David Deutsch and Brett Hall: What Would Count Against “Hard to Vary”?

Research record. GitHub source and revision history · Canonical published edition

Brett Hall’s critique of Michael Levin raises a question for all three: what would show that the capacity or cause they invoke is doing the work?

Bottom line. David Deutsch’s “hard to vary” test helps expose arbitrary explanations. It cannot, by itself, establish that a proposed cause or capacity is indispensable. Brett Hall is right to ask Michael Levin what distinguishes persuasion from ordinary physical change. The same question reaches Hall’s claim that only humans possess explanatory creativity and consciousness. Each claim needs evidence that can distinguish it from a serious rival. Otherwise, a conjecture can become a reason to fund the wrong investigation, mistake fluency for understanding in education, or discount signs of suffering in animals that cannot explain what they feel.

Start with Deutsch’s strongest example

Deutsch compares two explanations of the seasons. In the myth, the seasons follow the movements of Persephone. The story can be rewritten to fit opposite seasons in the two hemispheres. The account based on Earth’s tilt connects the seasons to the angle of sunlight, which has independently observable effects. Its details have jobs. Change those details and other things the theory gets right may stop following. That is what Deutsch means when he says the details “play a functional role.” Deutsch, The Beginning of Infinity, chapter 1.

Here is where I would put the pressure. Converting metres to feet changes a description without changing the explanation. Replacing one proposed mechanism with another can preserve the fit to familiar evidence while changing what should happen in a new case. A rigid model can also fail because its tightly connected assumptions are wrong. So tell me what the comparison holds fixed, which variations make genuinely different claims, and where those claims predict different results. The seasons example works because those independent comparisons are available.

Hall explicitly says testability is necessary for scientific theories. Deutsch, for his part, explains why a failed prediction could reveal a problem in the apparatus or a background assumption. Popper also allowed researchers to revise an auxiliary assumption. The Internet Encyclopedia of Philosophy describes a legitimate repair as one that “allows for the generation of new, falsifiable predictions.” If a thermometer gave the wrong reading, check it against another instrument. If a rescue changes only the story we tell about the result already in hand, ask what the rescue now puts at risk. Hall on experimental testability, Deutsch on experimental tests, IEP on auxiliary and ad hoc hypotheses, Popper, §§20 and 22.

Apply Hall’s question in both directions

Repairing a clock and convincing a colleague of Euler’s identity both change a system. Hall asks what “persuadability” lets us distinguish once it covers both. A graded concept could do useful work if, for example, the proposed differences predict which systems learn a new rule, correct a mistake, or transfer what they learned. Those abilities can come apart. The label gets its meaning in the comparison. Hall, Categories and Continuums, episode 257. My bubble-sort analysis works through a case where local rules produce a surprising result and the exact mechanism of clustering still matters.

Around minute 52 of that discussion, Hall conjectures that only humans possess explanatory creativity, consciousness, and minds, with “mind” tied to his demanding sense of explanatory ability. He treats the conjecture as open to criticism. That leaves three questions to investigate. Can a system generate and correct explanations? Does it experience anything? How far does its explanatory ability reach? Deutsch’s own account allows for nonhuman persons and treats the connection between creativity and consciousness as conjectural. A definition of “mind” can stipulate which ability qualifies. Whether the ability is exclusive to humans, or required for experience, remains a question about the systems themselves. Hall, episode 257, Deutsch, The Beginning of Infinity, chapter 7.

When a conjecture becomes a rule

In his essay on animals, Hall distinguishes pain from suffering, conjectures that suffering requires a kind of explanatory interpretation, and urges caution about cruelty. Consider a welfare committee that makes the conjecture its screening rule. Under that rule, signs of distress could be dismissed because the animal has never demonstrated the required style of reasoning. Hall, Humans and Other Animals.

Robyn Crook’s octopus study shows what such a rule could screen out. Octopuses avoided a place associated with a noxious stimulus. Those exposed to that stimulus later preferred a place associated with local pain relief, while controls given the same local anesthetic without the prior stimulus showed no such preference. The comparison bears on whether the response reflects an aversive state that relief changes, beyond a withdrawal reflex or a general attraction to the drug. A committee considering precautions has evidence here that deserves weight even while the exact character of octopus experience remains open. Crook, iScience, 2021.

