Contribution
This paper takes the explanatory pluralism of Kendler’s 2005 paper (LIT-552, cited as reference 1 for its opening premise) and gives it a method. It imports the mechanistic philosophy of science of Bechtel, Craver and Wimsatt into psychiatry: explanation is the decomposition and reassembly of multilevel mechanisms. It then uses Marr’s levels to settle the relation between biological and psychological explanation as implementation rather than replacement. No new empirical result is claimed.
Key insight
Two different integration problems are at stake, and the paper keeps them apart. The first is part–whole: a disorder’s causes sit at different levels (molecules, neurons, persons, societies) of one mechanism, and these levels interact non-additively and in loops. The answer is to decompose where the system is locally decomposable and then stitch the pieces together. The second is perspectival: brain and mind are two descriptions of the same process. The answer is that the psychological description states the function, and the biology is how the function is implemented. On that view a gene-to-protein finding has not explained schizophrenia until someone can say how the protein abnormality produces delusions and hallucinations. The paper puts this as the resident’s question: “how does this abnormality in protein Y lead to the characteristic symptoms?”
Assumptions
- Mechanistic explanation is the right model for the biological and social sciences relevant to psychiatry. Bechtel is quoted: “the decomposition required by mechanistic explanation is reductionist, but the recognition that parts and operations must be organized into an appropriate whole provides for a robust sense of a higher level of organization”.
- Levels are part–whole relations (“molecules make up a membrane, neurons make up a circuit … individuals make up a society”).
- The brain is functionally “a machine designed by evolution”, so reverse-engineering applies, and function can only be stated in psychological terms (“Neurochemical terms will not work”).
- Marr’s scheme generalises. It was designed for information-processing systems such as vision. The paper grants that “many psychiatric problems are not currently amenable to an information-processing perspective” but holds that “Marr’s underlying logic is nonetheless valid for psychiatry across the board”.
Key results
These are arguments, with illustrative evidence cited.
- Against law-based explanation. A physics-derived law model, with “all real causal processes … understood from one perspective and one set of laws”, “has been counterproductive” in mental health and has fed both hard reductionism and hard emergentism. The home-heating parable sets the three views side by side: the reductionist dismantles the thermostat, the emergentist models the temperature curve, and the mechanist models the whole house.
- Decomposability criteria (after Bechtel and Richardson, and Wimsatt). An “easily decomposable” system shows aggregativity (parts combine additively), “intersubstitutable” parts with context-independent functions, and no causal loops. Loops can be intra-level or inter-level. The most important is “recursive, or top-down, causality”, in which the system’s output feeds back on its own components.
- Alcohol dependence fails all three. Risk factors sit at four levels (biological/genetic, psychological, social, cultural/economic). Genetic effects vary with religion, marital status and social environment, so additivity and intersubstitution fail. There are causal loops: genes shape the response to ethanol, which shapes expectancies and consumption; tolerance produces a positive feedback loop; temperament selects risky peers, the “brain has feet”; and “reverse” cultural transmission makes offspring of heavy drinkers abstain more often.
- Other disorders. Sensitivity to adversity is context-dependent. Neuroticism generates the stressors that raise depression risk. A fearful child’s avoidance prevents habituation. In panic patients, the illusion of control over CO₂ reduced panic symptoms (Sanderson, Rapee and Barlow 1989).
- Nondecomposability is not defeat. Science has handled nondecomposable systems before (the citric acid cycle) by first finding “subareas of local decomposability”. “The naive emergentism that opines that the system is so complex and interrelated that we cannot possibly study any part in isolation is just plain wrong”, and the reductionist belief that the parts “simply fit together” is “no less wrong”. With Wimsatt, more detail should reveal more emergence: etiological pathways will be “deeply recursive”, a “causal thicket”, and reassembly involves Craver’s “explanatory oscillation” across levels.
- Implementation, not replacement. In replacement, the higher-level construct becomes redundant, as temperature is to mean kinetic energy. In implementation, the lower-level theory supplies “the mechanistic details of how the functions proposed by the more abstract theory actually get accomplished”. Following Gold and Stoljar on Kandel’s Aplysia work, Hatfield on vision and Bechtel on memory, neuroscience has progressed by implementing processes psychologists first described. Conclusion: “biology will implement but not replace psychology”, and the two “will coevolve”.
