# Keyboard Shortcuts?

×
• 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 (or swipe left)

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

\title {Social Cognition \\ Lecture 08}

\maketitle

# Social Cognition

\def \ititle {Lecture 08}
\def \isubtitle {Social Cognition}
\begin{center}
{\Large
\textbf{\ititle}: \isubtitle
}

\iemail %
\end{center}

What models of minds and actions, and of behaviours,

and what kinds of processes,

underpin mental state tracking in different animals?

Way forward:

1. Construct a theory of behaviour reading

2. Construct a theory of mindreading

Back to the theory of behaviour reading ...

## Ingredients for a Theory of Behaviour Reading

\section{Ingredients for a Theory of Behaviour Reading}

\section{Ingredients for a Theory of Behaviour Reading}

What could infants, chimps and scrub-jays reason about, or represent, that enables them, within limits, to track others’ perceptions, knowledge, beliefs and other propositional attitudes?

The standard question:

Do nonhuman animals represent mental states or only behaviours?

The ‘Logical Problem’ ...

Alternative question:

How might nonhuman animals model others’ behaviours, and how might they predict them?

‘a better conception of ‘not mindreading’ would be more disparate and less dependent on common sense than the current conception of behaviour reading’

\citep[p.~322]{heyes:2014_animal}

Heyes (2015, p. 322)

identifying mental states

Because ascriptions of mental states are based on observed behaviours.

identifying effects of actions

predicting when an event of interest will occur

learning how to do things

Our primary concern here with behaviour reading is as a potential basis for abilities to track others’ mental states without representing them. But behaviour reading is plausibly important in other ways. In mindreaders, behaviour reading is thought to be useful or even necessary for identifying intentions and other mental states (\citealp[p.~861]{Newtson:1977dw}; \citealp[p.~708]{Baldwin:2001rn}). Behaviour reading may also matter for efficiently representing events \citep{Kurby:2008bk}, identifing the likely effects of actions \citep{Byrne:1999jk}, predicting when an event likely to be of interest will occur \citep[p.~121]{Swallow:2008cf}, and learning through observation how to do things \citep{Byrne:2003wx}. And of course a special case of pure behaviour reading, ‘speech perception’, underpins communication by language in humans.
Start with a case in which behaviour reading is clearly involved. I take Byrne’s study to demonstrate that chimps are capable of sophisticated behaviour reading. But how might they represent behaviours?
The procedure for preparing a nettle to eat while avoiding contact with its stings is shown in \vref{fig:byrne_2003_fig1}. It involves multiple steps. Some steps may be repeated varying numbers of times, and not all steps occur in every case. The fact that gorillas can learn this and other procedures for acquiring and preparing food by observing others’ behaviour suggests that they have sophisticated behaviour reading abilities \citep[p.~513]{Byrne:2003wx}. If we understood the nature of these behaviour reading abilities and their limits, we might be better able to understand their abilities to track mental states too.

Byrne 2003, figure 1

‘great apes [are] able to acquire complex and elaborate local traditions of food acquisition, some of them involving tool use’ \citep[p~513]{Byrne:2003wx}
So even quite sophisticated behaviour reading is possible without any reliance on communication by language. We can therefore think of behaviour reading as foundational for any kind of radical interpretation.
[background] ‘Nettles, Laportea alatipes, are an important food of mountain gorillas in Rwanda (Watts 1984), rich in protein and low in secondary compounds and structural carbohydrate (Waterman et al. 1983). Unfortunately for the gorillas, this plant is 'defended' by powerful stinging hairs, especially dense on the stem, petioles and leaf-edges. All gorillas in the local population process nettles in broadly the same way, a technique that minimizes contact of sting- ing hairs with their hands and lips (Byrne & Byrne 1991; figure 1). A series of small transformations is made to plant material: stripping leaves off stems, accumulating larger bundles of leaves, detachment of petioles, picking out unwanted debris, and finally folding a package of leaf blades within a single leaf before ingestion. The means by which each small change is made are idiosyncratic and variable with context (Byrne & Byrne 1993), thus presum- ably best learned by individual experience. However, the overall sequence of five discrete stages in the process is standardized and appears to be essential for efficiency (Byrne et al. 2001a).’ \citep[pp.~531--2]{Byrne:2003wx}
‘Like other complex feeding tasks in great apes, preparing nettles is a hierarchically organized skill, showing considerable flexibility: stages that are occasionally unnecessary are omitted, and sections of the process (of one or several ordered stages) are often repeated iteratively to a criterion apparently based on an adequate size of food bundle (Byrne & Russon 1998).’ \citep[pp.~532]{Byrne:2003wx}

