The Question AI Can't Answer

After eight years working at the intersection of AI and professional learning, the question I'm asked most often is: "How do we use AI without losing what makes us effective?"

It's the right question. But organisations are mostly asking it too late - after deploying the tools, not before.

The ones getting it right share a common instinct: they treat AI as infrastructure for learning, not a shortcut around it. They use it to surface more problems to solve, not fewer. They push their people to engage more critically with AI outputs; questioning, testing, refining, rather than accepting and forwarding. In doing so, they're actually strengthening the cognitive habits that AI, unchecked, tends to erode.

The risk isn't that people will become dependent on AI. The risk is that they'll stop noticing they have.

What I've seen work is building visible decision points into how teams use AI; moments where a person must bring their own judgment rather than ratify the machine's. Not as friction for its own sake, but as designed practice for the skills that remain irreducibly human: contextual reasoning, ethical discernment, relationship navigation, and the willingness to sit with ambiguity long enough to understand it.

Sustainable performance was never about output volume. It was always about building people who get better over time. AI doesn't change that. It just raises the stakes for organisations that forget it.

Institute Reflection: AI, Judgment, and the Value of Staying Curious

At the Institute for Sustainable Human Performance, this perspective highlights a shift that is becoming increasingly important: the most valuable organizations may not be those that automate decisions the fastest, but those that deliberately preserve opportunities for people to think, question, and learn.

Recent research suggests that the way people interact with AI shapes not only the quality of decisions but also the development of human expertise. Lee et al. (2025) found that knowledge workers who rely heavily on generative AI report investing less effort in critical evaluation, particularly when they have higher confidence in GenAI. This raises an important question: if AI provides immediate certainty, do we lose the habit of exploring alternatives?

Rather than accepting AI outputs as final answers, people should pause, question assumptions, and contribute their own judgment. Research by Fang and Feng (2026) supports this distinction, showing that people who actively evaluate and refine AI-generated content develop deeper engagement and are less likely to become dependent on the technology than those who simply accept its outputs.

Equally important is the willingness to remain with uncertainty. Leonardi and Leavell (2026) argue that one of the hidden risks of generative AI is the creation of artificial certainty: the tendency for fluent, confident answers to make users feel that ambiguity has disappeared when it has merely been hidden. They suggest that expertise increasingly depends on the ability to keep uncertainty active long enough to explore multiple interpretations, challenge assumptions, and recognize what is still unknown. In this sense, "sitting with ambiguity" is not hesitation but a valuable cognitive practice.

From our perspective, the question AI cannot answer is often the most important one: What are we missing? Organizations that intentionally create space for curiosity, reflection, and constructive uncertainty are likely to build stronger judgment, greater adaptability, and more sustainable performance than those that simply pursue faster answers.

References

• Fang, X., & Feng, J. (2026). Research on the application behavior of generative artificial intelligence learning of college students based on self-determination theory. Frontiers in Psychology, 17, Article 1805498.

• Lee, H.-P. H., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems.

• Leonardi, P. M., & Leavell, V. A. (2026). Knowing enough to be dangerous: The problem of artificial certainty for expert authority when using AI for decision-making and planning. Organization Science.