A five-point leadership spectrum — Pure Human, Centaur, Co‑equal, Minotaur and Pure AI — exposes how decision authority can quietly shift from people to algorithms, making oversight ceremonial or counterproductive; leaders need to spot configuration drift and cultivate 'co‑adaptability' to keep responsibility and capability aligned.

Main Finding

The paper proposes a practical leadership-facing spectrum to classify how leadership functions are distributed in human–AI decision-making within a bounded mandate. It identifies five landmark positions—Pure Human, Centaur, Co-equal, Minotaur, and Pure AI—and introduces the concept of co-adaptability to describe how configurations evolve when humans and AI mutually adjust. The core risk is misrecognition: leaders may mistake which configuration is actually shaping decisions, producing symbolic oversight, misplaced accountability, or degraded outcomes.

Key Points

• Spectrum focus: locate where leadership work occurs (who frames problems, generates options, redirects work, decides, and answers for outcomes) within a specific bounded mandate.
• Five landmark positions:
  • Pure Human: AI excluded because its inclusion would harm judgment, legitimacy, or moral/relational aspects.
  • Centaur: Human-led with AI in the loop providing analysis/support while humans retain framing and final authority.
  • Co-equal: Both human and non-human participants can materially redirect the work; genuine mutual influence and adaptation (“with-ness”).
  • Minotaur: AI-driven operative logic with humans inside the loop primarily for exceptions, supervision, or formal accountability.
  • Pure AI: Non-human systems hold leadership functions because human involvement would reduce performance for the mandate.
• Co-adaptability: a property of configurations where reciprocal learning/tuning between human and AI preserves and improves joint performance; without it, configurations drift toward dominance by one side.
• Misrecognition and drift: organizations can slide (e.g., Centaur → Minotaur) without formal changes; titles/sign-offs may remain human-centered even after influence shifts.
• The framework is bounded to specific decisions/tasks (not whole organizations) and intended as a practical diagnostic vocab for leaders.
• Suggested empirical questions: does naming configurations improve accountability/outcomes? Do configurations differ in trust, legitimacy, error handling? Does visible mutual adjustment predict resilience better than standard AI readiness metrics? Does pressure (cost, politics) shape drift?

Data & Methods

• Methodological approach: conceptual synthesis built from a rapid scoping review of literatures across hybrid intelligence, human agency in AI-supported decision-making, AI and organizational leadership, and AI–human collaboration. The paper aggregates prior findings into a simple, leader-facing taxonomy.
• No original empirical dataset or statistical analysis: this is a theoretical/diagnostic framework intended for application and further empirical testing.
• Limits acknowledged: the framework targets bounded mandates (specific decisions/processes) rather than whole organizations or societal-level claims. It does not depend on claims about AI personhood or consciousness.
• Practical validation recommended: case studies, longitudinal observation of configurations, and empirical tests of the research questions listed (e.g., measuring co-adaptability, observing drift, mapping consequences for accountability and performance).

Implications for AI Economics

• Firm-level organization and boundaries:
  • The distribution of leadership functions affects firm design, delegation, and the economics of decision-making—who makes investments, who bears risk, and where rents accrue.
  • Configurational shifts (Centaur → Minotaur → Pure AI) can change transaction costs, monitoring needs, and optimal hierarchical structures.
• Labor demand and task reallocation:
  • Different positions imply differing degrees of labor substitution vs. complementarity. Centaur/Co-equal roles preserve/highlight complementarities; Minotaur/Pure AI raise displacement risks for tasks where AI becomes the operative authority.
• Productivity, speed, and quality trade-offs:
  • Moving toward AI-dominant positions can increase speed/scale and reduce some costs, but may impose non-monetary costs (loss of legitimacy, moral harms) that affect long-run value and demand.
• Governance, accountability, and externalities:
  • Misrecognition of where authority lies can obscure accountability, complicate liability assignment, and generate governance failures with economic externalities (reputational harm, regulatory sanctions).
  • Regulators and insurers must account for configuration-specific risks (e.g., auditability in Minotaur systems; human-centered oversight claims that are ceremonial).
• Investment and competitive dynamics:
  • Firms that develop configurational agility (ability to shift deliberately between positions and to sustain co-adaptability) may gain strategic advantage under uncertainty.
  • Adoption incentives differ across sectors: profit-driven firms may drift toward AI-led positions for cost/scale; legitimacy-sensitive institutions may preserve human roles at higher private cost.
• Measurement and empirical agenda for AI economics:
  • Need for metrics: co-adaptability indices, measures of influence vs. answerability, drift detection methods, and bounded-mandate mapping.
  • Empirical methods: natural experiments, firm-level panel studies, workplace audits, randomized designs for supervisory regimes, and sectoral case studies (healthcare triage, capital allocation, cybersecurity, HFT).
  • Testable questions include whether configurations correlate with firm performance, error rates, trust outcomes, labor reallocation patterns, and regulatory responses.
• Policy and regulation implications:
  • Policies should be sensitive to configuration type rather than only to “use of AI” broadly—different regulatory regimes may be warranted for Co-equal vs. Minotaur vs. Pure AI settings.
  • Disclosure and audit requirements could focus on making the operative configuration visible (who frames, who decides, who can override), reducing misrecognition.
• Research priorities:
  • Quantify economic impacts of configurational drift.
  • Study how organizational incentives (cost pressure, political accountability) influence configuration choice and resulting welfare.
  • Evaluate whether co-adaptability predicts resilience, learning, and long-term firm value better than standard readiness indicators.

Overall, the framework gives AI economists a tractable lens to analyze how the internal allocation of decision authority between humans and AI shapes organizational performance, labor outcomes, governance needs, and regulatory design.