When AI Starts to Think for Itself: What Leaders Need to Know

When AI Starts to Think for Itself: What Leaders Need to Know AI is no longer limited to the predefined rules of early expert systems, nor merely to the powerful but often puzzling and unpredictable intuition-like outputs of generative models. In the summer of 2025, some commercial systems began to implement structured reasoning: the ability to work through multi-step problems, generate intermediate conclusions, and revise them along the way. After decades of limited progress toward machine reasoning, this shift is striking. One way to understand it is through the lens of early 20th-century philosopher Ludwig Wittgenstein. He argued that reasoning is not a private mental process governed by formal internal rules, but a “language game” shaped by socially learned norms of justification. From this perspective, AI systems do not “discover” logic internally; they learn what counts as valid reasoning from training data and feedback. In that sense, reasoning models mirror and operationalize human standards of argument and inference. In this session, Professor Miguel Lobo traces the evolution of computation and AI from rule-based systems that execute explicit logic step by step, to machine learning models, to reasoning models that extend this architecture to deliberative, multi-step problem solving. Finally, we consider the economic implications. As models become more capable of reasoning, compute requirements rise not only during training but also during deployment. The result is more constrained access to frontier capabilities, and a shifting competitive landscape that will determine which business models are viable. For leaders deploying AI, managing teams that rely on it, or making strategic technology investments, understanding these conceptual and technical shifts is essential to making informed decisions.