--------
--------
My topic is mindreading, the process of tracking another’s mental states.
Let me start with a canonical illustration of mindreading in action.
--------
So where the nonmindreader uses facts to generate predictions about actions, the
mindreading attributes beliefs.
--------
In many cases, you will get the same predictions whether you are mindreading
or just using facts. But in cases like this, where there are false belief,
the predictions can come apart.
\textbf{This means that we can detect mindreading by measuring action predictions.}
--------
In what follows I’m going to focus on belief.
The existence of mindreading processes for tracking beliefs raises many questions:
\begin{enumerate}
\item Why is belief-tracking in adults sometimes but not always automatic? (or automatic to varying
degrees)
\item How and why is automatic belief-tracking limited in ways that nonautomatic belief-tracking is not?
\item How and why are inhibition, attention and working memory involved in belief-tracking?
\item Why is there an age at which children pass some false belief tasks but systematically fail
others?
\item What feature or features distinguish the tasks these children fail from those they
pass?
\end{enumerate}
I think there is more than one kind of mindreading process which tracks beliefs.
That’s why my title is ...
--------
NB: mindreading defined as *tracking* mental states.
--------
--------
In this experiment by Brian Hare and colleagues, a subordinate chimpanzee makes predictions about a
dominant chimpanzee’s ability to retrieve food. They found that the subordinate’s predictions take
into account whether the dominant’s view was blocked while the food was placed. This could be
explained by the Third Principle. For the subordinate to predict that the dominant will not be able
to recover the food, it is sufficient to think: because the dominant did not encounter the food,
she will not be able to retrieve it.
‘In informed trials dominant individuals witnessed the experimenter hiding
food behind one of the occluders whereas in uninformed trials they could
not see the baiting procedure. In misinformed trials, dominants witnessed
the experimenter hiding food behind one of the occluders, and once the
dominant’s visual access was blocked, the experimenter switched the food
from its original location to the other occluder’ \citep{Hare:2001ph}.
--------
‘Mean percentage ±SE of ... (b) trials in which subordinate subjects chose not
to approach as a function of whether the dominant competitor was informed,
uninformed, or misinformed about the location of the food in experiment 1.’
--------
We need to talk about how to evaluate, and write about, scientific research.
Consider how the results of Experiment 1 is reported
‘subordinate subjects retrieved a significantly larger percentage of food when dominants lacked accurate information about the location of food
‘(Wilcoxon test: Uninformed versus Control Uninformed: T=36, N=8, P<0.01; Misinformed versus Control Misinformed: T=36, N=8, P<0.01)’
\citep[p.~143]{Hare:2001ph}
--------
‘the jays were much more likely to re-cache if they had been observed by a conspecific while they
were caching than when they had cached in private. By re-caching items that the observer had seen
them cache, the cachers significantly reduce the chance of cache theft, as observers would be unable
to rely on memory to facilitate accurate cache theft’ \citep[p.~516]{Clayton:2007fh}.
--------
First show that window open vs window closed affects caching behaviour.
Then familiarise ravens with the peephole, letting them use it while humans
cache and letting them ‘steal’ the human-cached food.
Now test them by allowing them to cache when there’s apparently (from the sound,
which is actually a recording)
a raven on the other side.
--------
Conclusion 1:
‘Peephole designs can allow researchers to overcome the confound of gaze cues’ \citep{bugnyar:2016_ravens}.
--------
--------
Researchers sometimes use the term ‘theory of mind’.
‘In saying that an individual has a theory of mind, we mean that the individual [can ascribe] mental states’
\citep[p.\ 515]{premack_does_1978}
--------
New term. Is this is the same thing as mindreading?
--------
Mindreading : tracking others’ mental states
--------
A \textit{theory of mind ability}
\label{df:tom_ability}
is an ability that exists in part because exercising it brings benefits
obtaining which depends on exploiting or influencing facts about others’ mental states.
--------
Not all theory of mind abilities depend on theory of mind cognition.
For example, preening others may be worthwhile in part because
it influences their attitudes towards oneself and thereby
strengthens social bonds \citep{Clayton:2007ob}.
Where this is so, preening is a theory of mind ability.
It doesn’t follow, of course, that preening involves theory of
mind cognition.
--------
One might be driven to preen others without understanding that
preening is worthwhile because it influences others’ attitudes.
--------
A \textit{theory of mind ability}
\label{df:tom_ability}
is an ability that exists in part because exercising it brings benefits
obtaining which depends on exploiting or influencing facts about others’ mental states.
