Advances in Experimental Philosophy of Causation

Pascale Willemsen, Alex Wiegmann, Pascale Willemsen, Alex Wiegmann

• Éditeur: Bloomsbury Academic
• Année: 2022
• ISBN:
  9781350235823
• Édition: 1
• Sujet: Philosophy
• Sous-sujet: Philosophical Metaphysics

What is the connection between causation and responsibility? Is there a best way to theorize philosophically about causation? Which factors determine and influence what we judge to be the cause of something? Bringing together interdisciplinary research from experimental philosophy, traditional philosophy and psychology, this collection showcases the most recent developments and approaches to questions about causation. Chapters discuss the diverse theoretical ramifications of empirical findings in experimental philosophy of causation, providing a comprehensive survey of key issues such as the perception and learning of causal relations, omission, normative considerations, mechanism, voluntariness and legal theories of causation. With novel contributions from both experts and rising stars, Advances in Experimental Philosophy of Causation demonstrates the value of empirical work and opens new domains of inquiry at the cutting edge of the field.

1

Revisiting Hume in the Twenty-first Century: The Possibility of Generalizable Causal Beliefs Given Inherently Unobservable Causal Relations

Nicholas Ichien and Patricia W. Cheng

The present chapter is an introduction to a basic problem in causal induction: how is generalizable causal knowledge possible? We first present the problem of causal induction as posed by David Hume (1739/1987, 1748/1975) and clarify three confusions surrounding this problem. We go on to review empirical evidence from four perspectives that all provide support for Hume’s view: Causal relations are not in the input to the reasoner/cognitive system. What is in the input are merely the states of the candidate causes and the state of the outcome-in-question due to all its causes present in the context. Making a causal inference about a target candidate cause therefore requires decomposing the observed outcome into contributions from the target cause and from other causes of the outcome in the context. Nature does not tell the reasoner how to decompose the observed outcome—the task is up to the reasoner. After establishing the problem of causal induction from the perspective of cognitive science, we explain what the assumption of causal invariance is and why it is a necessary constraint for rational causal induction. We end the chapter by relating our analysis of causal invariance to normative causal inference in statistics. This chapter does not assume any background knowledge of work on causal induction in philosophy or psychology. Our intention is for it to be of interest to upper-level undergraduate students, graduate students, and anyone else who enjoys thinking through the problem our mind solves when it aims to infer a generalizable causal relation.

People often have the compelling intuition that they directly “see” causation, and thus have no need to infer causation. If they see an unfortunate person killed by a volcanic eruption, overtaken by a pyroclastic flow, it may seem hard to deny that they perceived the reality of the volcanic eruption killing the person.¹ If they see a moving ball hit a stationary ball, and the stationary ball starts to move away, they “see” the true “launching” into motion of one ball by motion in the other. If their right-hand fingers scratch a mosquito bite on their left arm, and their left arm feels relief from the itch, they directly perceive the relieving of the itch by their scratching.

Hume (1739/1987) argues that counter to our compelling intuition that a moving ball launches a stationary ball when we observe the former hitting the latter, the causal aspect of that intuition is an inference in our mind and is absent in the observation itself. Hume (1748/1975, p. 37) also brings attention to an assumption so intuitive that we may be unaware of making it: Whenever we generalize from a learning context to an application context, we assume, “the future will resemble the past.” He goes on to state its implication, “If there is any suspicion that the course of nature may change, … all experience becomes useless …” Together, Hume’s two points raise the question: If causal perceptions and beliefs are mental constructs absent in the observations in our experience, on what basis would one expect these mental constructs to capture the unchanging course of nature, such that experience is not useless?

Three confusions clouding the nature of the problem of causal induction

It is tempting to conclude that the compelling perception of causal relations renders inference unnecessary, at least in cases in which causation appears “observable.” However, the conclusion that causation is observable involves three sources of confusion.

The first is a confusion between the input of the causal induction process and its output. The conclusion mistakes the compelling perception of causation, as illustrated in our examples, to be the input to the causal induction process, when it is in fact the output to be explained (see Henle, 1962, for an example in which confusion about the input to a cognitive process, deductive inference in her case, creates confusion about the process itself). This confusion may be due to the vagueness of Hume’s criterion for what he does or does not find “evident” in the observations (1739/1987, pp. 649–50). In contemporary information-processing language, a paraphrase of Hume’s thesis that causal relations are not evident in the observations would be: Causal relations are not in the input available to a process that infers cause and effect—the construct we label the causal-induction process. Given that our sensory input does not contain causal relations, but we “know” causal relations, there must be a downstream process that does the work of arriving at the causal output from its noncausal input.

The unobservability of causation is a specific form of the general challenge of formulating adaptive knowledge: reality in the world does not come represented (Goodman, 1955; Hawking and Mlodinow, 2010; Kant, 1781/1965). All our perceptions and conceptions of reality are our representations of it, formulated within an infinite search space. Consider our perception of a cube. The 2-dimensional image cast by a cube on our retina is ambiguous in that it can map onto an infinite number of differently shaped 3-dimensional objects (e.g., see Pizlo, 2001). Yet, despite the inherent under-determination of the distal object, we perceive a cube. Narrowing down to this adaptive percept in the infinite space of possible distal objects illustrates the application of potent constraints in the form of a priori assumptions, in this case the default assumption that the distal object has the simplest form that is consistent with the image (i.e., the object is a “parsimonious explanation” of the image).

Thus, with respect to the stereoscopic vision process, 3-dimensionality is “unobservable”—a shorthand for “being absent in the input to a process”—and is the to-be-explained output of the process. Likewise, causation is “unobservable” with respect to the causal induction process, and the perceived “necessary connection” between a cause and an effect is the to-be-explained output (Hume, 1739/1987).

A second source of confusion is that the apparent examples of observable causation often involve prior causal knowledge at a more abstract level than the particular causal relation in question. Although a reasoner may be witnessing a pyroclastic flow hitting someone for the first time, they almost certainly know, at a more general level, from knowledge of landslides and fires, that being struck by massive flows of hot or heavy matter can be fatal. Lien and Cheng’s (2000) hierarchical consistency hypothesis explains how consistency and inconsistency of covariations between potential cause and effect variables across representations at different levels of abstraction can explain conclusions of causality or noncausality ostensibly based on a single instance. Their paper presents evidence showing that information beyond what is in the single instance gets recruited, more specifically, that judgments involving a single instance can be explained by retrieval from causal schemas in long-term memory formed by past causal inferences, rather than by “observable” causality. (See Rips (2011) for a review of evidence and arguments against perception of causality as the source of the causal knowledge.)

A third source of confusion is that examples of observable causation concern situations in which only one cause is perceived to be present (i.e., the reasoner assumes no background causes). In such cases, an inferential process, either inductive (Cheng, 1997) or deductive, can account for the causal percept. Deduction such as the following would reach our intuitive causal conclusions:

Premises:1.effect e occurred in situation x

2.effects do not occur without a cause

3.c is the only candidate cause in situation x

Conclusion: c caused e. In other words, the fact that we humans are able to judge causation in situations involving one single plausible cause does not imply that we do not have a general causal-induction process capable of inferring new causal knowledge in situations involving more than one plausible cause. The single-cause situation may be regarded as a trivial case of the application of that inference process.

To illustrate that causation in situations with a single plausible cause is not observed but inferred, we review the striking “ph...