Convergence Point: Wu & He and the Difference You Cannot See
The outputs match. The mechanism does not.
The Dates Are Just Terrible.
This one happened during exactly the week Modal Path Ethics was trying to stop saying stupid shit about loci.
- 3 November: Peter van Inwagen and the Ball Of It All separates material composition from organized continuation. A locus does not need to be an additional material substance sitting above its constituents.
- 5 November: Linqian Wu and Diaoxuan He submit Same Behavior, Different Mechanism: What Finite Experiments Can Certify About Integrated Information.
- 8 November: Modal Path Ethics Gets a Paper Route tightens nonredundancy. A complete finer-grained account may fully realize an institution, provided it preserves the historically integrated organization whose activity is being examined.
- 9 November: Amartya Sen, Martha Nussbaum, and Modal Path Ethics states the resulting position plainly: a bearer can be relationally constituted, describable at finer grains, and still undergo an impairment attributable to its own organized continuation.
Turns out Wu and He have produced two systems that behave exactly the same.
And underneath that behavior, they are different.
The Same Thing Happens.
Wu and He begin from a simple problem in mechanism-level measurement.
Suppose you care about how a system computes, rather than only what comes out.
Can behavior tell you?
Sometimes, absolutely the fuck not.
Their cleanest construction compares a recurrent four-unit network with a feed-forward network. At the tested response delay, the two systems have exactly the same input-output law from every initial state.
Watch them from outside forever. You learn nothing that distinguishes them.
Internally, however, they have different causal organizations. Under the integrated-information quantity Wu and He study, one has Φ=0.28 bits and the other has Φ=0. Record an internal unit and the tie begins to break within a small number of trials.
That is already enough to cause me trouble.
Because Modal Path Ethics has spent months asking whether an organization survives, whether a replacement preserves it, and whether a higher-level bearer is doing explanatory work unavailable from an inventory of its parts considered separately.
The easy test would be:
Does it still behave the same way?
Except Wu and He have just removed that option.
Van Inwagen Already Took Away My Ball.
Three days earlier, Peter van Inwagen had forced a different concession.
Modal Path Ethics does not need to prove that every locus is an additional composite material object.
A hospital can be completely realized through people, records, equipment, permissions, practices, and their relations.
A complete sufficiently fine description can contain all of that. Nothing ghostly needs to hover above anyone.
The locus claim concerns the organized continuation carried through those relations.
That was the whole point of the katamari.
- Break the organizing relations while leaving the constituents present and some characteristic continuations disappear.
- Exchange constituents while preserving the organization and some continuations survive.
- Name an arbitrary collection and nothing comparable happens.
The relevant question is what actual organization exists and what its disruption changes.
Paper Route then sharpened the same point against a stronger reductionist response.
Take the newsroom apart conceptually. Describe every journalist, payment provider, record, transaction, machine, permission, and coordinating relation.
If that description is complete, the institution does not need to add another physical ingredient.
But notice what the reduction had to retain in order to explain the injury:
the organization of the work.
- Which resources sustained it.
- How decisions became publications.
- How records participated in continuing practice.
- How the loss of operative support disabled that integrated activity.
The descent explains the institution's realization. It does not make the organization disappear.
That is the current nonredundancy claim.
Treating the other relevant loci separately fails to preserve the organized continuation at issue.
A sufficiently complete relational account can preserve it.
Now Wu and He make the epistemic problem harder.
What if two candidate realizations preserve the same outward behavior while differing internally?
Behavior Has Lost Jurisdiction.
Wu and He call their result a measurement problem.
Modal Path Ethics calls it instrument jurisdiction.
If the claim concerns mechanism, an instrument that sees only behavior does not automatically have jurisdiction to settle it.
More behavioral data will not help when the experimental design makes the mechanisms observationally equivalent.
Their Proposition 1 makes this exact: for the selected input-output interface, the recurrent and feed-forward systems remain indistinguishable at any sample size. Internal recording supplies evidence that the black-box experiment cannot.
Then they make it worse.
Partial internal access can also fail.
Hide one unit and wildly different values of the mechanism-level quantity can remain compatible with exactly the same observations. No amount of repeating that same inadequate experiment removes the uncertainty.
In another construction, several hidden components can only be observed through a product of their parameters. Longer observation still cannot separate them. One intervention can.
Their general lesson is:
What an experiment touches determines what it can establish.
The amount of data comes second.
Φ: Not a Locus Detector.
There is an extremely tempting way to ruin this immediately.
Do not do this:
Wu and He proved that integrated information identifies the true locus.
But no. They did not.
They explicitly decline to establish that Φ is consciousness. Their subject is whether a mechanism-level quantity can be identified and certified under finite, partial, noisy experimental access. And their results actually make a universal scalar locus test look somehow even less attractive.
Their chosen measure contains sharp boundaries. Tiny changes in underlying mechanism can cause large jumps in reported Φ when the selected partition changes. A surrogate can fit interventional behavior almost perfectly and still report the mechanism-level quantity badly near those boundaries.
Modal Path Ethics should take the warning. The locus diagnostic still requires more than integration:
- actual extance,
- historical formation,
- causal coordination,
- a warranted grain,
- characteristic successor structure,
- and nonredundancy in the relevant respect.
Kit Fine's separate problem also remains. Evidence that an organization continues does not automatically settle numerical identity through replacement.
Wu and He do not answer those questions.
They tell us something prior:
If your claim is about causal organization, your evidence has to reach causal organization.
The Newspaper Looks Different.
The Novaya Gazeta locus claim came from the actual history connecting editorial practice, records, decisions, support relations, production, circulation, adaptation, and attempted substitution. Its organization had to be independently evidenced before its injury could be attributed.
Wu and He give us a hostile test for any future replacement claim.
- Suppose a replacement newsroom produces indistinguishable outputs.
That establishes behavioral continuity.
It does not, by itself, establish preservation of the organization whose continuation mattered.
Sometimes behavioral equivalence will be enough for the practical question being asked. Sometimes it will not.
The grain and claim decide what evidence is required.
That is a much cleaner rule than declaring either behavior or mechanism universally supreme.
The Ruling.
Linqian Wu and Diaoxuan He were not trying to identify moral bearers. They were asking when finite experiments can establish facts about causal mechanism.
Modal Path Ethics was independently tightening a theory in which moral harm must attach to an actually established bearer whose organized continuation can survive constituent change, fail through relational disruption, and remain fully physically realized at finer grains.
The projects meet at one hard constraint:
Behavioral equivalence does not establish organizational equivalence when organization is the object of the claim.
Wu and He make that limitation exact.
Modal Path Ethics adds historical and ethical structure they are not trying to supply.
The best part of their paper is a workflow.
Before attempting to measure a mechanism-level property:
- declare the system,
- declare the grain,
- declare the time scale,
- declare the claim,
- ask whether the planned observations can distinguish the relevant mechanisms,
- identify what remains observationally equivalent,
- test whether the metric is stable across that uncertainty,
- then report only the claims the evidence can actually certify.
That belongs almost intact in future locus work.
Before saying:
This organization survived.
This replacement preserved the relevant structure.
or:
These two systems instantiate the same causal organization.
ask what evidence could distinguish that claim from its alternatives.
If the answer is:
Their outputs look the same,
we may not have measured the thing we are talking about.