Trends in Cognitive Sciences
Feature ReviewLogic, probability, and human reasoning
Section snippets
The nature of deductive reasoning
To be rational is to be able to make deductions – to draw valid conclusions from premises. A valid conclusion is one that is true in any case in which the premises are true [1]. In daily life, deductions yield the consequences of rules, laws, and moral principles [2]. They are part of problem solving, reverse engineering, and computer programming 3, 4, 5, 6 and they underlie mathematics, science, and technology 7, 8, 9, 10. Plato claimed that emotions upset reasoning. However, individuals in
Problems for logic as a theory of deductive reasoning
An ancient proposal is that deduction depends on logic (see also 16, 17, 18, 19, 20). Sentential logic concerns inferences from premises such as conjunctions (‘and’) and disjunctions (‘or’). Like most logics, it has two parts: proof theory and model theory [35]. Proof theory contains formal rules of inference for proofs. One major rule of inference in most formalizations is:
A → C
A
therefore, C
where A and C can be any sentences whatsoever, such as:
‘2 is greater than 1’ → ‘1 is less than 2’.
Proof
Probability logic
As a consequence of the preceding arguments, some cognitive scientists propose that probability should replace logic. Their theories differ in detail but overlap enough to have a label in common – the new paradigm 25, 26, 27, 28, 29. We refer to the paradigm as ‘probability logic’ or ‘p-logic’ for short. It presupposes that degrees of belief correspond to subjective probabilities 45, 46, 47, 48, 49, an idea that not all psychologists accept 50, 51. It focuses on conditionals, and one p-logician
The theory of mental models
During World War II, Kenneth Craik proposed that individuals simulate the world in mental models to make predictions, but that reasoning depends on verbal rules [59]. A more recent theory – the mental model theory – postulates that simulation underlies reasoning too 30, 31, 32, 33. It is a simple idea – people simulate possibilities – and most of its ramifications are integrated in a computer program, mReasoner [60], which is in the public domain at http://mentalmodels.princeton.edu/models/.
An evaluation of p-logic and mental models
P-logic has the great merit of allowing that reasoning can be tentative, uncertain, and probabilistic. It has inspired much ingenious research. However, what is the standing of its four key hypotheses?
Ramsey's test assesses the probability of conditionals. Perhaps reasoners use the test [45], but their estimates of conditional probabilities tend to violate the complete joint probability distribution (see below).
Conditionals have a defective truth table. Some experiments corroborate its
Do probabilities enter into pure deductions?
By ‘pure’ deductions, we mean those that make no reference to probabilities. P-logic, however, holds that probabilities are ubiquitous and that they unconsciously enter into pure deductions 21, 22, 23, 25, 26, 27, 28, 29. By contrast, the model theory implies that probabilities enter into the contents of reasoning only if invoked explicitly. We examine this proposal for conditional, syllogistic, and causal reasoning.
According to p-logic, a conditional such as:
If the FDA approves a drug then it
The probabilistic machinery of reasoning
At this point, readers may suspect that the model theory is deterministic through and through. In fact, it postulates that the machinery underlying reasoning – even for pure deductions – is probabilistic [30]. One illustration concerns evaluations of consistency – an important task because inconsistent beliefs can lead to disaster 109, 110. The only general way to use formal rules of inference [18] to establish the consistency of a set of assertions is to show that the negation of one assertion
Reasoning about probabilities
The model theory can explain how people reason about probabilities of various sorts. Consider the following inference:
A sign of a particular viral infection – a peculiar rash – occurs only in patients who are infected, but some patients with the infection do not have the rash. Is the infection more likely than the rash? (Yes.)
This deduction follows from the mental model of the possible individuals. Here is a numerical example:
There is a box in which there is at least a red marble, or else there
Concluding remarks
We began with two questions: does logic underlie human deductions and how do probabilities fit together with them? Despite the importance of logic to mathematics and the theory of computability, unconscious logical rules do not appear to be the basis of everyday reasoning. Arguments for this claim motivated the new paradigm in which reasoners rely instead on probability logic. It focuses on conditional assertions and postulates that individuals assess them by imagining that their if-clauses are
Acknowledgments
The authors thank Ruth Byrne, Rebecca Schwarzlose (TiCS editor), and two anonymous reviewers for their constructive criticisms of an earlier draft. They are also grateful to Igor Douven, Niki Pfeifer, Gernot Kleiter, and Klaus Oberauer for helping them to clarify their views about the four key hypotheses of the new paradigm. This research was supported by a Jerome and Isabella Karle Fellowship from the Naval Research Laboratory to S.S.K.
