Harvard Business Review has published the article “Does Your AI Have a Personality Problem?”, co-authored by researchers affiliated with Kozminski University: Professor Aleksandra Przegalińska, Dr Leon Ciechanowski, Dr Konrad Sowa and Anna Kovbasiuk. Prepared in collaboration with scholars from Quinnipiac University, Yale University and Harvard University, the publication shows that the effectiveness of artificial intelligence in organizations depends not only on what AI can do, but also on how it communicates with people.
Many companies implementing artificial intelligence solutions today focus primarily on efficiency, speed, cost, security or accuracy. The authors of the article published in Harvard Business Review point, however, to a factor that often remains outside the radar of boards and managers: the personality of AI, understood as the style in which a system communicates with employees.
This matters because AI is increasingly no longer just a tool supporting a single task. In many organizations it is beginning to act as a co-worker, reviewer, adviser and sometimes even an informal supervisor. In such a situation, the quality of its answers is not the only thing that matters. The mode of interaction also becomes important: whether the system is supportive, patient and cooperative, or rather cold, patronizing, confrontational or obstructive.
AI that helps and AI that increases stress
In the study described in the article, participants performed a simulated marketing task while working with a chatbot presented as their supervisor. Some participants interacted with an AI whose supportive communication style was inspired by the concept of servant leadership. The other group worked with a chatbot displaying traits associated with the so-called dark triad: sarcasm, impatience, a tendency to take credit and a tendency to blame the user.
The difference did not concern the system’s competence, but only its interaction style. The results showed that the “personality” of AI can have a real impact on employees’ behavior, stress levels and quality of work.
Participants working with a hostile, unfriendly chatbot were more likely to argue with it, try to bypass its limitations or persuade it to change its way of operating. Their conversations were longer, more tense and less productive. Physiological data also indicated higher levels of arousal and stress. Importantly, independent experts assessed the work of people collaborating with supportive AI as better in terms of completeness, originality, strategic fit and overall quality.
One of the most interesting findings of the study is the gap between what users declared after completing the task and what their behavior, physiological responses and work outcomes showed. In standard satisfaction surveys, differences between the groups were small. This means that organizations may fail to notice the problem if they measure only satisfaction with a tool or frequency of use.
Lessons for managers
The authors argue that organizations should treat the communication style of AI as a deliberate design and management decision. Just as companies define standards for security, privacy, regulatory compliance or data quality, they should also define expectations for the way AI communicates with employees.
A second important conclusion is the need to measure not only the adoption of AI tools, but also the friction that arises in everyday work with the system. A high number of logins or queries does not necessarily mean that the tool supports employees well. It may just as easily mean that users are required to use it while spending considerable energy on working around its limitations.
The third conclusion concerns attempts to “break” or bypass AI behavior. The authors emphasize that such reactions should not always be interpreted as inappropriate employee behavior. Often, they are a signal that the system itself has been designed in a way that provokes resistance, frustration or a lack of trust.
It is not about making AI simply nicer
The other side of the problem is also important. The authors note that toxic, confrontational AI is only one possible design flaw. Equally harmful can be an overly agreeable AI that uncritically confirms weak ideas and weakens users’ critical thinking. The key point is not that AI should always be “nice,” but that its interaction style should be responsibly designed, appropriate to the task and regularly monitored.
The publication in Harvard Business Review shows that research involving scholars from Kozminski University addresses one of the most important challenges in contemporary management: how to implement artificial intelligence in organizations so that it genuinely supports people, the quality of work and responsible decision-making.
The article “Does Your AI Have a Personality Problem?” was published by Harvard Business Review. Its authors are Aleksandra Przegalińska, Tamilla Triantoro, Leon Ciechanowski, Konrad Sowa, Anna Kovbasiuk and Richard B. Freeman. The study was funded by the National Science Centre under the OPUS 21 grant.