already considered : The Teleological Stance

An account of pure goal ascription is an account of how you could in principle infer facts about the goals to which actions are directed from facts about joint displacements, bodily configurations and their effects (e.g. sounds). Such an account is a computational theory of pure goal ascription.

pure goal ascription

Infer The Goals from The Evidence

The Goals: facts which goals particular actions are directed to...

The Evidence: facts about events and states of affairs that could be known without knowing which goals any particular actions are directed to, nor any facts about particular mental states ...

‘an action can be explained by a goal state if, and only if, it is seen as the most justifiable action towards that goal state that is available within the constraints of reality’

\citep[p.~255]{Csibra:1998cx}

Csibra & Gergely (1998, 255)

1. action a is directed to some goal;

2. actions of a’s type are normally means of realising outcomes of G’s type;

3. no available alternative action is a significantly better* means of realising outcome G;

4. the occurrence of outcome G is desirable;

5. there is no other outcome, G′, the occurrence of which would be at least comparably desirable and where (2) and (3) both hold of G′ and a

Therefore:

6. G is a goal to which action a is directed.

OK, I think this is reasonably true to the quote. So we’ve understood the claim. But is it true?

Byrne 2003, figure 1

Recall Byrne’s study. To understand this action, the Teleological Stance is insufficient. Why?

‘Like other complex feeding tasks in great apes, preparing nettles is a hierarchically organized skill, showing considerable flexibility:

stages that are occasionally unnecessary are omitted,

and sections of the process (of one or several ordered stages) are often repeated iteratively to a criterion apparently based on an adequate size of food bundle’

(Byrne, 2003 p. 532)

First, because understanding it involves identifying a hierarchical structure of goals. That is, assigning goals to parts of the action and relating these to the goals of larger units of action.
Second, because understanding this action requires segmentation: identifying where one action begins and another ends.
So the Teleological Stance is just part of a theory of behaviour reading.

segmentation: Where does an action begin an end?

Behaviour reading involves at least three components: segmentation, categorisation and structure extraction. First, it is necessary to segment continuous streams of bodily movements into units of interest. Adult humans can readily impose boundaries on continuous sequences of behaviour \citep{Newtson:1976ni,Zacks:2001ua,Kurby:2008bk}, as can infants \citep{Baldwin:2001rs}. Segmentation is probably complicated by the fact that units are not performed one after another but can overlap, as when you bat away a fly while squeezing a lemon and talking on the phone. A further complication is that identifying unit boundaries probably involves taking into account the context in which a behaviour occurs \citep[p.~847]{Newtson:1977dw}. How could segmentation be achieved? Commencement and completion of a goal or subgoal typically coincide with dramatic changes in physical features of the movements such as velocity \citep{Zacks:2001vo}. Baldwin and Baird express this idea graphically with the notion of a ‘ballistic trajectory’ that provides an ‘envelope’ for a unit of action \citep[p.~174]{Baldwin:2001rs}. Research using schematic animations has shown that adults can use a variety of movement features to group behavioural chunks into units \citep{Zacks:2004vd,Hard:2006gr}.

categorisation: What is the type of that action?