Glossary
- Bayesian net
- a directed graph in which each node represents a variable and arrows from one node to another represent conditional dependencies. It captures the complete joint probability distribution in a parsimonious way.
- Consistency
- a set of assertions is consistent if they can all be true at the same time.
- Counterexample
- in an inference, a possibility to which the premises refer but which is inconsistent with the conclusion.
- Deductive reasoning
- a process designed to draw a conclusion that follows
References (160)
General intelligence predicts reasoning ability even for evolutionarily familiar content
Intelligence
(2011)- et al.
Reasoning ability is (little more than) working memory capacity?!
Intelligence
(1990) - et al.
Framing human inference by coherence based probability logic
J. Appl. Logic
(2009) - et al.
Mental models of Boolean concepts
Cogn. Psychol.
(2011) - et al.
Toward a unified theory of reasoning
- et al.
Are humans good intuitive statisticians after all? Rethinking some conclusions of the literature on judgment under uncertainty
Cognition
(1996) - et al.
The Adams family
Cognition
(2010) - et al.
Probability in reasoning: a developmental test on conditionals
Cognition
(2015) - et al.
Illusory inferences: a novel class of erroneous deductions
Cognition
(1999) - et al.
Conceptual illusions
Cognition
(2010)
Strategies in syllogistic reasoning
Cogn. Sci.
Distinct neural substrates for deductive and mathematical processing
Brain Res.
Formal Logic: Its Scope and Limits
The psychology of moral reasoning
Judgm. Decis. Mak.
The psychological problem of Sudoku
Think. Reason.
Strategic changes in problem solving
J. Cogn. Psychol.
Kinematic mental simulations in abduction and deduction
Proc. Natl. Acad. Sci. U.S.A.
A theory of reverse engineering and its application to Boolean systems
J. Cogn. Psychol.
Mathematical Epistemology and Psychology
How We Reason
Thinking and Deciding
Mathematical Reasoning: Patterns, Problems, Conjectures, and Proofs
The influence of affect on higher level cognition: a review of research on interpretation, judgement, decision-making and reasoning
Cogn. Emot.
Models and cognitive change in psychopathology
J. Cogn. Psychol.
Working memory involvement in propositional and spatial reasoning
Think. Reason.
A Border Dispute: The Place of Logic in Psychology
The Psychology of Proof
Reasoning
Bayesian Rationality
Précis of Bayesian Rationality: the probabilistic approach to human reasoning
Behav. Brain Sci.
A Treatise on Probability
New paradigm psychology of reasoning
Think. Reason.
Questions and challenges for the new psychology of reasoning
Think. Reason.
Rationality in the new paradigm: strict versus soft Bayesian approaches
Think. Reason.
The new psychology of reasoning: a mental probability logical perspective
Think. Reason.
Dynamic inference and everyday conditional reasoning in the new paradigm
Think. Reason.
Mental Models
Deduction
The erotetic theory of reasoning: bridges between formal semantics and the psychology of deductive inference
Philos. Perspect.
Computability and Logic
On computable numbers, with an application to the Entscheidungsproblem
Proc. Lond. Math. Soc. Ser.
Engines of Logic: Mathematicians and the Origin of the Computer
Nonmonotonic Reasoning
Reasoning with exceptions: an event-related brain potentials study
J. Cogn. Neurosci.
Logic as Marr's computational level: four case studies
Top. Cogn. Sci.
Probabilistic single function dual process theory and logic programming as approaches to non-monotonicity in human vs artificial reasoning
Think. Reason.
The empirical study of norms is just what we are missing
Front. Psychol.
Logic, models, and paradoxical inferences
Mind Lang.
Cited by (163)
Probabilistic and deductive reasoning in the human brain
2023, NeuroImageA question-guided multi-hop reasoning graph network for visual question answering
2023, Information Processing and ManagementCitation Excerpt :Reasoning in neural networks has attracted increasing attention in the community (Chen et al., 2021a). It is the core component of V&L tasks and human ability (Johnson-Laird et al., 2015). In visual question answering, improving inference is a core step towards a reliable VQA system that can span various questions in real-world V&L scenarios.
Relations between geometric proof justification and probabilistic reasoning
2022, Learning and Individual DifferencesThe reemergence of the language-of-thought hypothesis: Consequences for the development of the logic of thought
2023, Behavioral and Brain SciencesExploring the Future of the Engineer Profession: The Effect of the Entrepreneurial Engineers' Mental Models on Their Opportunity Recognition Capabilities
2024, IEEE Transactions on Engineering ManagementOn the Cognitive Logic of Human Propositional Reasoning: Merging Ranking Functions
2024, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)