Implicit and
explicit Bayesian inference
Description
A major current revolution in cognitive science concerns the rapid
ascendance of Bayesian modeling of probabilistic (or uncertain) reasoning (Chater,
Tenenbaum, & Yuille, 2006). In the last couple
of years, Bayesian modeling has surpassed both symbolic and connectionist methods
in frequency of conference presentations and publications. An entire special
issue of Trends in Cognitive Sciences
(July 2006) was devoted to Probabilistic Models of Cognition. Bayesian modeling
is being successfully applied to a wide range of problems in cognitive science
including sensorimotor control (Körding & Wolpert, 2006), vision (Yuille & Kersten, 2006), conditioning (Courville, Daw, & Touretzky, 2006), induction and
inference (Tenenbaum,
Griffiths, & Kemp, 2006), and language (Chater
& Manning, 2006). In each of these areas, Bayesian models are
accounting for (and in some cases even predicting) subtle data patterns in a
wide variety of psychological experiments.
A general conclusion emerging from this work is that people (and
other animals) optimize their performance by conforming to Bayes’
rule that specifies how posterior conditional probabilities (of a hypothesis
being true, given a pattern of data) are computed from the product of the
likelihood of those data given the truth of the hypothesis and the prior
probability of the hypothesis. This likelihood x prior product is divided by
the sum of similar products for all relevant hypotheses (a sum known as the marginalized
probability of the data). Part of the appeal of this approach is that, unlike
the emphasis on representational structure in symbolic models and soft
statistical constraints in connectionist models (Shultz,
2003),
Bayesian approaches emphasize both structure and statistics – effectively, by
computing statistics over structures, in a way that describes how knowledge is
modified by new evidence.
This resurgence of interest in Bayesian methods is somewhat
surprising given the Nobel-Prize-winning work showing that people are rather
poor Bayesians, subject to, e.g., the base-rate fallacy and the representativeness heuristic (Kahneman, Slovic, & Tversky, 1982; Kahneman & Tversky, 1996; Tversky & Kahneman, 1974, 1981). There is also
evidence that people confuse the direction of conditional probabilities, e.g.,
the probability of a symptom given a disease vs. the probability of a disease
given a symptom (Eddy,
1982).
Even experienced medical professionals deviate from Bayes
in these ways, creating medical inefficiencies and sometimes disastrous
outcomes.
Clearly, the new evidence that people are Bayesian optimizers
needs to be reconciled with this older work suggesting that people routinely
ignore prior probabilities and confuse conditional probabilities. An hypothesis
currently being floated to explain this discrepancy is that people are
implicit, but not explicit, Bayesians (Chater
et al., 2006). This seems a bit implausible given that the
distinction between implicit vs. explicit distinction is rather vague and may
not make much of a psychological difference in any case. It also seems to
conflict with experiments showing that prior probabilities are more
likely to be used if made more explicit (Bar-Hillel & Fischoff, 1981; Fischoff, Slovic, &
Lichtenstein, 1979; Gigerenzer, Hell, & Blank,
1988).
Interestingly, however, the implicit-explicit hypothesis does seem to be
testable. One could, for example, design psychological experiments that a) take
a modern experiment showing implicit conformity to Bayes’
rule and add a condition in which the numeric features are explicit, or b) take
an older experiment showing explicit deviations from Bayesian rationality and
add a condition in which the numeric features are implicit. According to the
implicit-explicit hypothesis, such new conditions should reverse the usual
trends.
Another, perhaps related hypothesis worth considering is
that what upsets explicit Bayesian reasoning is the use of probabilities in
problem descriptions. There is evidence that people do much better on otherwise
explicit uncertainty problems if the numerical information is presented in
terms of frequencies, rather than probabilities, although subjects still don’t
come very close to Bayesian norms (Chase,
Hertwig, & Gigerenzer,
1998; Gigerenzer & Hoffrage,
1995; Gigerenzer & Todd, 1999).
Qualifications
Ability to design, set up, and run
a psychology experiment. Ability to use statistical packages
such as SPSS for ANOVA.
References
Bar-Hillel,
M., & Fischoff, B. (1981). When do base
rates affect predictions? Journal of
Personality and Social Psychology, 41, 671-680.
Chase, V. M., Hertwig, R., & Gigerenzer, G.
(1998). Visions of rationality. Trends in Cognitive Sciences, 2,
206-214.
Chater, N., & Manning, C. D. (2006). Probabilistic
models of language processing and acquisition. Trends in Cognitive Sciences, 10, 335-344.
Chater, N., Tenenbaum, J. B., & Yuille, A.
(2006). Probabilistic models of cognition: Conceptual foundations. Trends in Cognitive Sciences, 10,
287-291.
Courville, A. C., Daw, N. D., & Touretzky, D.
S. (2006). Bayesian theories of conditioning in a
changing world. Trends in
Cognitive Sciences, 10, 294-300.
Eddy,
D. M. (1982). Probabilistic reasoning in clinical medicine: Problems and
opportunities. In D. Kahneman, P. Slovic
& A. Tversky (Eds.), Judgment under uncertainty: Heuristics and biases (pp. 249-267).
Cambridge: Cambridge University Press.
Fischoff, B., Slovic, P., & Lichtenstein, S.
(1979). Subjective sensitivity analysis. Organizational behavior and human decision
processes, 23, 339-359.
Gigerenzer, G., Hell, W., &
Blank, H. (1988). Presentation and content: The use of base rates as a
continuous variable. Journal of
Experimental Psychology: General, 14, 513-525.
Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian
reasoning without instruction: Frequency formats. Psychological Review, 102, 684-704.
Gigerenzer, G., & Todd, P. M. (1999). Simple heuristics that make us smart.
New York: Oxford University Press.
Kahneman, D., Slovic, P., & Tversky, A.
(1982). Judgment under
uncertainty: Heuristics and biases. Cambridge: Cambridge University Press.
Kahneman, D., & Tversky, A. (1996). On the reality
of cognitive illusions. Psychological
Review, 103, 582-591.
Körding, K. P., & Wolpert, D. M. (2006). Bayesian decision theory in sensorimotor
control. Trends in Cognitive
Sciences, 10, 319-326.
Shultz, T. R.
(2003). Computational
developmental psychology. Cambridge, MA: MIT Press.
Tenenbaum, J. B., Griffiths, T. L., & Kemp, C.
(2006). Theory-based Bayesian models of inductive learning and
reasoning. Trends in Cognitive Sciences,
10, 309-318.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics
and biases. Science, 185, 1124-1131.
Tversky, A., & Kahneman, D. (1981). The framing
of decisions and the psychology of choice. Science, 211, 453-458.
Yuille, A., & Kersten, D. (2006). Vision as Bayesian inference:
analysis by synthesis? Trends in
Cognitive Sciences, 10, 301-308.