Post on 18-Aug-2018
Mechanistic Explanation in Cognitive Science and Neuroscience
Edouard Machery University of Pittsburgh
Three Issues
(1) What is a mechanistic explanation?
(2) Are explanations in neuroscience mechanistic?
(3) How are explanations in neuroscience and in psychology related?
Plan
1. Mechanistic explanation
2. Explanation in neuroscience
3. The relation of psychology and neuroscience
Plan
1. Mechanistic explanation
2. Explanation in neuroscience
3. The relation of psychology and neuroscience
Hempel’s Deductive-Nomological (DN) Model of Explanation
Explanantia: Sentence (in a language L) of a Law(s) of Nature Sentence describing the initial conditions .
Explanandum: Sentence describing the event or generalization to be explained
Relevance and Explanation
All males who take birth control pills regularly fail to get pregnant !
John Jones is a male who has been taking birth control pills regularly !
John Jones fails to get pregnant !
(Salmon, 1981)
Underlying Issue
What seems amiss is that we want to explain an event by citing its causes:
- causes are asymmetric
- causes are relevant
Upshot
Another model of explanation is needed, which acknowledges the explanatory role of causes and
does not require laws.
Mechanism
A mechanism is a physical object that does something (for artifacts, its function) by virtue of the organization of its parts and of their activities
(what they do).
Mechanistic Explanation
That is, (1) to identify the parts of a mechanism and (2) their activities, which produce the
phenomenon to be explained.
Mechanistic Explanations and then Failure of the DN Model
The mechanistic model of explanation acknowledges the explanatory role of causes and does not require laws, from which a phenomenon
can be deduced.
No Laws At All?
Generalizations describing the behavior of parts are however needed: “Laws in situ” (Cummins) or “mechanistically fragile generalizations” (Craver).
Constitutive Explanation
Mechanistic explanations are constitutive (in contrast to merely etiological): An explanation of
a phenomenon P involves determining what a mechanism is.
Decomposition and Recomposition
Discovering explanations depends on two complementary discovery heuristics: decomposition and recomposition.
Multi-Level Explanation
Mechanistic explanations are hierarchical and multi-level in that the activities of the parts of
mechanisms need to be explained as well.
Plan
1. Mechanistic explanation
2. Explanation in neuroscience
3. The relation of psychology and neuroscience
Case Study: The Action Potential
The discovery of the explanation of the action potential consists in understanding the
mechanism responsible for it: replacing “filler terms” with mechanism parts and their
interactions, etc.
Contrast with Cummins’s Functional Explanation
To explain a capacity φ of a system (characterized extensionally as a I-O function) is to identify a set of subcapacities that together
constitute φ.
A Narrow Perspective on Explanation in Cognitive Science and
Neuroscience
It is dubious that all explanations in neuroscience are mechanistic. In particular, computational
explanations are not mechanistic.
Plan
1. Mechanistic explanation
2. Explanation in neuroscience
3. The relation of psychology and neuroscience
The Issue
Are psychological explanations (particularly, the kind of explanation provided in cognitive science) autonomous from neuroscientific explanations?
The Claim
Psychological explanations are not autonomous because they are mechanisms sketches
(incomplete, to-be-filled-in) of neuroscientific mechanistic explanations.
The Argument1. Psychological explanations are functional explanations.
2. Functional explanations are sketches of mechanistic explanations.
3. Mechanistic explanations of cognitive competences are neuroscientific explanations.
4. If type-A explanations are sketches of type-B explanations, type-A explanations improve (get more explanatory) as their explanantia are connected to type-B explanantia
5. type-A explanations are autonomous from type-B explanations iff type-A explanations do not improve (get more explanatory) as their explanantia are connected to type-B explanantia.
6. Psychological explanations are not autonomous.
The Argument1. Psychological explanations are functional explanations.
2. Functional explanations are sketches of mechanistic explanations.
3. Mechanistic explanations of cognitive competences are neuroscientific explanations.
4. If type-A explanations are sketches of type-B explanations, type-A explanations improve (get more explanatory) as their explanantia are connected to type-B explanantia
5. Type-A explanations are autonomous from type-B explanations iff type-A explanations do not improve (get more explanatory) as their explanantia are connected to type-B explanantia.
6. Psychological explanations are not autonomous.
Functional Explanation
Explaining a capacity by “functional analysis” consists in identifying subcapacities and their organization, which are constitutive of the possession of the capacity to be explained.
Functional Explanation in Psychology
Explaining a psychological capacity by “functional analysis” consists in identifying psychological
subcapacities and their organization, which are constitutive of the possession of the capacity to
be explained.
The Argument1. Psychological explanations are functional explanations.
2. Functional explanations are sketches of mechanistic explanations.
3. Mechanistic explanations of cognitive competences are neuroscientific explanations.
4. If type-A explanations are sketches of type-B explanations, type-A explanations improve (get more explanatory) as their explanantia are connected to type-B explanantia
5. Type-A explanations are autonomous from type-B explanations iff type-A explanations do not improve (get more explanatory) as their explanantia are connected to type-B explanantia.