A second component of behaviour reading is categorisation. Adult humans spontaneously label units of behaviour as ‘running’, ‘grasping’, or ‘searching’ (say). This is categorisation: two units which may involve quite different bodily configurations and joint displacements and which may occur in quite different contexts are nevertheless treated as equivalent. How are categories identified in pure behaviour reading? One possibility is that some categorisation processes involve mirroring motor cognition. When a monkey or a human observes another’s action, there are often motor representations in her that would normally occur if it were her, the observer, who was performing the action (see \citealp[][]{rizzolatti_functional_2010,rizzolatti_mirrors_2008} for reviews). Further, in preparing, performing and monitoring actions, units of action are represented motorically in ways that abstract from particular patterns of joint displacements and bodily configurations \citep[e.g.][]{rizzolatti:1988_functional, Rizzolatti:2001ug, hamilton:2008_action, cattaneo:2010_state-dependent,Umilta:2001zr, villiger:2010_activity, koch:2010_resonance}. And, in observation as performance, what is represented motorically depends on contextual factors such as the presence of particular objects on which certain actions might be performed (\citealp{Gallese:2011uq}). Taken together, these findings indicate that %there may be a deep connection between processes involved in action performance and one process of categorisation in behaviour reading. %Perhaps one process by which units of action are categorised is the process by which, in other contexts, your own actions are prepared.

structure extraction: Are these actions related to those?

A third component of behaviour reading is structure extraction. Many actions can be analysed as a structure of goals hierarchically ordered by the means-ends relation A behaviour reader should be able to extract some or all of this structure. But how? Units of behaviour that are all involved in bringing about a single outcome are more likely to occur in succession than chunks not so related. This suggests that transitional probabilities in the sequence of units could in principle be used to identify larger structures of units, much as phonemes can be grouped into words by means of tracking transitional probabilities \citep{Saffran:1996aj,Gomez:2000jr}. We know that human adults can learn to group small chunks of behaviour into larger word-like units on the basis of statistical features alone \citep{Baldwin:2008mw}. A statistical learning mechanism required for discerning such units is automatic \citep{Fiser:2001cx}, domain-general \citep{Kirkham:2002cj} and probably present in human infants \citep{saffran:2007_dog} as well as other species including monkeys \citep{Hauser:2001oo}, pigeons \citep{herbranson:_artificial}, songbirds \citep{abe:2011_songbirds} and rats \citep{Toro:2005ma,murphy:2008_rule}. It is therefore plausible that at least some animals’ pure behaviour reading abilities enable them to extract some of the hierarchical structure of actions.

goal ascription: What is the goal of those actions?

Here I think the teleological stance (which may involve motor processes) is helpful.
Let me recap the four ingredients with a diagram ....

What are the limits of pure behaviour reading? While little is currently known about pure behaviour reading outside of communication by language, it is perhaps reasonable to assume that structure extraction depends on domain-general learning mechanisms. After all, such mechanisms appear sufficient and there is currently little evidence for domain-specific mechanisms. %(even in adult humans, the most abundant lab animals) This assumption allows us to make conjectures about the limits of pure behaviour reading.
One limit concerns the kinds of structure which animals of a given species might extract. How complex are the patterns of behaviour which can be extracted?

There is a non-adjacent dependency in my behaviour when, for example, my now having a line of sight to an object that is currently unobtainable because of a competitor’s presence results in me retrieving the object at some arbitrary later time when the competitor is absent. In this case, my retrieving the object depends on my having had it in my line of sight, but there is an arbitrary interval between these events. The hypothesis is that structures involving non-adjacent dependencies are relatively difficult to learn and identify, and that difficulty increases as the number of non-adjacent dependencies increases.% \footnote{% Compare \citet[][]{vries:2012_processing}. Of course, whether non-adjacent dependencies are intrinsically difficult depends on the cognitive architecture \citep{udden:2012_implicit}. There is evidence that monkeys \citep{ravignani:2013_action} and chimpanzees \citep{sonnweber:2015_non} can learn patterns involving one non-adjacent dependency. }

The Birdsong Limit?

To generalise, suppose we assume that extracting structure in pure behaviour reading involves domain-general learning mechanisms only. Then whenever an animal’s abilities to track others’ mental states are underpinned by pure behaviour reading only, her abilities to learn and identify patterns should be subject to domain-general limits. Since birdsongs are discriminable and involve diverse behavioural structures \citep{berwick:2011_songs}, we might take the \emph{Birdsong Limit} as a rough working hypothesis: structures not found in birdsong cannot be extracted in pure behaviour reading.
‘The current study tested the hypothesis that a non-human primate species could detect abstract, non-adjacent dependencies in acoustic stimuli, even when dependencies occurred over an arbitrary variable number of intervening sounds ... Squirrel monkeys consistently recognized and generalized the pattern ABnA at different levels, showing sensitivity to arbitrary-distance dependencies’ (\citealp{ravignani:2013_action}; see also\citealp{sonnweber:2015_non}).

