Mindreading: Signature Limits, and Development
s.butterfill@warwick.ac.uk & c.sinigaglia
Lin et al (2010, figure 3)
‘Fig. 3. Latency of decision window for the target as a function of cognitive load and presence of a competitor (Experiment 2).’
The task is a director study. ‘competitor present’ means there are two possible objects
only one of which the director can see.
The ‘decision window’ in ‘the time difference between first noticing the target and finally reaching for it’
Note that the decision window widens under cognitive load only when a competitor object is
present, i.e. only when the task requires perspective taking!
Minimal theory of mind is the psychological equivalent
of impetus mechanics.
Kozhevnikov & Hegarty (2001, figure 1)
\begin{center}
\includegraphics[width=0.25\textwidth]{fig/signature_limits_table.png}
\end{center}
\begin{center}
\includegraphics[width=0.3\textwidth]{fig/low_2012_fig.png}
\end{center}
There is some evidence that this prediction is correct.
Jason Low and his collegaues set out to test it.
They have now published three
different papers showing such limits; and Hannes Rakoczy and others
have more work in progress on this.
Collapsing several experiements using different approaches,
the basic pattern of their findings is this ...
Take non-automatic responses first; in this case, communicative responses.
When you do a false-belief-identity task, you see the pattern you also
find for false-belief-locations tasks.
But things look different when you measure non-automatic responses ...
The non-automatic responses all show the signature limit of minimal
models of the mental.
This is evidence for the hypothesis that
Some automatic belief-tracking systems rely on minimal models of the mental.
I also hear that quite a few scientists have pilot data that speaks
against this signature limit.
One particular task for future research will be to examine whether other
automatic responses to scenarios involving false beliefs about identity,
such as response times and movement trajectories, are also subject
to this signature limit.
Windows light up, you will look in anticipation
Low et al, 2014 figure 2
Another experiment got similar results (rabbits quack):
this graph is nice because it illustrates how you can dissociate
the proactive gaze from the verbal response (make it uniformly good,
or uniformly bad)
Just say that you can do this with other stimuli and paradigms, and we
have done this with infants and would like to do it with adults.
These findings complicate the picture: is helping driven by automatic processes only?
If not, why do we predict that the signature limit of minimal theory of mind
is found in this case too?
Edwards and Low, 2019 figure 1
Here’s what the participants (P) see. They’re looking at a video of an agent (A).
The person can see two balls. The subject can also see two balls.
You’re told that at the end of each clip, you will see either the red ball
or the blue ball. You have to press a key corresponding to which ball you see.
Ignore the agent. He’s completely irrelevant to the task.
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 1
Key finding is P-A- vs P-A+
automatic prediction
Edwards & Low 2019 tests both predictions simultaneously.
(Long history that I’m not mentioning.)
Kovacs-like but with two balls, red and blue, always one goes.
Beliefs concern which belief remains
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 4 (part)
Key finding is P-A- vs P-A+
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 2
Kovacs-like identity starring the robota
Edwards and Low, 2019 figure 1
Edwards and Low, 2019 figure 4 (part)
Two Systems Theory of Mindreading
Two (or more) processes concerning X are distinct:
the conditions which influence whether they occur,
and which outputs they generate,
do not completely overlap.
One process makes fewer demands on scarce cognitive resources than the other.
(Terminology: fast vs slow)
further evidence (only indirect) ...
Edwards and Low, 2017 figure 7a
Edwards and Low, 2017 figure 7a
Edwards and Low, 2017 figure 7a
Any attempt to answer questions
about when children first come to know facts about objects or minds
immediately runs up against an initially puzzling
pattern of findings.
Suppose you ask when humans first know simple facts in a particular domain.
If you measure responses such as verbal predictions (or final key presses
given in response to verbal instructions),
in almost any domain, you will find evidence for a developmental transition.
Importantly, you will find this initially puzzling pattern even if you use the same scenario and
subjects.
For example, here is Jason who wants to retrieve a red robot from under of these boxes.
