Design AI systems that grow people and knowledge

When we introduce an artificial intelligence system into an organization, we should ask ourselves what skills we want to develop in the people who will use it. The time saved matters. It also matters what happens to knowledge, professional autonomy, and the ability to make informed decisions.

Two readings published in September 2026 offer useful perspectives for addressing this question. Working with AI, from the research group at the University of Auckland, proposes a model for designing collaboration between people and AI systems. Educating minds with generative AI, by Laura Desiree Di Paolo, Andy Clark, and Thomas Wachter, calls for reconsidering the conditions and purposes of learning. I read them as an invitation to make explicit the choices that accompany technological adoption.

Collaboration requires a project

The Auckland white paper relates five dimensions: person, AI system, task, organization, and society. It proposes evaluating success also through workers' experience and the quality of collaboration. In the illustrative case, 30 participants assemble a valve with a robot handing over the tools. They can experiment with different configurations and choose how to interact. A deliberate error by the robot reduces reported trust; in the subsequent trial, with correct behavior, trust rises again. The result concerns a limited laboratory activity: it does not demonstrate lasting effects on company productivity, nor does it allow generalization to generative systems. [1]

For those running a business, these elements suggest a concrete question: can the worker truly intervene in the process? In an AI project for customer service, for example, it is necessary to establish who verifies a response, when a case should be escalated to a person, and how to preserve what has been learned from a mistake. This is an example of design that I propose, not an application verified by the white paper.

Modern computing is social. Every decision about how a system operates affects work relationships, influences the distribution of responsibilities, and can change the value attributed to skills. For this reason, training must also support the people who define goals and processes, in addition to those who use the interface.

Evaluate competence beyond the result

Educating minds with generative AI is a perspective article subjected to scientific review, not a new experiment. The authors describe the school as an environment in which social practices, materials, and technologies contribute to shaping thought. They identify a gap between AI design and learning sciences, and a second one between measurable performance and personal development. They also question the concentration of explanation, feedback, and assessment in a single interface. The long-term cognitive and social consequences remain to be studied. [2]

The distinction also concerns adult education. A well-written paper by a generative system does not, by itself, tell us whether the person presenting it has understood the problem. In designing DKTS pathways, I consider it essential to ask the person to explain their choices and tackle new situations. The assessment must make visible the contribution of AI and that of the person using it.

Let's imagine a corporate training exercise: the system proposes a solution to an organizational problem, the participants check the premises and discuss which information is missing. Then each person tackles a different case, explaining their decision. It is a teaching proposal to be tested and evaluated, even after some time, to understand which knowledge remains available.

Knowledge and responsibility in decision-making

I often use four verbs: ask, obtain, evaluate, decide. They serve as a reminder to keep the relationship with AI systems conscious. The question already requires a choice: defining the problem, recognizing the context, and understanding what information is needed. Evaluation requires knowledge. Decision involves the responsibility of a person and the organization in which they operate.

Artificial Intelligence is a discipline. Its adoption requires a culture capable of connecting technological skills and understanding of people. In my experience, the meeting between humanistic education and technology has made it possible to conceive problems from different perspectives. It is this ability that I would like to continue to cultivate in professional training as well.

For DKTS, enhancing knowledge also means asking where the experience gained during work remains. If a system produces a summary, who verifies its sources? If an error emerges, how does it become shared knowledge? If an expert recognizes an exception, how do we make their reasoning understandable to colleagues? These are questions to ask before automating a process.

The same criterion must guide inclusion. From my point of view, access to training and the ability to influence decisions are conditions to be verified for all people, including those with less digital familiarity. The experience of those who know a trade can help evaluate what a system proposes; it must find space in design and discussion.

Experiment and verify over time

My proposal is to start from a recognizable problem and define in advance how we will evaluate the result. Alongside service times and quality, we observe the ability to recognize errors, explain choices, and handle a case without assistance when needed. A limited trial allows us to understand what to modify before extending the use of the system.

First experiment, then validate and continue to observe. Needs can change with people's experience and system updates. Evaluation criteria must also remain open to revision.

We need intelligences: human, artificial, collective, organizational, and scientific. Making them work together requires care for the conditions in which people learn and make decisions. The value of innovation is also seen here: in the knowledge that remains within organizations and in people's ability to use it responsibly.

References

[1] Yuqian Lu, Regina Lee, Rui Zhou, Lixin Jiang, Andrew McDaid, and Amy Lawrence. Working with AI: A Design Framework for Human-AI Collaboration. University of Auckland, September 2026; arXiv version deposited on September 25, 2026. White paper, peer review not indicated.
https://doi.org/10.48550/arXiv.2609.31793

[2] Laura Desiree Di Paolo, Andy Clark, and Thomas Wachter. Educating minds with generative AI. Communications Psychology, 4, 128, September 14, 2026. Perspective article peer-reviewed.
https://doi.org/10.1038/s44271-026-00522-8