Jeffrey (1983) proposed a generalization of conditioning as a means of updating probability distributions when new evidence drives no event to certainty. His rule requires the stability of certain conditional probabilities through time. We tested this assumption (“invariance”) from the psychological point of view. In Experiment 1 participants offered probability estimates for events in Jeffrey’s candlelight example. Two further scenarios were investigated in Experiment 2, one in which invariance seems justified, the other in which it does not. Results were in (...) rough conformity to Jeffrey (1983)’s principle. (shrink)

It is easy to construct pairs of sentences X, Y that lead many people to ascribe higher probability to the conjunction X-and-Y than to the conjuncts X, Y. Whether an error is thereby committed depends on reasoners’ interpretation of the expressions “probability” and “and.” We report two experiments designed to clarify the normative status of typical responses to conjunction problems. © 2004 Cognitive Science Society, Inc. All rights reserved.

We consider a panel of experts asked to assign probabilities to events, both logically simple and complex. The events evaluated by different experts are based on overlapping sets of variables but may otherwise be distinct. The union of all the judgments will likely be probabilistic incoherent. We address the problem of revising the probability estimates of the panel so as to produce a coherent set that best represents the group’s expertise.

We describe a logic of preference in which modal connectives reflect reasons to desire that a sentence be true. Various conditions on models are introduced and analyzed.

We propose a theory that relates perceived evidence to numerical probability judgment. The most successful prior account of this relation is Support Theory, advanced in Tversky and Koehler. Support Theory, however, implies additive probability estimates for binary partitions. In contrast, superadditivity has been documented in Macchi, Osherson, and Krantz, and both sub- and superadditivity appear in the experiments reported here. Nonadditivity suggests asymmetry in the processing of focal and nonfocal hypotheses, even within binary partitions. We extend Support Theory by (...) revising its basic equation to allow such asymmetry, and compare the two equations’ ability to predict numerical assessments of probability from scaled estimates of evidence for and against a given proposition. Both between- and within-subject experimental designs are employed for this purpose. We find that the revised equation is more accurate than the original Support Theory equation. The implications of asymmetric processing on qualitative assessments of chance are also briefly discussed. (shrink)

Eric Martin and Daniel N. Osherson present a theory of inductive logic built on model theory. Their aim is to extend the mathematics of Formal Learning Theory to a more general setting and to provide a more accurate image of empirical inquiry. The formal results of their study illuminate aspects of scientific inquiry that are not covered by the commonly applied Bayesian approach.

Alternative models of idealized scientific inquiry are investigated and compared. Particular attention is devoted to paradigms in which a scientist is required to determine the truth of a given sentence in the structure giving rise to his data.

Most speakers experience unclarity about the application of predicates like tall and red to liminal cases. We formulate alternative psychological hypotheses about the nature of this unclarity, and report experiments that provide a partial test of them. A psychologized version of the ‘vagueness-as-ignorance’ theory is then advanced and defended.

We advance a theory of inductive reasoning based on similarity, and test it on arguments involving mammal categories. To measure similarity, we quantified the overlap of neural activation in left Brodmann area 19 and the left ventral temporal cortex in response to pictures of different categories; the choice of of these regions is motivated by previous literature. The theory was tested against probability judgments for 40 arguments generated from 9 mammal categories and a common predicate. The results are interpreted in (...) the context of Hume’s thesis relating similarity to inductive inference. (shrink)

We advance a model of human probability judgment and apply it to the design of an extrapolation algorithm. Such an algorithm examines a person's judgment about the likelihood of various statements and is then able to predict the same person's judgments about new statements. The algorithm is tested against judgments produced by thirty undergraduates asked to assign probabilities to statements about mammals.

A paradigm of scientific discovery is defined within a first-order logical framework. It is shown that within this paradigm there exists a formal scientist that is Turing computable and universal in the sense that it solves every problem that any scientist can solve. It is also shown that universal scientists exist for no regular logics that extend first-order logic and satisfy the Löwenheim-Skolem condition.

A glance at the sky raises my probability of rain to .7. As it happens, the conditional probabilities of each state given rain remain the same, and similarly for their conditional probabilities given no rain. As Jeffrey (1983, Ch. 11) points out, my new distribution P2 is therefore fixed by the law of total probability. For example, P2(RC) = P2(RC | R)P2(R)+P2(RC | ¯.

A model of idealized scientific inquiry is presented in which scientists are required to infer the nature of the structure that makes true the data they examine. A necessary and sufficient condition is presented for scientific success within this paradigm.

1. Characterizing randomness. Consider a physical process that, if suitably idealized, generates an indefinite sequence of independent random bits. One such process might be radioactive decay of a lump of uranium whose mass is kept at just the level needed to ensure that the probability is one-half that no alpha particle is emitted in the nth microsecond of the experiment. Let us think of the bits as drawn from {0, 1} and denote the resulting sequence by x with coordinates x0, (...) x1, . . .. Now wouldn’t it be odd if there were a computer program P with the following property? (shrink)

Because the conjunction pandq implies p, the value of a bet on pandq cannot exceed the value of a bet on p at the same stakes. We tested recognition of this principle in a betting paradigm that (a) discouraged misreading p as pandnotq, and (b) encouraged genuinely conjunctive reading of pandq. Frequent violations were nonetheless observed. The findings appear to discredit the idea that most people spontaneously integrate the logic of conjunction into their assessments of chance.

