Source-linked AI summary
Measuring the "Interaction Gap" in Drama Therapy with AI
Sora Kang
TL;DR
The paper questions whether AI elicits a more authentic self, noting that AI- and human-directed performances are shaped by different social pressures. It proposes the Interaction Gap as a drama-therapy lens for interpreting these differences and leaves task design, measurement, and ethical positioning open.
Problem
AI-directed speech is itself a performance shaped by social pressures, so comparing it with human-directed speech may provide diagnostic information beyond asking which context is more truthful.
Method
The paper defines the Interaction Gap as within-patient differences in verbal and non-verbal self-presentation across identical drama-therapy tasks with AI and human partners.
Results
The paper proposes the Interaction Gap as a diagnostic lens grounded in drama therapy’s use of role and aesthetic distance.
Takeaways & Limitations
The lens reframes AI as a comparable interaction partner and identifies task design, measurement signals, and ethical positioning as directions for further exploration.
Takeaways & Limitations
Patient-facing interpretation may risk mistaken self-diagnoses, while measurement creates observer-effect, surveillance, and privacy concerns.
Abstract
from arXiv · showhide
Generative AI is increasingly being introduced into expressive arts therapy, where it is often credited with offering a non-judgmental environment that supports psychological safety. Existing HCI work has largely positioned AI as a co-creative material or as a bridge/mediator into human-led care. This paper explores a different position. When a patient performs the same drama therapy task with an AI partner and with a human partner, the resulting self-presentations tend to differ in patterned ways. We propose treating this difference, the Interaction Gap, as a diagnostic lens within drama therapy. Rather than asking which context elicits a truer self, the lens reads the difference between the two performances as information about the social pressures shaping self-expression in each context. We sketch a starting point for task design and measurement signals grounded in drama therapy's existing use of role and aesthetic distance, and raise provocations for workshop discussion: the observer effect and privacy paradox that measurement introduces, and the question of whose lens the gap is.
1 Introduction
The paper questions whether AI elicits a more authentic self, arguing instead that AI- and human-directed performances reflect different social pressures. It names their systematic difference the Interaction Gap and proposes it as a diagnostic lens for drama therapy.
- AI-directed speech is itself a performance shaped by social pressures that differ across users and contexts.
- The Interaction Gap treats differences between AI- and human-directed self-presentations as diagnostic information rather than evidence that one context is more truthful.
- The lens extends drama therapy’s role-centered and aesthetic-distance tradition by comparing the same task with AI and human partners.
- The paper contributes a gap-as-lens framing, task-design considerations, candidate measurement signals, and provocations about observer effects, privacy, and whose lens the gap represents.
2 Background
Prior work commonly treats AI as expressive material, co-creator, or bridge to human care, while related studies document systematic differences between AI and human interaction. Drama therapy contributes role and aesthetic distance as the conceptual basis for comparing these contexts.
- Generative AI has entered expressive arts therapy as expressive material and as a co-creative collaborator in art-based and storymaking practices.
- Conversational studies report that people may disclose more to virtual interviewers and that chatbots and therapists adopt different communicative postures.
- Other systems position AI as a mediator or bridge supporting self-disclosure before human-led mental health care.
- Related theater work uses LLMs for improvisation, scriptwriting, and actor reflection, but these systems are not therapeutic settings.
- Drama therapy’s aesthetic-distance mechanism places therapeutic insight between affective overwhelm and detached intellectual analysis, using role-mediated expression to soften defenses.
3 Motivation
Prior actor-training interviews motivated the paper by describing AI interactions as less socially frictional than human collaboration. These accounts suggest a tendency, not a uniform effect, and come from actors rather than patients.
- Participants in prior actor-training work described AI interactions as different in texture from interactions with human collaborators.
- Participants associated AI with less pressure to prevail logically, manage others’ reactions, or perform an idealized version of themselves.
- The authors characterize this pattern as a tendency toward reduced social friction rather than a uniform effect across users.
- Because the motivating participants were actors rather than patients, their role-play experiences involve different stakes and vulnerabilities.