6. Psychological explanations are not autonomous.
The Argument1. Psychological explanations are functional explanations.
2. Functional explanations are sketches of mechanistic explanations.
3. Mechanistic explanations of cognitive competences are neuroscientific explanations.
4. If type-A explanations are sketches of type-B explanations, type-A explanations improve (get more explanatory) as their explanantia are connected to type-B explanantia
5. Type-A explanations are autonomous from type-B explanations iff type-A explanations do not improve (get more explanatory) as their explanantia are connected to type-B explanantia.
6. Psychological explanations are not autonomous.
The Argument1. Psychological explanations are functional explanations.
2. Functional explanations are sketches of mechanistic explanations.
3. Mechanistic explanations of cognitive competences are neuroscientific explanations.
4. If type-A explanations are sketches of type-B explanations, type-A explanations improve (get more explanatory) as their explanantia are connected to type-B explanantia
5. Type-A explanations are autonomous from type-B explanations iff type-A explanations do not improve (get more explanatory) as their explanantia are connected to type-B explanantia.
6. Psychological explanations are not autonomous.
Autonomy
Different types of autonomy:
(1) ontological autonomy
(2) evidential autonomy
(3) explanatory autonomy
Evidential Heteronomy
(1) Findings at a lower level can confirm or undermine high-level theories.
(2) Findings at a lower level are necessary to confirm or undermine high-level theories.
Explanatory Heteronomy
(Potential) explanations couched in higher-level terms improve to the extent that their explanantia are connected with (shown to be identical to or
constituted by) the explanantia of lower-level explanations.
Evidential and Explanatory Heteronomy
Even if psychology is evidentially heteronomous in either sense (1) or (2), it does not entail that it is
explanatory heterogeneous.
The Argument1. Psychological explanations are functional explanations.
2. Functional explanations are sketches of mechanistic explanations.
3. Mechanistic explanations of cognitive competences are neuroscientific explanations.
4. If type-A explanations are sketches of type-B explanations, type-A explanations improve (get more explanatory) as their explanantia are connected to type-B explanantia
5. Type-A explanations are autonomous from type-B explanations iff type-A explanations do not improve (get more explanatory) as their explanantia are connected to type-B explanantia.
6. Psychological explanations are not autonomous.
P&C’s Strategy
Examine three forms/presentations* of psychological explanations:
- task analysis
- functional analysis
- boxological models
1. A system S has a particular capacity because of the capacities of its parts and their organization.
2. Hence, the subcapacities that constitute S’s capacity are capacities of the parts of a system.
3. Hence, a functional explanation is a mechanism sketch.
P&C’s Argument for Premise 2
Problem
1 does not entail 2:
While the HL properties of a system depend on the properties of LL parts, they need not be
properties of these parts.
Upshot
But the argument against the autonomy of psychology requires that functional analyses
essentially are mechanism sketches.
Conclusion
Still little reason to deny the explanatory autonomy of psychology.
Claims to the contrary confuse dependence or evidential relations with explanatory ones.
Two Distinctions
1. How-possibly vs. how-actually mechanistic explanations
2. Mechanism sketches vs. mechanism schemata.
How possibly
- describe a mechanism that might be responsible for the explanandum
- heuristically useful
- does not explain adequately
How-Possibly vs. How-Actually Mechanistic Explanations
1. ConfusionsSchemata Sketches
how possibly
Complete description of
a possible mechanism
Incomplete description of
a possible mechanism
how actually
Complete description of
the mechanism
Incomplete description of
the mechanism
How actually
- describe a component that is responsible for the explanandum
- explains adequately
How-Possibly vs. How-Actually Mechanistic Explanations
Mechanisms Sketches vs. Mechanism Schemata
Sketches
Some components or capacities have not been specified.
1. Confusions
Craver mischaracterizes the distinction between how-possibly and how-actually mechanism
descriptions.
1. Confusions
Two distinctions:
- how possibly and how actually mechanism description.
- successful vs. unsuccessful explanation
1. Confusions
How-possibly vs. how-actually mechanism descriptions
Explaining how a phenomenon might be produced vs. is produced
1. ConfusionsSuccessful Unsuccessful
how possibly
epistemically possible and sufficient for
the phenomenon
epistemically impossible or insufficient for
the phenomenon
how actually
true description of
the mechanism
false or incomplete
description of the mechanism
1. Confusions
Craver tends to identify how-possibly mechanism descriptions with mechanism sketches and how-actually mechanism descriptions with mechanism
schemata.
Rebuttal
Functional analyses that are not mechanism sketches are neither actual nor merely possible description of the causal structure of system, if a description of a causal structure identifies parts
of a system and their causal interactions.
Rebuttal
Unclear why P&C take the first horn of the dilemma to be true, but a possible explanation is
that they take functional analysis to be underdetermined without evidence about the
mechanism bringing about the capacity.
Rebuttal
But even if this underdetermination is true, it is irrelevant to the issue at hand. We may use
information about parts of a system to identify the one true functional analysis, but this analysis need not be a mechanism sketch in that it need
not describe properties of the parts of the mechanismm.