- Concessions. Hard reductive and emergent perspectives can still be useful. Therapies can come from gene findings without any account of mechanism, and CBT “emerged from psychological constructs that contained no biology”. Such perspectives give only part of the picture.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Psychiatric disorders have causes at several levels, inside and outside the person | strong as a summary of cited epidemiology and genetics | alcohol-dependence risk factors, refs 11–30 |
| C2 | Those causes are non-aggregative: genetic effects depend on biochemical and psychosocial context | moderate-to-strong: cited G×E and epistasis findings | refs 31–36, 48–56 |
| C3 | Psychiatric etiology contains causal loops, including top-down loops | moderate: cited examples | refs 37–47, 58–61 |
| C4 | Hard reduction will not work, because of non-aggregativity and causal loops | argued from C2–C3 | Implications of Nondecomposability |
| C5 | Local decomposition followed by reassembly can still make progress | argued by analogy to cell biology and memory research | Bechtel; citric acid cycle |
| C6 | Biology will implement, not replace, psychology in explaining psychiatric disorders | philosophical argument, with historical cases | Marr; Gold & Stoljar; Hatfield; Bechtel |
| C7 | More detailed knowledge will reveal more cross-level interaction, not more simplicity | a prediction, after Wimsatt | quote, ref 63 |
Concepts
- mechanistic explanation: the explication of causal mechanisms and of “how those mechanisms are actually instantiated in the world”.
- hard reductionism / hard emergentism: the views that psychiatric illness is best explained solely in molecular-neuroscience terms, or solely in mental or social terms that “cannot be deduced from biology”.
- level: a part–whole (component–system) relation within one mechanism, distinct from a perspective on the same process.
- easily decomposable system: aggregative, with intersubstitutable, context-independent parts and no causal loops.
- top-down (recursive) causality: the output of the whole system feeds back on its own components.
- local decomposability: a subsystem simple enough to study in isolation.
- explanatory oscillation (Craver): moving back and forth between levels while reassembling a mechanism.
- implementation vs replacement: a lower-level theory either fills in how a higher-level function is done, or makes the higher-level construct redundant.
Connections
- Kendler 2005 (LIT-552), unread; relation declared as
extends. This paper cites it as reference 1 for its opening premise that etiological models “need to be pluralistic or multilevel”. Its two sections develop what the 2005 abstract names but this record cannot read. The 2005 paper’s “patchy reductionism” and “piecemeal integration” become local decomposition and reassembly. Its mind–brain propositions become the Marr-based implementation thesis. The 2005 propositions against substance dualism and epiphenomenalism are not argued here. - Engel (LIT-548), unread. Cited (reference 3, with Adolf Meyer) as an “earlier integrationist account”. Kendler offers the mechanistic approach as its “conceptually rigorous descendant”, which implies that Engel’s model lacked a method of integration. The paper does not say this outright.
- Borsboom (LIT-547). Both reject a single common cause and both put causal loops at the centre. Kendler’s loops cross levels (gene → sensitivity → expectancy → environment). Borsboom’s run among symptoms, and Principle 2 pushes genes and environments into the external field or into the constitution of a symptom. Borsboom cites Kendler’s later papers for the absence of common causes.
- Hofmann and Hayes (LIT-544). Both reject eliminative reductionism. Both cite the illusion-of-control panic study as a psychological, top-down effect. Kendler’s “biology implements psychology” is compatible with their demand that processes be stated in modifiable psychological terms.
- Salkovskis, Sighvatsson and Sigurdsson (LIT-538). The fearful child whose avoidance “prevent[s] the habituation of the initial fear response” is a maintenance loop of exactly the kind their Theory A is built on, with avoidance protecting a threat meaning.
- Mental causation (LIT-147). Pernu asks whether interventionist evidence supports nonreductive mental causation. Kendler’s implementation relation offers a non-eliminative alternative to that framing: the psychological level is not reduced away, because it states what the mechanism does.
Bearing on the record
- It supports the convergence proposed in the batch report (disorder is maintained by self-reinforcing loops, not a single latent cause). Kendler adds that the loops cross levels, including gene–environment and top-down loops. That would widen the candidate THEORY from “loops among symptoms” to “loops across levels”.
- Candidate THEORY, Proposed: “In explaining psychiatric disorders, biological findings implement psychological functions rather than replacing them.” Source: this paper. Promote it only with further sources and with cases from psychiatry, not vision or memory.
- No ML instruction. Marr’s levels are well known in machine learning, but here they are used for psychiatric explanation. Nothing here belongs in the anthology.
- Filed as THEORY-048, Proposed. Its cross-level loops are also one source of THEORY-058.
Limitations
- Illustrative, not systematic. The alcohol-dependence sketch is “much simplified”, and the loop examples are selected cases.
- Marr’s generalisation is asserted. The paper concedes that many disorders are not yet amenable to an information-processing account, then holds that the logic applies “across the board”.
- No account of how reassembly is done beyond the analogies to cell biology and memory; “stitching” is named, not specified.
- The implementation thesis is argued from neuroscience successes in perception, learning and memory, not from any psychiatric case.
Open questions
- Are there psychiatric constructs that biology will replace rather than implement, as kinetic energy did temperature? The paper names none and rules none out.
- Do cross-level loops of Kendler’s kind fit inside a symptom-network model (LIT-547), or do they need nodes that are not symptoms?
- What would show that a reassembled multilevel model of a disorder is complete?
Corrections
- none to a seeded skim (there was no seed)
- Reference slip. Reference 53 (Blatt & Zuroff, Clin Psychol Rev 12:527–562) is dated 1969; volume 12 of that journal is 1992.