Hare et al (2001, figure 1)

Although not designed to test such limits, some existing experimental designs involve features which plausibly exclude explaining subjects’ performance in terms of pure behaviour reading only. To illustrate, consider the sequence of events in the ‘misinformed’ condition of \citet[][Experiment 1]{Hare:2001ph}. A competitor observes food being placed [$A$], the competitor’s access is blocked [$B$], stuff happens [$X^N$], food is moved [$C$], more stuff happens [$Y^N$], and the competitor’s access is restored [$D$]. Finding evidence that chimpanzees can learn to identify patterns of this form [$ABX^NCY^ND$] and use them to predict the conspecifics’ behaviours would represent a major discovery.
‘Before the subordinate’s door was raised, thus allowing her to enter the middle cage, we closed the dominant’s door, regard- less of whether it had been open before, and it remained closed until the subordinate’s hand first touched the floor of the middle cage.’ Hare et al (2001, p. 142)

A competitor observes food being placed

the competitor’s access is blocked

stuff happens

food is moved

more stuff happens

the competitor’s access is restored

challenge

If nonhumans aren’t mindreading, explain how they else could be tracking mental states in the ways we know they do by (a) constructing a better theory of behaviour reading, (b) generating predictions from it, (c) and confirming those predictions.
Earlier I mentioned this way of setting up the debate. And I argued that this is not a productive way to shape the debate.

What do infants, chimps and scrub-jays reason about, or represent, that enables them, within limits, to track others’ perceptions, knowledge, beliefs and other propositional attitudes?

The standard debate

-- Is it mental states?

-- Or only behaviours?

What models of minds and actions, and of behaviours,

and what kinds of processes,

underpin mental state tracking in different animals?

Current theories of behaviour reading do not support tracking mental states in any of the more sophisticated experimental settings used to test for mindreading abilities.
And given the possibility of having a minimal theory of mind, there is no need for them to do so.

nonhuman mindreading: all the evidence points to its existence

Humans are not alone (or unique) in being able to represent mental states.

The domain: what is a theory of social cognition a theory of?

Social cognition:

cognition of
others’ actions and mental states
in relation to social functioning.

Our goal for this course: construct a theory of social cognition. But what questions does such a theory aim to answer.

How in principle could someone infer facts about actions and mental states from non-mental evidence?

What is the relation between an account of radical interpretation* and a theory of social cognition?

A theory of radical interpretation* is supposed to provide a computational description of social cognition.

I’ve told you what this is, but I’ll remind you in a moment.

Infer The Mind from The Evidence

The Mind: facts about actions, desires, beliefs, emotions, perspectives ...

The Evidence: facts about events and states of affairs that could be known without knowing what any particular individual believes, desires, intends, ...

The Intentional Stance (Dennett)

Davidson’s Theory

The Teleological Stance & Your-Goal-Is-My-Goal

Minimal Theory of Mind

also an implicit theory associated with perception of emotion

‘the intentional stance ...

‘first you decide to treat the object whose behavior is to be predicted as a rational agent;

‘then you figure out what beliefs that agent ought to have , given its place in the world and its purpose.

‘Then you figure out what desires it ought to have, on the same considerations,

‘and finally you predict that this rational agent will act to further its goals in the light of its beliefs’

Dennett (1987, 17)

\citet[p.~22ff]{Marr:1982kx} distinguishes:
\begin{itemize}
\item computational description---What is the thing for and how does it achieve this?
\item representations and algorithms---How are the inputs and outputs represented, and how is the transformation accomplished?
\item hardware implementation---How are the representations and algorithms physically realised?
\end{itemize}
One possibility is to appeal to David Marr’s famous three-fold distinction bweteen levels of description of a system: the computational theory, the representations and algorithm, and the hardware implementation.
This is easy to understand in simple cases. To illustrate, consider a GPS locator. It receives information from four satellites and tells you where on Earth the device is.
There are three ways in which we can characterise this device.

