The objection
CLAIM-174 makes a factor an agent at a level when it is one system
there, when a goal-directed description beats a device description by
more than the Occam cost, and when its policy would adapt if its
actions’ consequences changed. Its defeated_if names two kinds of
counterexample. One is a factor that passes the tests and is
nonetheless, “by independent marks, not an agent there”.
Boyle states, and returns to, what he calls the inscrutability paradox:
If something that looks like General AI emerges from transparent, well-understood programming and technology, that may lead us to doubt that it could have autonomous consciousness. The machine is merely doing or saying that which we have programmed it to do! (p. 30)
If we know how you were designed, if we understand how you “think” because we programmed you to do so, how can we see you as anything but a mannequin or a wind-up doll rather than an entity with free will? (p. 54)
He also states the other horn. An inscrutable system meets “fear and incomprehension” (p. 54), and “our fears may overshadow our respect” (p. 272). In the Conclusion he calls the paradox “the central catch-22” (p. 246).
If he is right, then for artifacts the marks a reader would reach for, what people and courts will grant, are not independent of the test. They track what is known about how the thing works. A system that passes CLAIM-174’s three tests and whose mechanism is published will meet “This is all just programmed!” (p. 246). That verdict would look like the first kind of counterexample, but it is evidence about the judges, not about the system. So the defeater needs marks that are not mechanism-sensitive, and for artifacts it is not clear what they would be.
What would tell them apart
CLAIM-174 can also lose an agency verdict when a mechanism becomes known, but only where knowing it makes a device description shorter than the goal-directed one. A chess engine whose code is public still has a shorter goal-directed description (it plays to win) than any account of its search. CLAIM-174 keeps it an agent. The paradox predicts that many people would not.
So the two separate on systems whose mechanism is known and whose behaviour is still best compressed as goal-directed. A study that held behaviour fixed and varied only what subjects were told about the mechanism would test the paradox. CLAIM-174 predicts no change in such a study unless the information changes which description is shorter.
What it does not say
- It does not say CLAIM-174’s test is wrong, or that agency is in the eye of the beholder. It is about evidence for defeating the claim, not the claim’s truth.
- Boyle’s paradox is about consciousness, free will and personhood, not agency as such. It reaches CLAIM-174 most directly through its willed-agent clause (CLAIM-155). Its extension to agency in the thin sense is this record’s step, not Boyle’s.
- Boyle offers the paradox as a forecast with examples from fiction, the Chinese Room debate and current chatbots. He reports no study. The warrant is interpretive.
- It leaves CLAIM-174’s second kind of defeater untouched: a clear agent, such as a person, that fails one of the tests.
Note of 2026-10-30: what holds it back
Its premise, that attributions of agency to an artifact at matched behaviour change with whether its mechanism is known, is Boyle’s forecast, and no study in the record tests it; the study its section “What would tell them apart” describes has not been read or run. Boyle’s other track also cuts against the legal half: corporations, whose workings are known, are given legal personality for efficiency and with no claim to a mind (NOTE-1014, C1 and C5), and the claim does not answer it.