Keyboard Shortcuts?f

×
  • Next step
  • Previous step
  • Skip this slide
  • Previous slide
  • mShow slide thumbnails
  • nShow notes
  • hShow handout latex source
  • NShow talk notes latex source

Click here and press the right key for the next slide.

(This may not work on mobile or ipad. You can try using chrome or firefox, but even that may fail. Sorry.)

also ...

Press the left key to go backwards (or swipe right)

Press n to toggle whether notes are shown (or add '?notes' to the url before the #)

Press m or double tap to slide thumbnails (menu)

Press ? at any time to show the keyboard shortcuts

 

But One We Can Work Around

Part 1: characterisations of mental states generally

Part 2: characterisations of particular attitudes

Still the best work I have found on mindreading even more than 30 years later.

Start with

Perner’s Strategy

What is striking about Perner’s strategy is that there is a very clear separation between the scientific theory and the mindreading.
Perner invokes the scientific theory to ensure a shared understanding of mental states.
He then treats links between the scientific theory and mindreading as hypothesis

theory
(ensures shared understanding)

‘representation involves a representational medium that stands in a representing relation to its representational content.’
(Perner, 1991, p. 40)

conjecture: link to mindreading

diverse desire -> representational relation to situations

false belief -> representational relation to representations (metarepresentation)

Where other researchers seem to treat the link to mindreading as a philosophical claim, perhaps some kind of conceptual analysis, Perner treats it as a conjecture.

predictions

‘developmental levels can be theoretically tied in with my analysis of the concept of representation’
(Perner, 1991, p. 284)

alternative?: ‘Beliefs [...] represent nothing.’

(Davidson, 2001, p. 46)

Before I tell you more about what the alternative is, I first want to provide some motivation for it.

Perner’s Paradox

1. Ancient philosophers were deeply puzzled about the possibility of speaking and thinking falsely.

How can you think what is not?

2. Ancient philosophers could have passed false belief tasks.

This is what is funny about the film *The Invention of Lying*. The ancients thought there was an intellectual puzzle. But they were, we suppose, no less adept at deceipt than you or I.

3. To pass a false belief tasks is to understand a case of misrepresentation.

4. ‘Explicit understanding of representation (mentally modeling the representational relationship = metarepresentation) [...] is necessary for understanding cases of misrepresentation.’

(Perner, 1991, p. 102)

If you had an explicit understanding of representation, you shouldn’t be puzzled about the possibility of speaking or thinking falsely. That puzzle is, after all, exactly what the explicit understanding of representation is supposed to explain.

two models

Perner’s representational model

‘representation involves a representational medium that stands in a representing relation to its representational content.’
(Perner, 1991, p. 40)

Davidson’s measurement-theoretic model

We utter sentences to distinguish different beliefs.

This is like using numbers to distinguish temperatues

The numbers play no physical role.

The sentences play no psychological role.

The most sophisticated forms of mindreading involve ...

Hyp 1: ... a representational understanding.

Hyp 2: ... a measurement-theoretic understanding.

At this point you might object that I am making too much out of a difference that doesn’t really matter.
Does it really matter if, as researchers, we fail to specify how we understand mental states?
But you can see that the difference between the hypotheses matter because they generate different predictions, and these predictions have actually been considered quite widely.

predictions?

False signs matter because they involve a medium. (Since the measurement-theoretic view does not imply that they should not be correlated, perhaps there is also a theoretical reason why they should correlate on that view.)
Perner’s Representational ModelDavidson’s Measurement-Theoretic Model
Perner’s paradox

FB linked to false sign*

(inner saying?!‡‡)

FB correlates with alt. naming†

???

incompatible desires before FB‡

(competition?!††)

*Leekam, Perner, Healey, & Sewell (2008); Perner & Leekam (2008); Sabbagh (2006);
Doherty & Perner (1998); Rakoczy, Warneken, & Tomasello (2007)
††Priewasser, Roessler, & Perner (2013)
‡‡Geurts (2021)

We lack a shared understanding.

This is a practical problem.

But one we can work around.

step 1 : specify >1 theories of mental states (Perner’s Strategy+)

limit: no knowledge, intention ...!?

Part 1: characterisations of mental states generally

Part 2: characterisations of particular attitudes

So we can

dθ-both

action <- facts + objective values

dθ-belief

action <- facts + preferences

‘Diverse Desires’

dθ-desire

action <- subjective probability + objective values

‘Diverse Beliefs’

decision theory

action <- subjective probability + preferences

‘Belief--Emotion’

The advantage of decision theory is that we know exactly what subjective probability and preferences are.
This is because they are implicitly defined by the axioms that relate them to actions.
We also know from its history that it is possible to construct a family of simpler forms of decision theory ...
Now these map neatly on to Wellman & Liu (2004)’s theory of mind scale ...
For example, success on diverse desire might be taken as operationalising this model.
Note that we are not assuming anything about the notion of belief. Our notion, as researchers, is subjective probability. We note that ‘diverse beliefs’ is useful in relation to it because we are considering two hypotheses ...
Hypothesis 1: the ‘fact+objecive value model’ captures how things are from the mindreaders’ point of view.
Hypothesis 2: the ‘subjective probability+objecive value’ model captures how things are from the mindreaders’ point of view.
Wellman and Liu’s ‘Diverse Beliefs’ task is useful because the two hypotheses make different predictions about performance on it.
But note that we formulate the predictions and hypotheses without using the term ‘belief’ at all.
(The ‘Diverse Beliefs’ task is not sufficient to establish Hypothesis 2 against other alternative hypotheses, of course. For instance, Hypothesis 2 concerns a model on which agents are sensitive to the probabilities of things; this is not something that ‘Diverse Beliefs’ could be used to investigate. So of course more tasks are needed.)
And similarly for their ‘diverse belief’ task (in which you are told what another things and it is not what you think).