--------
This is evidence for mindreading = tracking;
it is at best indirectly evidence for mindreading = representing.
--------
‘In saying that an individual has a theory of mind, we mean that the individual [can ascribe] mental states’
\citep[p.\ 515]{premack_does_1978}
--------
The distinction between theory of mind abilities and cognition might help in describing some data. Scrub jays’ complex caching and infants’ imperative and declarative pointing are two behaviours which clearly manifest theory of mind abilities. It is a further question whether they also involve theory of mind cognition.
--------
Eventually, we want to understand,
What explains these abilities?
But first, let’s see some evidence.
--------
--------
Contrast {representing} a mental state with {tracking} one.
For you to \emph{track} someone’s mental state (such as a belief that there is food behind that rock)
is for there to be a process in you which nonaccidentally depends in some way on whether she has that
mental state.
Representing mental states is one way, but not the only way, of tracking them.
In principle it is possible to track mental states without representing them.
For example, it is possible, within limits, to track what another visually represents by representing her line of sight only.
More sophisticated illustrations of how you could in principle track mental states without representing them abound \citep[e.g.][pp.~571ff]{buckner:2014_semantic}.
What many
experiments actually measure is whether certain subjects can track mental states:
the question is whether changes in what another sees, believes or desires are reflected in subjects’ choices of route, caching behaviours, or anticipatory looking (say).
It is surely possible to infer what is represented by observing what is tracked.
But such inferences are never merely logical.
Example: toxicity / smell
--------
The standard debate
What could make others’ mental states intelligible (or identifiable) to a chimpanzee, infant or scrub-jab?
What could make others’ behaviours intelligible (or identifiable) to a chimpanzee, infant or scrub-jab?
--------
--------
Recall this experiment.
Could we explain success without appeal to the idea that chimpanzees
can ascribe knowledge and ignorance?
--------
‘an intelligent chimpanzee could simply use the behavioural abstraction […]: ‘Joe was present and oriented; he will probably go after the food. Mary was not present; she probably won’t.’’
\citep{Povinelli:2003bg}
--------
What’s that?
--------
\begin{quote}
For any food (x) and agent (y), if any of the following do not hold:
(i) the agent (y) was present when the food (x) was placed,
(ii) the agent (y) was oriented to the food (x) when it was placed,
and:
(iii) the agent (y) can go after the food (x)
then probably not:
(iv) the agent (y) will go after the food (x).
Also, if all of (i)–(iii) do hold, then probably (iv).
\end{quote}
--------
Doesn’t work as it stands because in the ‘misinformed’ condition
the dominant was oriented to the food when it was first placed
(just not when it was subsequently moved).
Right way to formulate this is tricky: should we say last placed?
Or should we say when the food was last placed at its current location,
the dominant was oriented to the food there?
--------
Predicting a non-behaviour is particularly challenging.
There are some conditions which could trump this; for example,
if the food is in a place where food is has recently been found, or if the food
is particularly smelly or noisy, or if a trail of grapes
leads to the food ...
So is the hypothesis that chimps ignore all such further factors,
or do we think that this claim should be elaborated?
--------
‘Don't go after food if a dominant who is present has oriented towards it’
\citep[p.~735]{Penn:2007ey}
By shifting from a prediction of the dominant’s behaviour to a
restriction on how the subordinate behaves, a lot of the problems vanish.
But this clearly not an interesting proposal: to see why, we need to think
about scientific methodology (predicting vs retrodicting the order in which
people entered the room).
--------
Maybe it’s not necessary that we actually construct detailed accounts of how each task
is solved using behavioural rules. Following Povinelli and colleagues, Lurz has argued
that the sorts of experiments we have been considering are in principle incapable of
eliminating a hypothesis about behaviour reading ...
‘since mental state attribution in [nonhuman] animals will (if extant) be based on observable
features of other agents’ behaviors and environment ... every mindreading hypothesis has ...
a complementary behavior-reading hypothesis.
Such a hypothesis proposes that the animal relies upon certain behavioral/environmental cues
to predict
[... the behaviour which], on the mindreading hypothesis,
the animal is hypothesized to use as its observable grounds for attributing the
mental state in question.’
\citep[p.~26]{lurz:2011_mindreading}; also \citep[p.~453]{lurz:2011_how}
--------
--------
‘since mental state attribution in [nonhuman] animals will (if extant) be based on observable
features of other agents’ behaviors and environment ... every mindreading hypothesis has ...
a complementary behavior-reading hypothesis.
