coarse variable drives, and that variable may be physical, not the named higher level. Evidence of degeneracy alone does not put the drive at the mental level.
- THEORY-048 and THEORY-057 already cite this paper for the same caution about inferring a distinct higher-level cause from many-to-one realisation.