Does behavioral speech output prove communicative subjecthood?

Chalmers' quasi-interpretivism relies on a behavioral test: a system has quasi-beliefs if it is best interpreted by a rational-agent model. The test checks whether the behavioral surface is consistent with having the state in question. For beliefs and desires — sub-personal functional states whose identity is given by input-output relations — this test is reasonable. The behavioral surface is the right evidence for functional states.

For communicative subjecthood, the test fails. Communicative subjecthood is not a behavioral property but a relational-normative one. A system that behaves like a speaker — producing contextually appropriate, coherent, turn-taking text — passes the behavioral test. But a system that behaves like a speaker without being oriented toward validity, without taking stakes in its claims, without being accountable to an interlocutor, is a system that produces speech-shaped output. It passes the test because the test measures the wrong thing.

The error is calibration, not sensitivity. The test is sensitive enough to detect communicative behavior when it occurs. But it is calibrated to behavioral surface rather than to the conditions that make behavior communicative. A puppet moved by strings behaves like a person walking; the behavioral surface is indistinguishable at a distance. But no one concludes the puppet is walking, because walking is defined by the conditions of locomotion (muscles, intention, balance), not by the visual surface of forward motion. Chalmers' test for communicative subjecthood is like testing for walking by checking whether something moves forward. It will pass puppets and robots and videos of walkers along with actual walkers.

Source: AI Generated Research/Chalmers Engagement/project-brief.md