‘Such a hypothesis proposes that the animal relies upon certain behavioral/environmental cues
to predict another agent’s behavior
[... the behaviour which], on the mindreading hypothesis,
the animal is hypothesized to use as its observable grounds for attributing the
mental state in question.’
\citep[p.~26]{lurz:2011_mindreading}; also \citep[p.~453]{lurz:2011_how}
--------
‘Behavior-reading animals can appeal only to ...
reality-based, mind-independent facts, such as facts about agents’ past behavior or
their current line of gaze to objects in the environment.
‘Mindreading animals, in contrast, can appeal to the subjective ways environmental
objects perceptually appear to agents to predict their behavior.’
\citep[p.~469]{lurz:2011_how}
Experimental implementations: e.g. \citep{karg:2015_goggles}
E.g. seeing an insect or a piece of fruit as an insect or piece of fruit
--------
There’s a competitor, two graps and some size-distorting glass ... can
Subject predict that Competitor will choose not the largest grape but
the one which appears largest?
‘The objective here is to test whether chimpanzees are capable of anticipating (as evidenced by their
looking behavior) the actions of a human competitor by understanding how an illusory stimulus (grape)
looks to the human.’ \citep{lurz:2011_how}
‘the competitor is absent when the grapes are placed inside the containers (as in the pretest stage),
he does not observe the change in apparent size of the grapes as they are placed inside the
containers, as the chimpanzee does. After the grapes are placed, the competitor enters the room, sits
at the table, looks at each grape through the sides of the containers’
\citep[p.~476]{lurz:2011_how}
‘dependent measure is anticipatory looking’.
In this case, the subjects (chimps) know which grape is which and how they appear.
Also a Krachun-like competitive version (in this case, the chimps have to
infer which is larger from the failed reach of the human competitor)...
Also a Krachun-like reaching measure:
‘After the competitor has acted in one of these two ways toward the container with the large grape,
the experimenter places opaque coverings completely over each container and moves the platform toward
the chimpanzee. Once the platform is in place before the chimpanzee, the chimpanzee is allowed to
choose one of the two containers (by pointing to it through a finger hole). Since the containers are
covered, the chimpanzee cannot make its choice by seeing the grape or grapes inside the containers.
Rather, to make its preferred choice (which we assume will be the remaining small grape in
competitor-first success trials and the large grape in competitor-first failure trials) the chimpanzee
must attend to the selection process of the competitor.’
\citep[p.~475]{lurz:2011_how}
--------
The Lurz and Krachun proposal looks really good because it identifies a gap
in the argument for the logical problem ...
--------
Progress ... but does this enable us to solve the logical problem?
--------
E.g. seeing an insect or a piece of fruit as an insect or piece of fruit
--------
Camoflauge ain’t mindreading.
--------
Return to the question; this is for discussion ...
Do goggles solve the ‘Logical Problem’?
‘“self-informed” belief induction variables [... are those] that, if the participant is
capable of mentalizing, he or she knows only through extrapolation from her own experience to be
indicative of what an agent can or cannot see and, therefore, does or does not believe’
\citep[p.~139]{heyes:2014_submentalizing}
--------
Experiment 2: chimp attempts to steal from human,
who will remove food if she sees the chimp approach.
One cover is transparent, the other opaque. But actually when they are down,
both look opaque from the chimp’s point of view.
--------
(This is just a photo of the apparatus, with covers up (top) and down (bottom).)
--------
‘Mean percentage of trials with the choice of the opaque box across the three conditions. In the
transparent condition, one lid was opaque, the other transparent. In the screen condition, one lid
was a screen, the other opaque. The control was like the transparent condition, but without the
experimenter's presence at the time of choice. Error bars indicate 95% CI. The horizontal line
indicates chance level (50%). *P < 0.05; **P < 0.01’
\citep{karg:2015_goggles}.
--------
Return to the question; this is for discussion ...
Do goggles solve the ‘Logical Problem’?
‘“self-informed” belief induction variables [... are those] that, if the participant is
capable of mentalizing, he or she knows only through extrapolation from her own experience to be
indicative of what an agent can or cannot see and, therefore, does or does not believe’
\citep[p.~139]{heyes:2014_submentalizing}
I think: what you ‘project’ from your own experience could be merely knowledge
of affordances (in Bugnyar et al) and appearances (in Karg et al). There’s no special
reason to think that what you project is mental states rather than a better understanding
of objects and their properties.
--------
In conclusion, ...
