Navigating codified, tacit and novel rules: Mapping the human-AI creativity frontier

This paper examines the boundary between human and machine creativity by analysing 593 tasks across 126 occupations in the cultural and creative industries. Theoretically, we propose an evolutionary conceptualisation of creativity, structured around three rule types corresponding to retention (codified), adoption (tacit), and origination (novel) phases. Empirically, using GPT-4, we generate synthetic annotations of the semantic content of task descriptions in the Australian Skills Classification. We derive indicators of cognitive and behavioural rules within tasks and their carriers (human, AI, or hybrid human-AI) to capture creativity, and indicators of AI autonomy feasibility and efficiency potential to capture GenAI exposure scenarios. The agent-rule matching suggests a specialisation of agents in carrying defined, tacit, or novel rules, as well as two mechanisms. The first is the structured novelty effect, whereby AI autonomy feasibility is higher when defined cognitive rules combine with novel rule creation. The second is the tacit knowledge boundary, whereby tacit rules are negatively associated with AI autonomy feasibility. In our interpretation, these mechanisms reflect an efficacy logic (matching the right agent to each rule type) that differs from an efficiency logic (optimising technically feasible gain potentials). Combining both logics, we derive a simulated classification of tasks into three categories. The mixed category, involving hybrid carrier configurations, predominates (86.3%), alongside limited replacement (2.7%) and AI immune (11.0%) categories. We identify two levels of human-AI complementarity. The first lies in rule types within tasks and engages different forms of creativity (combinatorial, exploratory, transformational). The second concerns efficiency potential when performing tasks.