The same mistake can travel in either direction. A school can mistake a polished answer for understanding and never check whether the learner can find and fix an error in a new problem. A public body can mistake a computational system’s explanatory fluency for authority to decide for people affected by it. A funder can pay to elaborate a speculative cause while the experiment that could distinguish it from a simpler account goes undone. These are risks of turning a classification into a decision rule. Evidence that bears directly on the decision should be able to overturn the classification or limit its use. Evolving God Out Of The Gaps follows that transfer of authority across several disputes. Synthbiosis works out when moral relevance can follow even while experience remains unresolved.

Show what the added cause does

The Stanford Encyclopedia of Philosophy describes one interventionist account of causal effect as “the difference made to the value of Y by a change or difference in the value of X.” The important question is which change accounts for the result under the stated conditions. If a proposed influence selects one physical pathway over another, tell me when that happens and how the pathway would differ without it. A valve can redirect flow using pressure already in the system. A proposed cause can change the timing, route, or probability of an outcome even when the energy balance gives us no way to distinguish the alternatives. SEP, Causation and Manipulability, §5.

Thermodynamics constrains how the change is physically realized. The accounting depends on the operation, boundary, and timescale. A memory can retain a distinction over a relevant interval without continuous power. Landauer’s familiar bound concerns a particular reset of an initially equiprobable bit in a symmetric memory at a specified temperature. The physical question is how a distinction is established, retained, read, or changed where that operation matters to the explanation. If the proposal requires departing from a well-tested constraint, say where and give us evidence for the departure. Parrondo, Horowitz, and Sagawa, Thermodynamics of Information, my Thermodynamic Monism audit.

Buchner’s cell-free fermentation shows why this matters. Sugar still fermented in an extract after intact yeast cells had been removed. The material made by yeast still mattered. The need for intact living cells during the reaction lost its warrant. Papineau traces a similar pattern in physiology: conservation of energy left room for some proposed vital or mental forces, while detailed work on physical mechanisms undermined the need to invoke those forces for the effects they were meant to explain. The historical evidence has force because an alleged requirement lost its job under investigation. Buchner’s 1897 paper, Papineau, The Rise of Physicalism.

Call the established account D and the proposed extra source P. If D and D plus P predict an observation equally well, that observation gives P no comparative advantage. If D produces a result P was said to be necessary for, the success counts against that specific necessity claim. Levin’s causal ingression proposal therefore owes a comparison that would pick out what the nonphysical patterns contribute. The same standard applies to a claimed supernatural intervention. Universal prehension and exhaustive physicalism, when offered as interpretations of the same process, face a different question. What makes the demonstrated process evidence for one interpretation over another? Levin, Ingressing Minds, SEP on panpsychism and the intrinsic nature argument. I examine the biological comparison in Religion With A Pipette and the claim that positive accounts of causation require universal feeling in Where the Necessity Case for Whitehead’s Prehension Fails.

Put universal explanation and my alternative to work

Deutsch’s physical Church–Turing proposal concerns what a universal computer can simulate. Discovering and correcting explanations under finite time, evidence, and resources is a further capacity. Give a candidate system an unfamiliar problem. Ask it to generate a rival, use evidence to correct an error, and transfer the result. Then change what information, feedback, or external support it has access to. The performance tells us which capacity has been demonstrated, under which conditions. A well-specified counterexample can defeat a human-only claim about a particular capacity. A successful finite test supports transfer within its tested scope. Deutsch, Quantum Theory, the Church–Turing Principle and the Universal Quantum Computer.

My candidate is organizational closure. In Montévil and Mossio’s account, constraints act on flows of matter and energy in ways that help maintain one another. I want to know whether some such dependencies also help sustain error correction and transfer. Disrupt a specified dependency under matched conditions and ask whether the relevant capacity degrades. If it survives, or a simpler model accounts for the same pattern, my proposal loses support for that role. A constraint can explain how a capacity is sustained only when the capacity depends on it in the proposed way. Montévil and Mossio, Biological Organisation as Closure of Constraints, Consciousness Naturalized, including the competing models and proposed tests.

What better evidence would change

I would give a distinct causal source more credit if it predicted a result the existing account missed and evidence traced the difference to that source. I would revise a species boundary when a system on the excluded side met the same well-specified test, with its supports and limits visible. For decisions about welfare or public authority, I would ask separately what the evidence establishes about the capacity, who bears the cost of getting the judgment wrong, and what protections the available evidence justifies. No label should be able to disqualify the evidence that could correct it. What Is Truth? develops the account of evidence and public correction behind that conclusion.