the postulates of the researcher

the language of the targets of research

But what would show that someone had the full model?
Note that success on both diverse desires and diverse beliefs is not sufficient. This is because they might be able to use both d-belief and d-desire models but not to use the full model.
A useful feature of decision theory is that it is almost completely neutral on the nature of the subjective probabilities and preferences. All that it requires are certain patterns. They do not have to be causes, nor do we even have to think of them as mental states in any robust sense. (As applications of decision theory in biology indicate; see, for example, Okasha (2011).)
This is useful because it means that adopting decision theory as a way of characterising states is orthogonal to which model of mental states we choose, Perner’s representational model or Davidson’s measurement-theoretic model.
Using decision theory to anchor thinking about the attitudes is compatible with adopting either model of what mental states are (and perhaps with rejecting both in favour of some alternative).

neutral on mental

We lack a shared understanding.

This is a practical problem.

But one we can work around.

step 1 : specify a theory of mental states (Perner’s Strategy+)

limit: no knowledge, intention ...!?

step 2 : exploit formal tools

limit ...

limits of decision theory

Why is decision theory not enough?

‘Polly has never ever seen inside this drawer. [...] So, does Polly know what is in the drawer?’

(Wellman & Liu, 2004, p. 539)

The problem is that in decision theory, subjective probabilities are defined in relation to your actions. There is no straightforward way of linking them to epistemic access.
And even if you could solve this problem ...

also missing:

This list of missing things is supposed to be formulated in the language of the researcher, not the language of the model.
  • strength of justification vs of confidence
  • In contrasting strength of justification vs of confidence, I’m thinking of the truth effect.
  • time and the need for planning
  • emotion
  • mood
  • humour
  • ethical constraints
  • ethical constraints: Sometimes things are strongly prohibited, so that it doesn’t matter how desirable they might be.
  • self-regulation
  • On self-regulation, I am thinking about the kind of cartwheels that Bandura (2002) has us turn in maintaining a self-image and how this links to moral disengagement.
  • ...

useful formal model would be too much to hope for

(e.g. Stalnaker, 1999 on knowledge)

p. A formal model like decision theory is too much to hope for. (see Stalnaker (1999) on the difficulty of giving a formal model of knowledge that is not merely ‘external’ but can be used to model possibilities for action) p.em-above Step 3: characterise features of attitudes in relation to particular limits of the best formal model p e.g. knowledge ??? p e.g. intention ~ meshing plans over time. (do not need to have a single notion, nor to assume that different limits come together) p.em-above this shows what is wrong with Phillips’ et al 4 signature features: they do not go beyond dθ-belief at all
: none of the tasks that Beaudoin, Leblanc, Gagner, & Beauchamp (2020) list
intention require you to go beyond decision theory (+goals).
Also, Phillips’ et al 4 signature features: they do not go beyond dθ-belief at all (Best not to see their work as concerned with Theory of Mind at its most sophisticated; that is probably not what they intend.)

Beaudoin et al. (2020, p. table 2 (part))

None of these tasks actually require going beyond decision theory; they do not require distinguishing intentions from preferences.

~Hunnius ?

you supermarket
covid so they one way-ed it
you like bananas
but you like chocolate more
and you hate being in the supermarket so have a strong preference for the shortest route
Ideal outcome: bananas and chocolate. Decision theory has a problem here (nothing to say about why it would be defective to frame your next action as a choice between bananas and chocolate).

frame 1 : chocolate or bananas?

frame 2 : chocolate only, bananas only, or chocolate+bananas?

Why is this a problem for decision theory? Nothing to say beyond that there are two ways of framing the situation.
Steve prefers chocolate to bananas.
Steve intends to buy both items.
Steve wants to leave the supermarket as quickly as possible.

Task : predict the route

Here the point is that we can operationalise a notion that goes beyond anything
Objection: why does success on the task indicate that you can coordinate the two frames rather than simply that you frame Steve’s actions in terms of the larger actions?
Reply: it doesn’t unless we add some background assumptions. The background assumptions are that you have to predict the actions in terms of the smaller units, and that the decision theoretic constraints apply to every unit of action.
Now the covid time is over and the one-way system is gone. But the route prediction should be the same.

frame 1 : chocolate or bananas?

frame 2 : chocolate only, bananas only, or chocolate+bananas?

Task : predict the route

the postulates of the
researchers

the language of the
targets

subjective probability + preferences

‘Steve intends to get the chocolate’

coordinating frames specifying different temporal perspectives

‘Steve wants chocolate more than bananas but intends to get both’

We lack a shared understanding.

This is a practical problem.

But one we can work around.

step 1 : specify a theory of mental states (Perner’s Strategy+)

limit: no knowledge, intention ...!?

step 2 : exploit formal tools

limits: time, emotion, mood, ethics ...

step 3 : specify which limits an operationalisation tests

The Myth of
Mind-
reading

Why It Matters & What to Do

It’s not just that we lack a shared understanding. I don’t think we could ever have a shared understanding because intention, knowledge and the rest are not one thing.
Nor do I think we should aim for one. Instead we should recognize that there are multiple coherent ways to model minds and actions, and we should design theory of mind tasks to work around this.
No research succeeds by unreflectively using the language of the targets of explanation in characterising physical cognition, colour cognition, or any other cognitive domain. Except mindreading. But that is something that we could change.