4 The Interaction Gap as a Diagnostic Lens
The Interaction Gap is the within-patient difference in self-presentation when the same drama therapy task is performed with AI and human partners. The proposal holds role-mediated distance roughly constant while varying the audience, but remains an undeveloped lens requiring operationalization and clinical validation.
- 4.1 The Gap as Systematic Divergence: The Interaction Gap compares a patient’s consistent self-presentation across the same task with an AI partner and with a human partner.
- 4.1 The Gap as Systematic Divergence: Self-presentation includes verbal signals such as disclosure depth and hedging, alongside non-verbal signals such as hesitation, prosody, gesture, and bodily spatial use.
- 4.1 The Gap as Systematic Divergence: The gap may differ across individuals but is expected to show patterned consistency within a given patient across sessions.
- 4.2 Why Drama Therapy: Using the same role in front of two audience types makes visible how the audience shapes role-mediated self-presentation.
- 4.2 Why Drama Therapy: The proposal remains a lens rather than a fixed framework, requiring measurement operationalization, clinical-population validation, and integration into therapeutic practice.
5 Design Considerations
The proposed task design keeps the drama-therapy role constant while comparing performance with AI and human partners. Measurement should cover verbal, pragmatic, and embodied signals, while preserving de-roling and interpreting divergence directionally.
- Task conditions: Patients perform through fictional characters rather than directly describing personal experience, using role distance to soften defenses around disclosure.The protective frame positions utterances as the character’s lines rather than the patient’s own.
- Task conditions: The AI and human partners should engage patients in the same role and scenario, with counterbalancing and temporal spacing to reduce order and learning effects.Minor scenario variations may also be considered.
- Task conditions: Comparable sessions should support voice and visual presence because text-only AI collapses the multi-modal signal space of drama therapy.Parity is needed to measure signals beyond verbal behavior.
- Task conditions: Both sessions should include de-roling, with the AI system designing this transition into the interaction rather than treating it as an afterthought.This follows drama therapy’s ethical requirement that patients exit the role at session end.
- Measurement signals: Candidate gap measures span lexical, pragmatic, and multi-modal signals, including self-reference, disclosure depth, topic avoidance, prosody, gesture, and spatial proxemics.Lexical signals are text-limited, pragmatic signals require qualitative coding, and multi-modal parity remains a design question.
- Measurement signals: The framework treats either direction of divergence as informative by comparing distributions across AI and human contexts rather than assuming one context is preferable.For example, greater hedging with either partner can constitute a measurable gap.
6 Discussion
The proposal raises an observer-effect and privacy paradox: analyzing AI sessions may reintroduce the social friction the AI context was meant to defer. It also leaves open whether gap analysis should primarily serve therapists or patients, since each orientation entails different system and safety requirements.
- The Observer Effect and the Privacy Paradox: Telling patients that AI conversations will be analyzed by therapists may make the AI feel like surveillance and tighten defenses, collapsing the gap.The measurement process can therefore alter the interaction it seeks to compare.
- The Observer Effect and the Privacy Paradox: Possible responses include asymmetric data flow that exposes only derived gap indicators or explicit honesty about measurement, with each trading measurement richness against privacy preservation.The paper frames this trade-off as a central design tension.
- Whose Lens?: A therapist-facing lens may support deeper clinical understanding, whereas a patient-facing lens may function as a reflective mirror across contexts.The two orientations require different data flow, visualization, consent, and timing.
- Whose Lens?: Therapist-facing designs raise surveillance concerns, while patient-facing designs raise risks of mistaken self-diagnosis and require guidance for interpreting one’s own gap.Patient-facing use may align more closely with expressive arts therapy’s self-knowledge orientation.
- Whose Lens?: In either orientation, the lens is intended for reflection within therapy rather than use as a phenotyping instrument that reduces patients to diagnostic labels.
7 Conclusion
The paper proposes the Interaction Gap as a drama-therapy lens that interprets systematic differences between AI and human interactions as information about the social pressures patients navigate across contexts. Task design, measurement, and ethical positioning remain open for development and discussion.
- The Interaction Gap treats systematic differences between AI and human interactions as information about the social pressures shaping patients’ self-presentation.The paper leaves the lens’s task design, measurement, and ethical boundaries open for further exploration.