Causal Inference Symposium, February 16, 2012
By COGS: the Cognitive Science Organization for Graduate Students at Berkeley
5:30 to 8:00 PM at: 5101 Tolman
Organizers: Mike Pacer and Andrew Critch
Speakers: Andrew Critch, Sarah Wellen, Elizabeth Bonawitz, Anna Waismeyer and Joshua Abbott.
Schedule:
| 5:30-5:40 | Pizza / Weclome |
| 5:40-5:45 | Introduction Mike Pacer, UC Berkeley, Pscyhology |
| 5:45-6:00 | Posing causal inference problems Andrew Critch, UC Berkeley, Mathematics |
| 6:05-6:20 | The Philosophy of Causality: A Conversation with Cognitive Science Sarah Wellen, Carnegie Melon University, Philosophy |
| 6:25-6:40 | Pizza / Discussion |
| 6:40-6:55 | What kids know about causality: Dispositional agency and causal language facilitate toddlers' causal representations Elizabeth Bonawitz, UC Berkeley, Psychology |
| 7:00-7:15 | Causal reasoning in the domestic dog: Will Pavlov's dogs ring the bell? Anna Waismeyer, University of Washington, Institute for Learning & Brain Sciences |
| 7:20-7:35 | Modeling order effects in causal learning Joshua Abbott, UC Berkeley, Psychology |
| 7:40-8:00 | Panel discussion |
Abstracts
- 5:45-6:00 -- Posing Causal Inference Problems
Andrew Critch, UC Berkeley, MathematicsAbstract: I'll quickly introduce the graph-theoretic framework of Bayes Nets, in which essentially any causal reasoning problem can be cast in pure mathematical language. This allows us to say precisely, and in full generality, what we mean by causality and what we mean by inferring it.
- 6:05-6:20 -- The Philosophy of Causality: A Conversation with Cognitive Science
Sarah Wellen, Carnegie Melon University, PhilosophyAbstract: Most of the major historical figures (e.g. Hume, Kant) who theorized about causality were interested both in the metaphysics/epistemology of causality and its psychological basis in human though. This talk will give a brief review of some historical and contemporary philosophical views on causality, with a particular emphasis on the interconnections between philosophical debates and theories of causal learning and reasoning. Topics that will be reviewed include Hume's challenge, counterfactual and process theories of causality, the distinction between type/token causality, and the interpretation of probabilistic views.
- 6:40-6:55 -- What kids know about causality: Dispositional agency and causal language facilitate toddlers' causal representations
Elizabeth Bonawitz, UC Berkeley, PsychologyAbstract: What kinds of events do young children treat as causal? While adults live in a world rife with causal connections, there are substantial constraints on toddlers' ability to infer that predictive relations (events that co-occur, that are associated with each other) might support effective manipulation (acting on 'A' may bring about 'B'). I will present a set of studies demonstrating that (outside of one object "launching" another) toddlers only represent associated events as causal when they involve a person or when causal language is used to describe the event. Follow-up predictive looking paradigms (measuring whether the child is "surprised" and looks longer) suggest that the differences in younger children's responding might not reflect mere failures of performance (e.g., due to the increased complexity of acting vs. looking) but genuine constraints on children's causal representations.
- 7:00-7:15 -- Causal reasoning in the domestic dog: Will Pavlov's dogs ring the bell?
Anna Waismeyer, University of Washington, Institute for Learning & Brain SciencesAbstract: Recently, both philosophers and psychologists have argued for an interventionist account of causal learning that integrates information from observations and interventions and lets us design new interventions (e.g., Gopnik & Schulz, 2007; Woodward, 2003, 2007, Styvers et al 2003). While, there is some recent evidence that domestic dogs will imitate an action on X to produce Y (Miller et al. 2010), these results on their own do not demonstrate behavior consistent with an interventionist account of causal learning. We use a new two-choice procedure to investigate whether domestic dogs use observations of the outcome of natural events without an agent to design their own interventions. Given only observations of covarying events, dogs did not demonstrate causal learning. However, dogs' performance improved when the activation of the objects was the result of their own interventions. This evidence suggests that domestic dogs do not develop causal models of the world based on observations of covarying events. Future research is needed to further examine this question.
- 7:20-7:35 -- Modeling order effects in causal learning
Joshua Abbott, UC Berkeley, PsychologyThe order in which people observe information has an effect on their subsequent judgments and inferences. However, many cognitive models are insensitive to these "order effects." In this talk, I'll review the kinds of order effects found in causal learning tasks and will describe a rational process model (a "particle filter") that captures these effects. This approach provides new insights into why order matters to people.