Race with the machine: can Artificial Intelligence help managers fight their limits?

Abstract

Intelligence, whether biological or artificial, is fundamentally constrained by energy. This paper explores the parallel energy taxes paid by human evolution and modern Artificial Intelligence (AI), arguing that both systems are shaped by the thermodynamic need to optimize information processing within strict metabolic limits. While the human brain solved this constraint by evolving energy-saving heuristics, manifesting today as cognitive biases, AI is currently overcoming its energy bottlenecks through massive scaling and architectural efficiency.

Managerial decision making is persistently vulnerable to cognitive bias, yet many debiasing interventions remain modest in impact because they target surface-level judgment errors rather than the structural constraints that make heuristics metabolically and organizationally attractive. This conceptual paper reframes cognitive bias as a constraint-induced default emerging from (i) evolutionary canalization of survival-oriented responses and (ii) the efficiency topology of neural networks that favors low-cost, fast, “good-enough” inference over globally optimal computation. In parallel, contemporary artificial intelligence systems face their own energy and scaling constraints, but they also offer a new design opportunity: AI can function as a metacognitive scaffold that externalizes and stress-tests managerial assumptions through structured dialogue. Building on discourse-oriented strategy research, we argue that AI is most valuable not as an oracle that replaces judgment, but as a dialogical partner that systematically generates counter-arguments, alternative stakeholder perspectives, and probabilistic scenario distributions to induce reflective “System 2” pauses in high-stakes workflows. Because AI can also introduce automation bias, hallucinations, and training-data bias, effective debiasing must be treated as a socio-technical design problem involving transparency, contestability, and governance. We conclude by outlining managerial design principles and a research agenda for testing when and how AI-supported dialogue improves decision quality under uncertainty.