Interactive AI-Generated Scriptwriting for Theatrical Productions Using Real-Time Audience Feedback

Thesis Abstract

This study investigates the integration of artificial intelligence (AI)-generated scriptwriting with real-time audience feedback to enhance theatrical production processes and audience engagement. The problem addressed is the limited capacity of traditional script development to adapt dynamically to live audience responses, potentially constraining dramaturgical relevance and affective impact. The aim is to develop and evaluate an interactive AI-assisted scripting framework that leverages real-time audience data to influence narrative progression, character development, and dialogue. Specific objectives include (1) designing a modular AI system that combines rule-based dramaturgic constraints with neural language models to generate scene variations; (2) implementing real-time feedback channels (voting, sentiment cues, biometric proxies) to steer script generation during rehearsals; (3) evaluating dramaturgical quality, coherence, and audience-affect resonance across multiple theatrical genres; (4) assessing production efficiency, cost implications, and creative agency for writers and directors; and (5) articulating ethical and methodological implications of AI-driven dramaturgy. The methodology adopts a mixed-methods, multi-site design anchored in constructivist epistemology and informed by activity theory. The population comprises professional playwrights, dramaturgs, and directors (n=60) and theatre audiences (n=900) across three contemporary productions in metropolitan theatres over two performance cycles. A stratified sampling approach yields 20 practitioners per site and 300 audience members per cycle. Data collection instruments include (i) an AI-assisted scripting platform with logging of all AI-generated variants and associated dramaturgical decisions; (ii) a standardized dramaturgical quality rubric assessing coherence, originality, emotional arc, and character consistency; (iii) audience response measures comprising real-time interaction logs, post-performance questionnaires (Likert scales and open-ended items), and biometric proxies (heart rate variability) for a subsample (n=120) to capture physiological arousal; (iv) semi-structured interviews with practitioners (n=30) to illuminate creative processes and perceived constraints; and (v) archival production metrics (rehearsal hours, changes to script, and run-time). Validity and reliability are addressed via triangulation, inter-rater reliability (Cohen’s kappa > 0.75) on the quality rubric, and pilot testing of instruments. Data analysis integrates quantitative and qualitative methods. Descriptive statistics characterize audience engagement and production metrics. Inferential analyses include repeated-measures ANOVA to compare audience affect and narrative satisfaction across AI-generated variants, and multiple regression to identify predictors of perceived dramaturgical quality. Content analysis and thematic analysis (per Braun and Clarke) of practitioner interviews and open-ended survey responses illuminate themes related to creativity, control, and ethical considerations. A thematic coding framework is iteratively refined through researcher triangulation. System performance is evaluated through precision and recall metrics for the AI generator against a curated ground-truth corpus, and ablation studies assess the contribution of real-time feedback channels to script adaptation. Expected findings indicate that real-time audience feedback materially influences AI-generated dramaturgic decisions, yielding higher audience-rated coherence and emotional engagement in AI-guided scenes compared with control scripts. The analysis anticipates measurable reductions in rehearsal time required to converge on effective variants and increased perceived agency among writers and directors. The study also expects nuanced ethical insights into authenticity, authorship, and audience manipulation risks, informed by Bell and Hume’s ethics of computational creativity and McLuhanian notions of media dramaturgy. The theoretical contribution positions the work within activity theory and socio-technical systems, extending Slater and Bannon’s concept of participatory design to live performance contexts, and integrates Baumgartner’s affective computing framework to interpret biometric responses. The research advances knowledge by providing a rigorously evaluated, scalable framework for AI-assisted dramaturgy that aligns machine generation with live audience reception, offering practical guidelines for rehearsal workflows, governance of AI interventions in creative teams, and criteria for assessing dramaturgical quality in AI-augmented theatre. Recommendations address governance, transparency, and sustainability, including the development of domain-specific safety filters, disclosure norms for AI involvement, and professional development curricula for playwrights and directors to competently harness AI-enabled script generation without compromising artistic integrity.