The Origins and Functions of Causal Thinking II—Causation, Agency and Intervention
California Institute of Technology, Pasadena :: 11-13
November 2005
Overview : Tentative
schedule : Titles and abstracts
: Accommodation and travel :
Attendance and enquiries
This is the second workshop in the series The
Origins and Functions of Causal Thinking, and will
focus on the topic 'Causation, Agency, and Intervention'.
Issues to be discussed include the following:
• The notion of intervention. Many researchers in philosophy, psychology, and artificial intelligence have taken the notion of intervention to be central to our understanding of causation. But what is an intervention? Is it possible to characterize an intervention in purely objective terms?
• Interventions in scientific domains. Is the same notion of intervention appropriate across scientific domains, such as physics, psychology, and the social sciences?
• Causation and agency. What is the relationship between our understanding of ourseleves as agents in the world, and our understanding of causation more generally?
• Interventions and causal learning. How do our interventions in the world help us to gain an understanding of causal relationships?
• Developmental perspectives. How do infants and children come to understand themselves as beings capable of intervening in the world? How does this interact with other aspects of their causal learning?
• Comparative perspectives. Are humans in some sense unique in their understanding of themselves as agents? Do, e.g., differences in tool use between humans and other animals point to differences in our understanding of causation and agency?
• Simulation. Do what extent is it possible to reproduce causal judgments using expert systems? What heuristics facilitate the functioning of such systems? What can this teach us about our own causal reasoning?
The speakers will include:
Dare Baldwin (Psychology, Oregon)
John Campbell (Philosophy, Berkeley)
Adam Elga (Philosophy, Princeton)
Joe Halpern (Computer Science, Cornell)
Richard Holton (Philosophy, MIT)
Jennan Ismael (Philosophy, Arizona/Centre for Time, Sydney)
David Lagnado (Psychology, U. C. London)
Daniel Povinelli (Psychology, Louisiana)
Huw Price (Centre for Time, Sydney)
Laura Schulz (Psychology, MIT)
The conference will last two and a half days, with twelve
90-minute sessions. Most sessions will comprise a
presentation (c. 45 minutes) followed by plenty of
time for discussion. The talks should be informal in tone,
suitable for an interdisciplinary audience, and should serve
as foci for productive discussion. There will also be round
table discussion sessions.
The conference is being organised by Chris Hitchcock
(cricky@caltech.edu)
and Jim Woodward (jfw@hss.caltech.edu)
of the Division of
Humanities and Social Sciences at Caltech. It is
co-sponsored by the National Science Foundation (grant #
SES-0522637), the McDonnell Foundation, and the Centre for
Time, as well as the Division of Humanities and Social
Sciences.
[top]
Tentative Schedule
[Click on a speaker's name to go to titles and abstracts]
Thursday, November 10 Friday, November 11 Saturday, November 12 Sunday, November 13 All talks will take place in Room 25, in the
basement of Baxter Hall. Breakfast and coffee will
be served outside this room.
6:00 – 7:00: Informal reception, Treasure
Room, Dabney Hall of the Humanities
8:30 – 9:00
Continental Breakfast
9:00 – 9:15
Welcome and Introduction
James Woodward
9:15 – 10:45
Speaker: John Campbell
Chair: Christopher Hitchcock
10:45 – 11:00
Break
11:00 – 12:30
Speaker: Huw Price
Chair: Adam Elga
12:30 – 2:00
Lunch
Dabney Garden
2:00 – 3:30
Speaker: Adam Elga
Chair: Peter Menzies
3:30 – 3:45
Break
3:45 – 5:15
Speaker: Joseph
Halpern
Chair: Ned Hall
5:15 – 5:30
Break
5:30 – 6:30
Octavian Forum Discussion
Introduction: Christopher Hitchcock
Beginning discussants: Christopher Hitchcock, Peter
Menzies, John Campbell, Huw Price,
Joseph Halpern
8:30 – 9:15
Continental Breakfast
9:15 – 10:45
Speaker: Richard
Holton
Chair: Maria Carla Galavotti
10:45 – 11:00
Break
11:00 – 12:30
Speaker: Jenann Ismael
Chair: Mathias Frisch
12:30 – 2:00
Lunch
Dabney Garden
2:00 – 3:30
Speaker: Daniel
Povinelli
Chair: James Woodward
3:30 – 3:45
Break
3:45 – 5:15
Speaker: David Lagnado
Chair: Alison Gopnik
5:15 – 5:30
Break
5:30 – 6:30
Octavian Forum Discussion
Introduction: James Woodward
Beginning discussants: James Woodward, Alison
Gopnik, Richard Holton, Jenann Ismael, Daniel
Povinelli, David Lagnado
Evening: Banquet
8:30 – 9:15
Continental Breakfast
9:15 – 10:45
Speaker: Dare Baldwin
Chair: Arif Ahmed
10:45 – 11:00
Break
11:00 – 12:30
Speaker: Laura Schulz
Chair: Noah Goodman
Baxter Hall is building #77 near the middle of the
campus
map. Dabney Hall is building #70 nearby.