When these windows light up,
he will reach through one of these windows to retrieve it.
In fact the red robot is in the right box.
But Jason has a false belief about it’s location: he falsely believes that it is in the left box.
We can measure whether you take Jason’s belief into account in two ways.
We can observe where you look in anticipation of Jason’s action when the windows light up
(initial proactive gaze).
And then, a few seconds later, we can ask you where you think Jason will look first.
Extending groundbreaking work by Clements & Perner (1994) and many following them,
Low and colleagues have done several variations on this, with 3- and 4-year-olds as well
as adults and found just the pattern here: an interaction with age.
At this point some are tempted to claim that this pattern of findings is not anomalous at all.
It’s simply a consequence of the fact that the some responses provide a more sensitive measure
of competence, or that other responses require some feature extraneous to the task.
But this claim is less persuasive given a further piece of the evidence.
For some carefully matched tasks, responses like initial proactive gaze and looking time
indicate very poor, below chance performance.
This is hard---not impossible, but hard---to reconcile with the claim that the
these responses simply provide more sensitive measures.
It is also hard to reconcile with the claim that the interaction with age is due to
extraneous features introduced when responses like verbal prediction are involved.
Low et al, 2014 figure 2
(Incidentally, I admint that I’m using a made up figure to illustrate a theoretical point.
But it’s actually not too far from how real data from even a single study
sometimes looks.)
I already showed you the evidence on this.+
Low et al, 2014 figure 2
Scott et al (2015, figure 2b)
Scott and colleagues (Scott, Richman, & Baillargeon, 2015) provided other evidence
suggesting that infants’ mindreading may be relatively sophisticated.
Specifically, 17-month-olds watched a thief attempt to steal a preferred
object (a rattling toy) when its owner was momentarily absent by substituting
it with a less-preferred object (a non-rattling toy). Infants looked longer
when the thief substituted the preferred object with a non-visually-matching
silent toy compared to when the thief substituted it with a visually-matching
silent toy. The authors postulated that infants can ascribe to the thief an
intention to implant in the owner a false-belief about the identity of the
substituted toy. The authors further suggested that infants make such ascriptions
only when the substitution involves a visually-matching toy and the owner will
not test whether the toy rattles on her return.
However, Scott et al.’s (Scott et al., 2015) explanations also
require postulating that infants take the thief to be strikingly inept;
despite having opportunity simply to pilfer from a closed box known to
contain at least three rattling toys, the thief engages in elaborate
deception which will be uncovered whenever the substituted toy is next
shaken and the thief, as sole suspect, easily identified. A further
difficulty is that factors unrelated to the thief’s mental states vary
between conditions, such as the frequencies with which toys visually matching
one present during the final phase of the test trial have rattled. These
considerations jointly indicate that further evidence would be needed to
support the claim that humans’ early mindreading capacity enables them to
ascribe intentions concerning false beliefs involving numerical identity.
The model of minds and actions underpinning automatic mindreading process does not
significantly change over development.
By contrast, the model of minds and actions underpinning nonautomatic mindreading process
does significantly change over development.
In the first three or four years of life, nonautomatic mindreading processes involve
relatively crude models of minds and actions, models which do not enable belief tracking.
What changes over development is typically just that the model underpinning nonautomatic
mindreading becomes gradually more sophisticated and eventually comes to enable belief
tracking.
Conjecture:
#. Automatic and nonautomatic mindreading processes both occur from the first year of life onwards.
#. The model of minds and actions underpinning automatic mindreading process does not
significantly change over development.
#. In the first three or four years of life, nonautomatic mindreading processes involve
relatively crude models of minds and actions, models which do not enable belief tracking.
#. What changes over development is typically just that the model underpinning nonautomatic
mindreading becomes gradually more sophisticated and eventually comes to enable belief tracking.
Low et al, 2014 figure 2
There is what Low et al’s results are really showing us.
(Low, Edwards, & Butterfill, 2020, p. figure 1)
(Low et al., 2020, p. figure 2)
(Low et al., 2020, p. figure 3)