keynotes

Causal varieties as heuristics

This talk explores causal varieties that present in scientific and everyday life contexts.  The varieties considered include types of causal influence, such as the stability, specificity, reversibility, and speed of causal relationships. This work provides an analysis of these varieties and explores their role as heuristics for causal explanation.  In particular, as most explanatory targets have an enormous set of causally relevant factors, how can attention to causal varieties be used as a heuristic to identify the causes that matter the most?

Inferring what happened through multi-modal mental simulation

Imagine walking into your dining room and noticing one of your favorite vases shattered on the floor. In the corner, you spot a few stray feathers on the ground. It doesn't take long to connect the dots: Whiskers, your cat, went for the bird, bumped into the vase, and gravity and physics did the rest. This seemingly unremarkable sequence of thoughts exhibits the components of an impressive cognitive processing capacity. Having observed something unexpected, you used your intuitive knowledge of how the world works (and how cats work) to imagine a plausible story that explains the data you observed. In this talk, I will shed light on how people can make such impressive inferential leaps. I present a computational model that sequentially simulates plausible hypotheses of what happened by considering both auditory and visual information. The model captures adult participants' responses, reaction times, and eye movements in a challenging physical inference task. We also tested the model with children and found that while 3-5 year-olds primarily relied on a simple heuristic in their inferences, 6-8 year-olds' responses were increasingly consistent with multi-modal simulation. I will conclude by highlighting the critical role that causal abstraction plays in human inference and demonstrate experimentally how we can tell what shape a learner's mental model takes.

Model assisted judgmental bootstrapping: Can incorrect predictions help elicit correct predictions?

In the “out of population prediction” problem, a model has been trained on one population but needs to be applied to a new population for which there are no training data available. How to train a model without training data? In a technique called “judgmental bootstrapping”, the model is trained on human guesses instead of the (missing) ground truth values. Surprisingly, the resulting model can be more accurate than human guesses, despite having been trained on them. Judgmental bootstrapping has been compared to individual expert judgments in domains like education, psychology, management, marketing, and finance, with generally good results.  In this work, we modify the technique of judgmental bootstrapping to see if it can be made even more useful for out of population prediction. In our experiment, participants make forecasts about a new population and models are trained on these forecasts and tested on new cases. In a control condition, participants make forecasts without any assistance. In the condition we call “model assisted judgmental bootstrapping,” participants can see model predictions when they make their forecasts, however, they are predictions for the old population. If people anchor too strongly on these old-population predictions, it may harm their forecasts and the models trained on them. However, if people are able to adjust the old-population predictions appropriately, it may lead to models that do better than human forecasts and models trained with ordinary judgmental bootstrapping.

Causal stories in criminal law: heuristics and justification

By telling stories that offer causal explanations people make sense of (surprising) events. Psychological research has revealed that stories play a central role in the evidential reasoning of legal fact finders (judges, jury members). According to Pennington & Hastie’s story model legal fact finders construct stories to make sense of the evidence. Stories also play a central role in legal epistemology. The scenario theory is a normative and prescriptive theory that builds on the story model. It instructs legal factfinders to construct and evaluate alternative scenarios.

In my talk I will discuss three questions:

1. Which epistemic functions can stories fulfill in legal cases? Can they play a heuristic role? Can they also play a justificatory role?

2. Does the scenario theory offer a rational approach to questions of evidence and legal proof in criminal cases?

3. What is the relation between the scenario theory and Bayesian probabilistic approaches to criminal cases. Are they compatible? If so, are they complementary?

Narratives, Heuristics, and the Currency of Thought

Decisions under risk—where potential outcomes can be listed and probabilities assigned—have received much more attention from decision researchers compared to decisions under (radical) uncertainty—where outcomes cannot be enumerated or their probabilities quantified. In this talk, I develop an approach to decision under uncertainty rooted in causal narratives, heuristics, and affective evaluations of the future (Johnson, Bilovich, & Tuckett, in press, BBS). I propose that narratives are a “currency of thought,” allowing us to explain the past, predict the future, evaluate that future, and motivate sustained action; narratives can also be readily communicated, facilitating social coordination and learning. Although this approach unites a variety of multidisciplinary theoretical perspectives, I also highlight a few of the many open questions it identifies.