1. computational description

First, we can explain how in theory it is possible to infer the device’s location from it receives from satellites. This involves a bit of maths: given time signals from four different satellites, you can work out what time it is and how far you are away from each of the satellites. Then, if you know where the satellites are and what shape the Earth is, you can work out where on Earth you are.

-- What is the thing for and how does it achieve this?

The computational description tells us what the GPS locator does and what it is for. It also establishes the theoretical possibility of a GPS locator.
But merely having the computational description does not enable you to build a GPS locator, nor to understand how a particular GPS locator works. For that you also need to identify representations and alogrithms ...

2. representations and algorithms

At the level of representations and algorthms we specify how the GPS receiver represents the information it receives from the satellites (for example, it might in principle be a number, a vector or a time). We also specify the algorithm the device uses to compute the time and its location. The algorithm will be different from the computational theory: it is a procedure for discovering time and location. The algorithm may involve all kinds of shortcuts and approximations. And, unlike the computational theory, constraints on time, memory and other limited resources will be evident.
So an account of the representations and algorithms tells us ...

-- How are the inputs and outputs represented, and how is the transformation accomplished?

3. hardware implementation

The final thing we need to understand the GPS locator is a description of the hardware in which the algorithm is implemented. It’s only here that we discover whether the device is narrowly mechanical device, using cogs, say, or an electronic device, or some new kind of biological entity.

-- How are the representations and algorithms physically realised?

The hardware implementation tells us how the representations and algorithms are represented physically.

Marr (1992, 22ff)

How is this relevant to my question? My question was, What is the relation between an account of radical interpretation* and a theory of social cognition?
I suggest that an account of radical interpretation* is supposed to provide a computational description of social cognition; it tells us what social cognition is for and how, in the most abstract sense, it is possible.

The Intentional Stance
can be (mis)interpreted as an attempt to provide
a computational description of social cognition.

‘the intentional stance ...

‘first you decide to treat the object whose behavior is to be predicted as a rational agent;

‘then you figure out what beliefs that agent ought to have , given its place in the world and its purpose.

‘Then you figure out what desires it ought to have, on the same considerations,

‘and finally you predict that this rational agent will act to further its goals in the light of its beliefs’

Dennett (1987, 17)

1. Humans can distinguish each others desires in ways unrelated to their purpose and place in the world.

BUT

2. The Intentional Stance provides no way to do this.

3. The Intentional Stance does not provide a correct computational description of human social cognition.

Objection 1

The Intentional Stance provides
no way to identify
false beliefs, ‘incorrect’ desires or failures of rationality.

Objection 2

The Intentional Stance provides
distinguish me from you.

1. Humans can sometimes identify false beliefs, ‘incorrect’ desires or failures of rationality.

BUT

2. The Intentional Stance provides no way to do this.

3. The Intentional Stance does not provide a correct computational description of human social cognition.

Objection 1

The Intentional Stance provides
no way to identify
false beliefs, ‘incorrect’ desires or failures of rationality.

Objection 2

The Intentional Stance provides
distinguish me from you.

The Intentional Stance (Dennett)

Davidson’s Theory

The Teleological Stance & Your-Goal-Is-My-Goal

Minimal Theory of Mind

also an implicit theory associated with perception of emotion

‘an action can be explained by a goal state if, and only if, it is seen as the most justifiable action towards that goal state that is available within the constraints of reality’

\citep[p.~255]{Csibra:1998cx}

Csibra & Gergely (1998, 255)

1. action a is directed to some goal;

2. actions of a’s type are normally means of realising outcomes of G’s type;

3. no available alternative action is a significantly better* means of realising outcome G;

4. the occurrence of outcome G is desirable;

5. there is no other outcome, G′, the occurrence of which would be at least comparably desirable and where (2) and (3) both hold of G′ and a

Therefore:

6. G is a goal to which action a is directed.

Objection 1

The Intentional Stance provides
no way to identify
false beliefs, ‘incorrect’ desires or failures of rationality.

Objection 2

The Intentional Stance provides
distinguish me from you.

