Uncertainty around AI takes different shapes and forms, but we can boil it down to three main categories that every manager should consider: state, effect and response uncertainty. State uncertainty occurs when managers lack sufficient information to predict market trends and changes. Managers experiencing this type of uncertainty face challenges in understanding AI’s current capabilities and potential developments. As AI evolves rapidly, it is difficult to distinguish between what is achievable now and what remains a distant possibility. This uncertainty is even more daunting because AI experts often have wildly different views on critical questions — like whether scaling AI has limits, whether AI confabulations can be fixed, or whether AI can ever truly reason. Effect uncertainty describes the difficulty managers have in predicting AI’s impact on business. Will AI disrupt your industry, or will it be just another tool? This uncertainty is compounded by the fact that current tests focus on narrow benchmarks that lack real-world relevance. As such, even developers do not know how enhancements — like an increased context window in upcoming models—will affect business outcomes or employee dynamics. Response uncertainty is the challenge for managers in determining how to react and the consequences of these actions to the many uncertainties surrounding AI. Should you take the leap as an early adopter, or is it wiser to wait and observe? Should your focus be on automation and cost-cutting, or augmenting human capabilities? This uncertainty extends to choices about models and approaches — whether to develop custom models, fine-tune existing ones, or integrate techniques like retrieval-augmented generation (RAG).
