Generative AI is fragmenting Kenya's performing-arts labour: veteran artists withdraw from recorded/digital spaces to protect communal affect, while younger practitioners adopt hybrid workflows that offload structural tasks to AI but preserve relational performance, risking unequal market capture and undervaluation of non-Western affective labour.

Main Finding

Generative AI introduces an experiential and ontological rupture in East African affective/performative labour: veteran practitioners perceive current Gen-AI as context-blind, inauthentic, and "robotic" (vernacular: “plastic”), prompting withdrawal to unrecorded, offline spaces; younger practitioners adopt a pragmatic 50/50 hybrid workflow that uses Gen-AI for structural tasks while safeguarding relational, intuitive affects. Participants describe an experiential “algorithmic reaction gap” — a persistent mismatch between machine-inferred emotional output and the relational affective cues it should respond to. The paper treats this as a report of lived experience, not a claim about theoretical limits of machine inference. Protecting African performance labour requires moving beyond Western discrete-emotion frameworks and designing governance and technology that respect collective relational epistemologies (e.g., Kibagenge, Utu).

Key Points

• Definitions and frame
  • Affective labour: relational and emotional dimensions of creative/performative work (Hochschild; Hardt).
  • Decolonial lens: critiques West-centric affective computing and epistemic mismatch with East African relational constructs.
  • Kibagenge and Utu: two interlocking East African relational constructs central to Kenyan performance contexts (collective relational norms and mutual personhood).
• Generational divide
  • Veterans: view Gen-AI as context-blind, “plastic” (inauthentic), and “robotic”; intentionally avoid recording and use offline spaces to preserve communal affect.
  • Younger, digitally native practitioners: pragmatic hybridization — about 50% of workflow handled by Gen-AI (structural tasks) and 50% retained for human relational/affective work.
• Conceptual contribution
  • Algorithmic reaction gap: participants’ term for the ontological/affective mismatch between machine responses and the lived emotional cues; experienced as unbridgeable with current systems.
  • Synthetic Empathy: term used to describe algorithmically generated affective responses that participants find shallow or inauthentic.
• Epistemic critique
  • Western affective computing relies on discrete-emotion models that do not map to collective, relational epistemologies like Kibagenge and Utu.
  • Current Gen-AI development and governance risk eroding communal practices and undervaluing affective labour rooted in non-Western relational norms.
• Empirical stance
  • Findings are descriptive of practitioners’ experiences and strategies; the paper does not assert impossibility of future technical solutions.

Data & Methods

• Sample: 16 Kenyan performing-arts practitioners spanning a generational range (veterans and younger digital natives).
• Approach: Qualitative, thematic analysis of interviews/focus groups using a decolonial analytic frame to interrogate Western-centric affective computing assumptions.
• Evidence types: practitioners’ narratives about workflows, trust, authenticity, use/avoidance of recording and Gen-AI tools, and vernacular terms (e.g., “plastic”) that capture affective evaluations.
• Analytic focus: how practitioners navigate Synthetic Empathy, the emergence of hybrid workflows, and the cultural epistemologies (Kibagenge, Utu) shaping those choices.
• Limitations: small, purposive sample focused on Kenya/East Africa; exploratory and interpretive—findings reflect experienced phenomena, not quantitative generalizability.

Implications for AI Economics

• Labour markets & task composition
  • Skill-biased technological change may be nuanced: Gen-AI automates structural creative tasks but leaves relational affective tasks resistant to automation, potentially creating hybridized job profiles rather than wholesale displacement.
  • Divergent adoption across generations can produce intra-sectoral inequality: veterans may withdraw from monetizable (recorded/digital) markets, while younger workers capture hybrid productivity gains.
• Valuation of affective labour
  • Market prices and platforms may undervalue collective, relational affective work when trained on or optimized for Western discrete-emotion proxies, risking wage suppression for affect-driven creators in non-Western contexts.
  • Need for new valuation metrics that capture collective relational value (e.g., community co-creation, offline performative worth).
• Intellectual property, consent, and data governance
  • Recording-avoidance strategies reduce available training data from non-Western affective practices, potentially biasing future models and increasing platform dependence on Western data.
  • Governance must address consent, community rights, and cultural appropriation — standard individual-focused consent regimes are insufficient for communal epistemologies like Kibagenge/Utu.
• Productivity, markets, and cultural industries
  • Hybrid workflows can raise productivity for certain tasks (e.g., editing, scaffolding), but the non-substitutability of affective labour limits full automation-driven productivity gains for cultural outputs that rely on communal presence.
  • Platform and aggregator models may capture surplus from structural-task automation, increasing market concentration unless governance or collective bargaining protects creators.
• Policy and design recommendations relevant to AI economics
  • Design incentives and procurement standards to include non-Western affective datasets and relational labels; fund participatory model-building with community epistemologies.
  • Develop governance frameworks that recognize collectivized data/affect rights (beyond individual consent), and protect unrecorded/offline cultural spaces where affective labour is performed.
  • Support alternative remuneration and valuation mechanisms (e.g., community-based royalties, public subsidies for cultural preservation) to avoid market-driven erosion of relational labour value.
  • Invest in capacity-building for practitioners to negotiate hybrid workflows and in regulatory tools to limit exploitative extraction of affective data.
• Broader economic research directions
  • Quantify how hybridization affects earnings, bargaining power, and market structure in cultural industries across age cohorts.
  • Model long-term dynamic effects of underrepresented affective data on global model performance and resulting distributional outcomes for non-Western creators.
  • Explore metrics to internalize cultural externalities and ecosystem services provided by communal affective labour.

(Interpretation and recommendations are drawn from the qualitative findings and framed as implications for AI economics and governance, acknowledging the study’s exploratory scope and contextual focus on Kenya/East Africa.)