causal learning
Causal Learning by Patricia W. Cheng and Marc J. Buehner (2012)
In the Oxford Handbook of Thinking and Reasoning, Szabolcs Kiss presents a detailed examination of causal learning through the perspectives of Cheng and Buehner.
Distinctions Regarding Causality
Causation vs. Association
Causation refers to a relationship where one event (the cause) directly affects another event (the effect). It is important to note that causation is not directly observable, but it can be inferred from observations.
Association is observable and can either indicate a causal relationship or not. An example of this distinction is the observed association between smoking and lung cancer, which does not imply that all associations indicate causation.
Hume's Distinction in Knowledge
Hume separates knowledge into analytic (based on definitions and logic) and empirical (based on sensory experience). He claimed that causal knowledge stems from empirical observations and argued that causality is fundamentally a mental construct.
Correlation vs. Causation
Basic statistical training highlights the principle that correlation does not equate to causation, affirming that one must not assume a causal relationship based solely on correlated data.
Observation vs. Intervention
There is a relevant distinction between observation (the passive act of seeing) and intervention (the active act of doing). Both methods contribute to the development of causal knowledge, thus enriching our understanding of causation.
Causal Inferences
Directionality of Causes and Effects
Causes are active agents that produce effects, whereas effects cannot instigate causes — this directionality is crucial in understanding causal relationships.
Distinguishing Observational and Interventional Learning in Rats
Experimental studies (such as those involving rats) show their ability to distinguish between observation and intervention, illustrating their capacity to learn about causal relationships through active participation (e.g., pressing a lever).
Diagnostic Causal Inference
This type of inference begins with an observed effect, from which possible causes are deduced, highlighting a reverse approach to causal understanding.
Experimentation as Intervention
Conducting experiments is a recognized method of intervention specifically aimed at uncovering causal relationships among variables.
Bayesian Inference and Causality
One key question in this framework involves assessing the probability of a causal link between two variables:
"How likely is it that a causal link exists between these two variables?"
Additionally, it considers the inverse relationship: "What is the probability with which a cause produces (alternatively, prevents) an effect?" (p. 223).
Causal Knowledge
Two-Way Street of Causation
Recent evidence contradicts the notion that the path from sensory experiences to causal knowledge is solely unidirectional. Instead, it suggests a reciprocal relationship where sensory experiences and causal knowledge continuously inform and shape one another (p. 229).
Computational Capabilities for Causal Reasoning
Three significant capabilities essential for effective causal reasoning in an agent include:
Ability to make deductive logical inferences.
Ability to compute statistical regularities.
Ability to represent and manage uncertainty in causal relationships.
Partitioning Events into Causes and Effects
A crucial question arises: "How do people arrive at their partitioning of the continuous stream of events into candidate causes and effects?" (p. 230).
Theories of Causal Understanding
The six main theories presented regarding causal understanding include:
A. Associationism (Hume)
B. Nativism (Kant, Leslie)
C. Constructivism (Piaget)
D. The Intervention Theory of Causation (Gopnik)
E. The Perceptual View of Causation (Michotte)
F. The Counterfactual View of Causation *(D. Lewis)