[top]
Titles and Abstracts
What's the Role of Action in an Understanding of Causation?
John Campbell (Berkeley)
In this talk I will be drawing heavily on the the interventionist approach to causation developed by Woodward 2003, Woodward and Hitchcock 2003, and the causal Bayes net tradition within which they are working.
You might read the interventionist analysis as characterizing the kinds of modal facts that a subject exploits in manipulation - facts about what would happen under various interventions. On this approach, a grasp of causal facts has implications for what the upshot will be of the subject's own actions. Similarly, there are implications for what the upshot will be of various 'natural experiments' that might happen, such as the collision of two inanimate objects. But, on this approach, you could in principle grasp what causation is, understood as the interventionist suggests, without having ever noticed that the causal facts have implications for your own actions.
Doesn't agency have a more immediate place in an understanding of causation? Can we really make anything of the idea of a subject who knows what causation is but to whom it has never occurred that causal facts might have implications for his own actions? If the interventionist analysis were a reductive definition, we could say that someone without the notion of cause comes to grasp the idea of cause by grasping the definition. But the interventionist analysis makes heavy use of the notion of cause. So we have to explain in what sense, if any, someone who grasps the notion of cause can be said to grasp the interventionist analysis. I will argue that what happens here is that the subject is provided with an implicit understanding of the notion of an intervention, by tacitly taking it that his own actions are interventions.
This raises the question whether the interventionist
framework, when glossed in this way, leaves open the
possibility of an understanding of one's own actions as
causal. Isn't the fact of one's own action being taken as a
primitive, presupposed by one's understanding of causation,
and incapable itself of being understood in causal terms? I
will argue that this is a mistake, that emphasizing the
importance of the fact of one's own agency in understanding
causation is consistent with giving an analysis of agency in
causal terms. And I will sketch out an account of the kind
of causal structure for agency that matters here.
[return to
timetable]
Growing and pruning causal arrows using boundary conditions
Adam Elga (Princeton)
Systems of fundamental laws sometimes impose constraints on boundary
conditions. For example, laws might rule out certain initial or final
states of the world, or might impose a probability distribution over
such states. Constraints on boundary conditions help to determine
which causal graphs correctly describe a given world. Sometimes,
imposing additional constraints can add causal arrows--as is
illustrated most dramatically in certain supertask pathologies, and in
cases that involve wormholes. Other times, imposing additional
constraints can remove arrows--as when probability distributions over
initial conditions make it appropriate to treat a sensitive system as
if it were isolated. Constraints on boundary conditions therefore are
an important part of the explanation of how simple macroscopic causal
structure arises from intricate microscopic fundamental dynamics.
[return to
timetable]
Causality, Responsibility, and Blame: A Structural-Model
Approach
Joseph
Halpern (Cornell)
I first review the basic definition of causality introduced
by Halpern and Pearl. This definition (like most in the
literature) treats causality as an all-or-nothing concept;
either A is a cause of B or it is not. We show how it can be
extended to take into account the degree of responsibility
of A for B. For example, if someone wins an election 11–0,
then each person who votes for him is less responsible for
the victory than if he had won 6–5. I then define a
notion of degree of blame, which takes into account an
agent's epistemic state. Roughly speaking, the degree of
blame of A for B is the expected degree of responsibility of
A for B, taken over the epistemic state of an agent. I also
briefly discuss the extent to which definitions reflect how
people use notions like cause, blame, and
responsibility in practice.
The first part of the talk represents joint work with Judea
Pearl; the latter half represents joint work with Hana
Chockler.
[return to
timetable]
Agency and Choice
Richard
Holton (MIT)
The person who suffers from anarchic hand syndrome loses
agency in one hand, but typically keeps it in the other. So
how do the two hands differ? Not, I suggest, because the
agent chooses what to do with one and not with the other.
Typically the well-behaved hand works without need of
choice. I use the case as a springboard for an account of
choice, and of why it might be important for cognitively
limited creatures like ourselves, and then return to ask
what choice has to do with agency. Perhaps the anarchic hand
is experienced as anarchic not because it acts in the
absence of choice, but because the kinds of things it does
are the kinds of things that would normally not be done
without choice. It thus presents a experience of the loss of
choice's efficacy. And that prompts some speculations about
the role of choice in the experience of causation. Could it
be that the most fundamental manipulation we experience is
that of manipulating our actions by our choices?
[return to
timetable]
A Noncausal Account of the Phenomenology of
Agency
Jenann
Ismael (University of Arizona/University of Sydney)
"The problem of action is to explicate the contrast between what an agent does and what merely happens to him, or between the bodily movements that he makes and those that occur without his making them." (Frankfurt, 'The Problem of Action'‚ in The Importance of What We Care About, p. 69.)
The most obvious and popular ways of 'solving' the problem of action (i.e., explicating the subjective difference -- the difference from the agent's perspective -- between what an agent does and what merely happens to him) look to causal differences. Frankfurt, for example, writes:
according to causal theories of the nature of action, which currently represent the most widely followed approach of [the] contrast, the essential difference is to be found in their prior causal histories - different versions of the causal approach may provide differing accounts of the sorts of events or states which must figure causally in the production of actions.
