AI Improved the Output. Did It Train the Person?
Imagine a company giving everyone an AI assistant. The reports improve. More difficult tasks become possible.
The next presentation calls it "upskilling".
Perhaps. But equipping people and educating them are not the same achievement.
A study published on 26 October in Nature Human Behaviour tested this distinction with 487 management consultants. Participants given access to ChatGPT and training in its use performed better on three types of data science task outside their usual skills.
Yet the researchers found no evidence that this group answered technical questions better without ChatGPT afterwards. Better assisted performance did not establish better underlying knowledge.
The experiment used GPT-4 in spring 2024. It was not a test of prolonged learning or today's models. Nor does it show that AI makes people less capable.
I am not arguing for a return to working without tools. A person plus AI can be a genuinely more capable working unit. Learning to work well with that system matters.
But if the promise is professional development, output quality alone is an incomplete measure.
A crane operator does not need to lift a beam by hand. I would still expect them to understand the load, recognise unsafe conditions and know when to stop.
The relevant question is not whether a person can reproduce everything the machine does. It is whether they can judge the work they are now able to produce.
For me, three outcomes deserve separate attention: better performance with AI, skill in using AI, and understanding that transfers to unfamiliar situations.
If training is the goal, I would build practice around that distinction.
Ask the person to explain a choice before seeing the model's explanation. Change an important condition and ask what must change in the solution. Introduce a plausible error and see whether they notice it. Revisit the task later.
Use AI to support those exercises, not simply to finish them.
None of this means turning every routine job into a classroom. If the objective is delivery, useful assistance may be enough. But buying capability and developing expertise are different investments.
A company can become more productive without having demonstrated that its employees learned the underlying subject.
That is not a failure of the tool. It is a reason to be precise about what was achieved.
Before announcing that we have trained the workforce, we should check that we have done more than equip it.
Research: Wiles et al., Nature Human Behaviour (2026). DOI: 10.1038/s41562-026-02566-6