I
a problem for minimal theory of mind
Will not go over because details not important
and already published a while ago but two things to note ...
In the model, mental states are relations to objects and their features,
not representations of objects.
This makes things a lot simpler but also limits what you can do with the model.
The other thing is that we borrowed the model from a left behind theory of mind
that everyone agreed had failed. I mention this because later in the talk I’m sharing
an unfinished search for a minimal model and one of the key techniques is to womble—that
is to re-cycle early, failed attempts to provide a theory of how things actually are in a domain.
[Low, Edwards, & Butterfill (2020, p. figure 1, part) based on Kovács et al. (2010)]
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
First thing to say: it’s a ball detection task.
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Inspiration: Action anticipation can influence reaction times.
NB: In this experiment, participants had to imitate a reach-to-grasp
movement which was modelled for them before the go-stimulus (the mug)
appeared.
WHat is measured? Movement initiation times. (NB not response times.)
Costantini et al, 2010 figure 1b
Anticipation associated with other people’s action possibilities
can also influence your movement initiation times.
Costantini et al, 2011 figures 3,4
Outcomes such as reaching for and grasping of a cup can be represented motorically.
As a body of research on mirror neurons and motor simulation more generally demonstrates,
motor representations of outcomes can generate expectations concerning another agent’s behaviour
These expectations are plausibly compared with the behavior that is actually observed.
And we conjecture that the result of this comparison modulates the strength of the motor representation of the outcome.
Within limits, this modulation will ensures that an outcome represented motorically is likely to be a goal of the observed action.
In this way, motor representations enable goal tracking.
So far we have nothing about belief-tracking. Where could that come in?
Jason’s pilot indicates that belief-tracking, which I suppose involves representing registrations, can influence behavioural expectations.
But how?
In principle, we might imagine that the belief-tracking process results in a second, independent behavioural expectation.
While we cannot rule this possibility out, it seems to add theoretical complexity.
After all, belief-tracking depend so on goal-tracking in this way: you can only track another’s mental states by tracking their actions. We therefore need the goal-tracking process to provide input to the belief-tracking process.
But if the goal-tracking process ignores the belief-tracking process, then false beliefs would cause it to make systematic errors about the goals of actions. Since these errors would feed into the belief-tracking process, it would seem that this process too should go wrong whenever there is a false belief. But then there would be no point in tracking beliefs at all.
Our proposal is therefore different: the belief-tracking process influences the process by which the behavioural expectation is generated.
The representation of registration means that behavioural expectation is generated as if things were as they are registered as being, rather than as they actually are.
Put colourfully, another’s registration can change the world as seen by your motor system.
(This might be why Katheryn has repeatedly found altercentric intereference effects of belief on motor actions.)
This picture has a radical implication about the nature of the automatic belief-tracking processes.
It implies that those belief-tracking processes must interface directly with motor processes.
This requires, in turn, that automatic belief-tracking represents objects and outcomes in the same format as the motor representations do.
Motor processes and belief-tracking must share a common representational format.
Although lots of details are not specified by the picture, its is does make readily testable predictions. For one thing, it implies that limits on what can be represented motorically are also limits on automatic belief-tracking.
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 1, part) based on Kovács et al. (2010)
Low et al. (2020, p. figure 2)
Nice replication, which is mildly interesting because we used a movie clips
with real people rather than smurfs.
Low et al. (2020, p. figure 3)
Six (2022, p. figure 11)
‘participants in the movement-restricted group were required to hold on to the elastic band
at the start of every experimental block.’
(Six, 2022, p. 98)
[Six (2022, p. 95) Experiment 1: ‘predictive gaze behaviour was modulated by the false belief state of the agent’]
So far, so good.
[In Experiment 1, we had no response to the videos required but there was a visual detection task - press a key when
one of the cups turns purple, which we expected would help to keep people attentive and ‘disguise the real purpose
of the expeirment’.]
The basic finding that false belief states modulate predictive gaze behaviour was partially replicated in Experiment 2,
(although differently from Experiment 1, ‘In the false belief condition, we saw that anticipatory looking times generally
sat around chance level.’ but this could be due to some technical difficulties with the equipment?)
[Ask Pieter: I think the attention check still involved a keypress, which the participant could do with the elastic band?]
However the motor mindreading prediction was not supported:
‘Our crucial manipulation in this experiment was the addition of a group that was motor-restricted and
therefore unable to grasp for a cup, which should in principle attenuate motor simulation and action prediction.
This, however, was not the case.’ (Six, 2022, p. 111)
[Also Six (2022, p. Experiment 3) did not find evidence that false belief states modulate predictive gaze behaviour.
I think this may be because we asked participants to predict the action on each trial
(‘Unlike previous experiments, this experiment featured a question prompt (“on which side of the screen
will the hand appear?”) after each trial upon participants were asked to respond vocally with “left” or “right”.’)
Notably, DLTS and explicit behavioural predictions were tightly correlated; this suggests to me that
asking the question may have altered which processes moved the eyes.]
Zani, Butterfill, & Low (2020)
| source | paradigm | measure | who bound? | as predicted? |
| Low &c, 2020 | Kovacs’ Smurf | RT | agent | Y |
| Six, 2022 Exp. 2 | Six’ Cups | looking time | participant | N |
| Zani, forthcoming Exp. 2 | Buttelmann/ helping | leaning | agent | suggestive |
How do agents ever perform optimally when time is pressing and cognitive resources such as working memory are scarce?
a. automatic → minimal
b. motor mindreading conjecture: minimal models of mind are implemented motorically