Anyone who wants to reduce the asymmetry of causation to
our experience of ourselves as agents in the world has to
solve the problem in a way that doesn't use the notion of
causation. I'm going to take a crack at doing that by taking
a clue from Anscombe that nobody agrees with and then
adapting it in a way that would have appalled her.
[return to
timetable]
Beyond Covariation: Cues for Learning Causal
Structure
David
Lagnado (University College, London)
How do people learn causal structure? Previous psychological
models have focused on covariation-based learning, but this
fails to capture the range of cues used to infer structure.
In a series of experiments we investigated the interplay
between temporal order, intervention and covariation. The
main findings were that (1) participants allowed temporal
order to override covariation information, leading to
spurious causal inferences when temporal cues were
misleading; (2) intervention and temporal order combined
additively to yield accurate causal inference, well beyond
that achievable through covariational data alone; (3)
participants were able to use double interventions to
disambiguate complex causal models. Together these studies
show that people use both intervention and temporal order
cues to infer causal structure, and that these cues dominate
the available covariational information. We endorse a
hypothesis-driven account of learning, whereby people use
cues such as temporal order to generate initial models, and
then test these models against the incoming covariational
data.
[return to
timetable]
Causal Reasoning in Chimpanzees: Lessons from Their
Conception of Weight
Daniel
Povinelli (Lousiana)
In a series on two dozen experiments, we assessed whether
adult chimpanzees possess a folk notion of weight that is
abstract in the sense that it is an object-based property,
independent of their immediate sensory experiences. The
results of experiments which required the apes to (a) sort
objects based on a dichotomous heavy/light distinction, (b)
infer the kind of effort it would take to displace objects
of radically different weights, (c) understand the
differential effects that heavy and light objects have in
simple collision events, revealed a virtual absence of
weight as a fully independent object property. In contrast,
the apes excelled at problems involving objects stability
and support when an explicit contrast of heavy and light was
not required. I suggest that these results are consistent
with the broader hypothesis that reasoning about genuinely
unobservable entities and processes (in both the domains of
physical objects and social behavior) is a human
specialization. This hypothesis argues that whereas both
chimpanzees and human excel at extracting statistical
regularities from directly perceivable events and objects,
only humans seek to explain those regularities on the basis
of causes.
[return to
timetable]
What Would God Know About Causation?
Huw Price
(University of Sydney)
Remember the shock (or thrill) of realising for the first
time that some people see you as a foreigner? Unless
you thought that they were just wrong, what you
discovered at that moment was that the local/foreigner
distinction is 'perspectival'. How it is applied depends on
where one stands; two different speakers may draw it
differently, without either being mistaken; and there's no
right way to apply it, from a perspective-free or
'god's-eye' point of view.
We humans have made many such discoveries, individually and
collectively. Some of them were major intellectual
achievements. Some of the lessons are still sinking in.
Others, perhaps, have yet to come over the horizon. And we
always learn something important, both about the world and
about ourselves, when we uncover a new case.
I'm interested in the possibility that causation is
perspectival in this way, and that the centrality of the
notion of intervention is a key reason for thinking
that it must be perspectival. In this talk I'll try to do
three things: (i) clarify the general notion of what it is
to be perspectival, in the relevant sense; (ii) sketch some
arguments for thinking that causation is perspectival; (iii)
discuss the kinds of things we learn about ourselves and
about the world, in this case, if we accept that causation
is perspectival.
[top]
Constructivism Revisited: From Action to Inference and Back Again
Laura Schulz (MIT)
The idea that children actively "construct" knowledge by experimenting in the world and revising their theories with evidence has had a profound influence on the field of cognitive development since Piaget. However, although researchers have suggested that children are "little scientists", children are poor at designing controlled experiments and little is known about how children's spontaneous behavior might generate evidence to support causal learning. In this talk I will suggest ways in which prior knowledge, patterns of evidence, and assumptions about intervention might affect both children's causal inferences and their spontaneous actions in the world. First, I will briefly discuss some preliminary data suggesting that even 12-month-old infants may treat human actions as "interventions" on a causal system. Second, I will suggest that children's spontaneous actions are affected by the quality of the evidence they observe. When children observe stochastic or ambiguous evidence, they spontaneously generate more variation and exploration in their actions than when they observe deterministic and unambiguous evidence. Finally, I will suggest that both the number of different actions children generate and the number of times any given action is repeated, decline with age. Thus younger children may generate more evidence about novel causal relations than older children and adults.
[top]
Accommodation and travel
Please click here or see this PDF file for further information.
[top]
Attendance and enquiries
This is an invitation-only workshop. However, a very limited number of additional participants may be accepted, at the discretion of the organisers. If you would like to be considered, please contact Chris Hitchcock (cricky@caltech.edu) or Jim Woodward (jfw@hss.caltech.edu) before October 20, 2005.
[top]
Last updated: 1 November 2005