Causal Thinking and the Embedded Perspective Jenann Ismael
Causal Thinking and the Embedded Perspective Jenann Ismael Centre for Time University of Sydney Jan. 9, 2009
Bertrand Russell n “The law of causality, I believe, like much that passes muster among philosophers, is a relic of a bygone age, surviving, like the monarchy, only because it is erroneously supposed to do no harm. ”
Cartwright n n Russell is correct that causal information is not reducible to dynamical information But specifically causal information is needed for choice, because specifically causal information is needed to distinguish effective from ineffective strategies.
Causes vs. Correlations n n There is a high positive correlation between smoking and lung cancer There is also a high positive correlation between having bad breath and lung cancer (because there’s a high correlation between smoking and having bad breath) Stopping smoking is an effective strategy for preventing lung cancer, but Taking medicine to prevent bad breath is not
Interventionism Information about the causal effects of A is information about what would happen if we were able to, sever A from its own past causes, and treat it as a free variable, and look at its future effects n This is put by saying that causal information is information about the results of hypothetical interventions n
Pearl n “The scientist carves a piece from the universe and proclaims that piece in…The rest of the universe is then considered out…This choice of ins and outs [or ‘endogenous’ and ‘exogenous’ variables] creates an asymmetry in the way that we look at things and it is this asymmetry that allows us to talk about ‘outside interventions’ and hence about causality”
n n “But if you include the entire universe in the model, causality disappears because interventions disappear - the manipulator and the manipulated lose their distinction” (Causality, p. xiv)
The Embedded Perspective n This is the perspective of an agent n n n Representing a world whose history is partly constituted by her activity, As that history unfolds In a world like ours, the embedded perspective is characterized by the further features that such an agent n n While in possession of information about its past, And deliberating about her future contributions to that history
Deliberation n n Since the choice between available alternatives is to be decided as the outcome of the deliberative process, the agent’s action is treated as a free variable in the course of deliberation Actions are represented in the model in hypothetical form, as potential outcomes of deliberation
The deliberative process then proceeds by tracing forward causal implications of potential decisions. n So the causal content of a model - what it says about what would happen under potential actions - plays an ineliminable role generating the decision n
In sum n n n The distinction between endogenous and exogenous variables has natural application in decision contexts in which an agent is deciding how to act. Her own actions are separated from external causes by the deliberative process and are correctly viewed in that context as free. But when these models are embedded in larger encompassing model of the universe as a whole, the variables no longer appear as free and the causal facts disappear
Modeling open subsystems n n n The models we apply in practice are always models of open subsystems of the universe Whenever we’re modeling some open subsystem of the universe, the scope of the model defines a division between endogenous and exogenous variables It is only when we retreat to the cosmological perspective that treats the universe as a closed system that the distinction between endogenous and exogenous variables disappears
n Other examples where structure of practical importance disappears when we retreat to an unembedded perspective n n The red dot on a map The distinction between nearby and far away The distinction between now and later The distinction between foreigners ad locals
A pragmatic view of causal relations We get causal relationships by supplementing laws with background information that supports robust connections between localizable events n The insight of interventionism and manipulability theories of causation is that the point of causal structure is to identify useful entry points for intervention n
Models as ppsfep’s n n We can generalize the insight to other cases by thinking of physical models as partially prepared solutions to frequently encountered problems Think of physical theories as a toolkit for constructing models and models as constructed on the fly, shaped by the constraints of the problem and tailored to the context of application
Inverting the order of priority n n Instead of focusing on the abstract models of a closed universe provided by textbook physics, we should be paying attention to the richer models of open subsystems that are the bread and butter of practical physics These are full of interesting physical structure that gets lost when we retreat to the cosmological perspective
- Slides: 16
