Week 7 Seminar: The self-model

The modeller starts modelling itself, and finds that introspection reports a summary rather than the process that produced it.

Part II — Interior models8 April 2027Week 7 lectureSlidesSchedule

A system that models its environment can model itself, and the moment it does there’s a constraint waiting: the self-model has to be smaller than the system. A complete model of a process, running inside that process, is not available. So what introspection reports is a summary — and summaries omit, without marking what they omitted.

Metzinger’s Being No One is the ambitious philosophical version. The empirical work is more unsettling than the philosophy.

Choice blindness#

Johansson and colleagues had people choose the more attractive of two faces, then covertly handed them the one they had rejected and asked why they picked it. Fewer than one in ten manipulations were caught. Under 20% across a whole session. People then explained their choice, citing features specific to the face they had actually turned down.

Beforehand, 84% said they would notice.

The objection, which you should arrive knowing, is Petitmengin and colleagues’ 2013 result: when participants are given room to examine their experience rather than pressure to produce an explanation, the effect largely dissolves, sometimes with recognition up to 45 minutes later. So the strong reading — introspection is simply unreliable — does not survive. Grade the strong version [C], and note that the weaker version is more interesting anyway: introspection is unreliable under the conditions where we usually deploy it, which is while being asked to account for ourselves.

The mirror test, as a lesson in method#

Kohda and colleagues reported in 2019 that cleaner wrasse pass the mark test. De Waal’s commentary in the same journal argued the marks were physically irritating and the behaviour ambiguous, so the result falls short of mirror self-recognition. Gallup, who devised the test, objected. Kohda’s group returned in 2022 with eighteen fish and a 94% pass rate.

We are not going to settle it. The reason it’s here is that it shows a behavioural assay carrying more interpretive weight than it can bear, and the argument is really about what the test was ever a test of. Which is a philosophical question wearing a methods section.

Reading#

The paper this seminar is built around:

• Johansson, Hall, Sikström and Olsson (2005). Failure to detect mismatches between intention and outcome in a simple decision task. Science.

The rest of the week’s reading:

• Petitmengin and colleagues (2013). A gap in Nisbett and Wilson’s findings? Consciousness and Cognition — the objection you are expected to arrive with.
• Nisbett and Wilson (1977). Telling more than we can know. Psychological Review.
• Kohda and colleagues (2019). If a fish can pass the mark test… PLOS Biology.
• de Waal (2019). Fish, mirrors, and a gradualist perspective on self-awareness. PLOS Biology.
• Kohda and colleagues (2022). Further evidence for the capacity of mirror self-recognition in cleaner fish. PLOS Biology.
• Metzinger (2003). Being No One. MIT Press — the philosophical version, and long.

Grades are not attached to papers here, because a grade belongs to a claim rather than to the paper it appears in — see evidence grades. Several of these papers contain claims at two different grades.