Establishing reasonable, prior distributions remains a significant obstacle for the construction of probabilistic expert systems. Human assessment of chance is often relied upon for this purpose, but this has the drawback of being inconsistent with axioms of probability. This article advances a method for extracting a coherent distribution of probability from human judgment. The method is based on a psychological model of probabilistic reasoning, followed by a correction phase using linear programming.

Compared to truthful answers, deceptive responses to queries are expected to take longer to initiate. Yet attempts to detect lies through reaction time (RT) have met with limited success. We describe a new procedure that seems to increase the RT difference between truth-telling and lies. It relies on a Stroop-like procedure in which responses to the labels true and false are sometimes reversed. The utility of this method is assessed in a laboratory study involving both statements of fact and attitude. (...) Three different ethnic groups participated: Americans of European ancestry, East Asians, and native Arabic speakers. For all three groups, our technique allowed identification of prevarication reliably better than chance. (shrink)

The volumes are self contained and can be used individually in upper-level undergraduate and graduate courses ranging from introductory psychology, linguistics, ...

We examine the prospects for finding “best possible” or “ideal” computing machines for various learning tasks. For this purpose, several precise senses of “ideal machine” are considered within the context of formal learning theory. Generally negative results are provided concerning the existence of ideal learning‐machines in the senses considered.

We consider drawbacks to scientific methods that prefer simple hypotheses to complex ones that cover the same data. The discussion proceeds in the context of a precise model of scientific inquiry.

The remarkable transition from helpless infant to sophisticatedfive-year-old has long captured the attention of scholars interested inthe discovery of knowledge. To explain these achievements, developmentalpsychologists often compare children's discovery procedures to those ofprofessional scientists. For the child to be qualified as a ``littlescientist'', however, intellectual development must be shown to derivefrom rational hypothesis selection in the face of evidence. In thepresent paper we focus on one dimension of rational theory-choice,namely, the relation between hypothesis confirmation and evidencediversity. Psychological research suggests cultural (...) variability inappreciating evidence diversity and lack of such appreciation by youngchildren. Before reaching conclusions about the ``little scientist''thesis, however, it is essential to normatively analyze the diversityissue. We undertake such an analysis within a Bayesianperspective. (shrink)

J. B. Predd S. R. Kulkarni H. V. Poor D. N. Osherson Department of Electrical Engineering Department of Psychology Princeton University Princeton University Princeton, NJ 08544 Princeton, NJ 08544 {jpredd,kulkarni,poor}@princeton.edu osherson@princeton.edu..

It is easy to construct pairs of sentences X, Y that lead many people to ascribe higher probability to the conjunction X‐and‐Y than to the conjuncts X, Y. Whether an error is thereby committed depends on reasoners' interpretation of the expressions “probability” and “and.” We report two experiments designed to clarify the normative status of typical responses to conjunction problems.

WORDNET, a ubiquitous tool for natural language processing, suffers from sparsity of connections between its component concepts (synsets). Through the use of human annotators, a subset of the connections between 1000 hand-chosen synsets was assigned a value of “evocation” representing how much the first concept brings to mind the second. These data, along with existing similarity measures, constitute the basis of a method for predicting evocation between previously unrated pairs.

Decision makers often rely on expert opinion when making forecasts under uncertainty. In doing so, they confront two methodological challenges: the elicitation problem, which requires them to extract meaningful information from experts; and the aggregation problem, which requires them to combine expert opinion by resolving disagreements. Linear averaging is a justifiably popular method for addressing aggregation, but its robust simplicity makes two requirements on elicitation. First, each expert must offer probabilistically coherent forecasts; second, each expert must respond to all our (...) queries. In practice, human judges (even experts) may be incoherent, and may prefer to assess only the subset of events about which they are comfortable offering an opinion. In this paper, a new methodology is developed for combining expert assessment of chance. The method retains the conceptual and computational simplicity of linear averaging, but generalizes the standard approach by relaxing the requirements on expert elicitation. The method also enjoys provable performance guarantees, and in experiments with real-world forecasting data is shown to offer both computational efficiency and competitive forecasting gains as compared to rival aggregation methods. This paper is relevant to the practice of decision analysis, for it enables an elicitation methodology in which judges have freedom to choose the events they assess. (shrink)

This paper provides a mathematical model of scientific discovery. It is shown in the context of this model that any discovery problem that can be solved by a computable scientist can be solved by a computable scientist all of whose conjectures are finitely axiomatizable theories.

One version of the problem of induction is how to justify hypotheses in the face of data. Why advance hypothesis A rather than B — or in a probabilistic context, why attach greater probability to A than B? If the data arrive as a stream of observations (distributed through time) then the problem is to justify the associated stream of hypotheses. Several perspectives on this problem have been developed including Bayesianism (Howson and Urbach, 1993) and belief-updating (Hansson, 1999). These are (...) broad families of approaches; the citations are meant just as portals. (shrink)