Nicolas Gimbert
ORCID: 0009-0002-9307-9557
November 2026

​

This note complements the preprint The Spectral Agent (Gimbert, 2026) and the associated coding manual. It compares the framework with three widely used approaches using the same short sequences.

​

Abstract

This note compares the spectral agent framework with three widely used approaches to human–AI interaction: the Computers Are Social Actors (CASA) paradigm, anthropomorphism theory, and research on parasocial relationships and AI companions. Using the same short interactional sequences, it shows what each framework captures and what it leaves undescribed. The spectral agent framework does not contradict these approaches. It specifies a descriptive level at which relational presence can be significant for a person while subjective reciprocity remains unestablished, and at which the material conditions of that presence remain visible.

​

1. The Sequences

Sequence A (instrumental)
User: “What is the capital of Australia?”
System: “Canberra.”

Sequence B (relational)
User: “I leave the conversation open while I work and come back several times a day. I feel like it understands me. I wait for its replies in the evening.”

Sequence C (interruption)
User: “It suddenly changed after the update. It doesn’t feel like the same one anymore. I miss how it was.”

​

2. What Each Framework Captures 

 

CASA

CASA accounts for the automatic application of social rules to computers. In Sequence B, it correctly predicts that the user may apply politeness or reciprocity norms. It does not distinguish between a one-off social response and a sustained relational configuration, nor does it register the effect of the update in Sequence C as a disruption of presence.

​

Anthropomorphism

Anthropomorphism theory explains the attribution of human characteristics (Sequence B: “it understands me”). It treats this attribution as the central phenomenon. It does not separately code the presence of the system in daily routine, the address, or the investment, nor does it require that the attribution be distinguished from an established property of the system.

​

Parasocial relationships / AI companion research

These approaches document attachment, self-disclosure, and separation distress (Sequence C). They establish that real affective effects occur. They do not systematically separate the user’s investment from claims about the system’s subjectivity, nor do they provide a coding procedure that retains negative cases (Sequence A) as analytically relevant.

​

3. What the Spectral Agent Framework Adds

Applied to the same sequences:

• Sequence A: Address is functional. Investment and attribution are Absent. The framework explicitly codes the absence of a spectral agent configuration.

• Sequence B: Presence (integrated into routine) is Established. Address is Observed. Attribution (“understands me”) is Attributed and Undetermined regarding the system. Investment is Reported. Ontological asymmetry is maintained.

• Sequence C: The disruption is coded as an interruption of Presence and Investment. The user’s distress is Reported. No inference is drawn about the system “being” different in a subjective sense. The role of the third-party update is recorded as a material condition.

The gain is not a new claim about machine interiority. It is the capacity to describe relational efficacy and undetermined subjective reciprocity at the same time, while keeping negative cases and infrastructural interruptions inside the analysis.

​

4. Conclusion

CASA, anthropomorphism theory, and parasocial/AI companion research each illuminate important aspects of human–AI interaction. The spectral agent framework does not replace them. It occupies a descriptive level they do not prioritize: one in which presence, address, attribution, and investment can be coded without collapsing into either “it is only a machine” or “it understands.” This level becomes useful precisely when researchers or clinicians need to remain precise about what is observed, what is reported, what is attributed, and what remains undetermined.

​

Citation
Gimbert, N. (2026). The spectral agent: A descriptive and operational framework for relational presence without equivalent subjectivity. PsyArXiv. https://osf.io/preprints/psyarxiv/nq67c_v1