The Teleological Stance does not face objection 2 (no way to distinguish you from me), at least not at first pass.
What about objection 1? This is tricky. By construction the Teleological Stance provides no way to identify false beliefs, ‘incorrect’ desires or failures of rationality.
But is this an objection ... ?

1. Humans can sometimes identify false beliefs, ‘incorrect’ desires or failures of rationality.

BUT

2. The Teleological Stance provides no way to do this.

3. The Teleological Stance does not provide a correct computational description of human social cognition.

The Teleological Stance isn’t trying to be a comprehensive theory of human social cognition. The idea, rather, is that human social cognition has many parts and one of them is a process of pure goal ascription and the Teleological Stance provides a computational description for that.

Dennett, Davidson:

We need a single theory covering all social cognition.

A better approach:

Social cognition involves a cluster of disparate abilities.

These include pure goal ascription.

There is no such thing as a theory of social cognition.

Instead we need a theory for each disparate ability.

The Intentional Stance (Dennett)

Davidson’s Theory

The Teleological Stance & Your-Goal-Is-My-Goal

Minimal Theory of Mind

also an implicit theory associated with perception of emotion

What do we perceptually experience of others’ mental states?

Evidence:

Humans have categorical perception of expressions of emotion.

Question:

Are expressions of emotion facial configurations?

Observation:

Facial configurations are not diagnostic of emotions (Aviezer et al)

Theory:

The objects of categorical perception are actions directed to the goals of expressing particular emotions (Butterfill, 2015).

Categorical perception of expressions of emotions

 level specification computational description The Teleological Stance representations and algorithms ... are broadly perceptual

The Intentional Stance (Dennett)

Davidson’s Theory

The Teleological Stance & Your-Goal-Is-My-Goal

Minimal Theory of Mind

also an implicit theory associated with perception of emotion

Btw, mTm is a description of a model of mind at the level of a computational theory; it is completely agnostic about representations and algorithms.
MTM overcomes both objections ... at a cost (it’s limited)

Objection 1

The Intentional Stance provides
no way to identify
false beliefs, ‘incorrect’ desires or failures of rationality.

Objection 2

The Intentional Stance provides
distinguish me from you.

1. Humans can sometimes identify false beliefs involving mistakes about numerical identity.

BUT

2. Minimal Theory of Mind provides no way to do this.

3. Minimal Theory of Mind does not provide a correct computational description of human social cognition.

dogma

the first

The dogma of mindreading: any individual has at most one model of minds and actions at any one point in time.
The first dogma of mindreading: there is just one mindreading process ... so we need just one true theory of radical interpretation
dual-process recap
So what does a Dual Process Theory of Mindreading claim? The core claim is just this:

Dual Process Theory of Mindreading (core part)

Two (or more) mindreading processes are distinct:
the conditions which influence whether they occur,
and which outputs they generate,
do not completely overlap.

\textbf{You might say, this is a schematic claim, one totally lacking substance.} You’d be right: and that’s exactly the point.
A key feature of this Dual Process Theory of Mindreading is its \textbf{theoretical modesty}: it involves no a priori committments concerning the particular characteristics of the processes.

dogma

the first

The dogma of mindreading: any individual has at most one model of minds and actions at any one point in time.
The first dogma of mindreading: there is just one mindreading process ... so we need just one true theory of radical interpretation

The Intentional Stance (Dennett)

Davidson’s Theory

The Teleological Stance & Your-Goal-Is-My-Goal

Minimal Theory of Mind

also an implicit theory associated with perception of emotion

The domain: what is a theory of social cognition a theory of?

Social cognition:

cognition of
others’ actions and mental states
in relation to social functioning.

Our goal for this course: construct a theory of social cognition. But what questions does such a theory aim to answer.

How in principle could someone infer facts about actions and mental states from non-mental evidence?

What is the relation between an account of radical interpretation* and a theory of social cognition?

A theory of radical interpretation* is supposed to provide a computational description of social cognition.

The domain: what is a theory of social cognition a theory of?

Social cognition:

cognition of
others’ actions and mental states
in relation to social functioning.

Our goal for this course: construct a theory of social cognition. But what questions does such a theory aim to answer.