Companies are increasingly bolder in implementing AI at work. The effect is not always positive
The pressure is high – in many companies, the use of AI-based tools is becoming an expected standard of everyday work. The problem is that decisions about implementing AI are often made too hastily. New research shows that working closely with AI can be counterproductive.
The study was conducted by Dr. Leon Ciechanowski and prof. Aleksandra Przegalińska from the Leon Koźmiński Academy. They invited 221 employees from Germany and the Netherlands representing various industries – from education and health care to IT – to participate in the experiment. Everyone worked on business ideas with the support of the same AI model (GPT‑4o). The key difference, however, was not the technology, but the way of working with it.
One group used AI as a classic, “silent” assistant. The second one worked with a chatbot designed as an active partner in the creative process. The finished concepts were then assessed by independent experts.
— Technologically, it was the same engine. The difference was only in how close AI was to people in the process, emphasizes Dr. Leon Ciechanowski in an interview with Business Insider.
AI: assistant vs. partner
What did the study show? In assistant mode, the system did not impose itself on the user and responded mainly to commands. The partner acted differently: he showed his own initiative, was able to interrupt, suggest a different line of thinking and sent messages on his own from time to time.
Language was also important. The assistant kept his distance, while the partner used inclusive forms such as “we” – “let's do”, “let's check”. As Dr. Ciechanowski explains, this treatment builds a sense of close relationship – like with a co-worker. And it has consequences.
More ideas, but…
The conclusions of the study may disappoint many managers. Although the affiliate model seems advanced, very close cooperation with AI may be counterproductive. Where the AI was “closer”, participants generated more ideas. The problem arose later when it was necessary to select the best concepts and plan their implementation.
— I didn't expect that such a small change in the way we cooperate would bring such big differences – admits Dr. Ciechanowski. — People working with a chatbot partner generated more ideas, but were worse at selecting them and describing their implementation, he adds.
The moment when cooperation with AI breaks down
It is at the stage of selecting solutions and their implementation that the partnership model most often fails. Participants who worked with AI as a “silent” assistant maintained greater distance and sober judgment than those who treated the chatbot as an active partner. Why this difference? The authors of the study indicate two mechanisms.
The first one is cognitive overload. — Selection and implementation of ideas are tasks that require long, continuous thinking: weighing pros and cons, comparing scenarios, planning next steps. Such a chatbot-partner fragments attention. You have to simultaneously analyze the idea and react to what the system does, explains the researcher.
Read also: AI leads to the abandonment of logical thinking – scientists have no illusions
The second factor is dilution of responsibility. In classic teams, the more people involved in the decision, the easier it is to feel that it's not just me. A similar effect occurs when AI becomes a partner: we transfer some of the responsibility to the tool.
— In cooperation with AI, a situation may arise in which people they subconsciously turn off critical thinking, giving part of the decision-making to the machine – says Dr. Ciechanowski. Although formally legal responsibility still rests with the person, psychologically it begins to be shared. As a result, the quality of selecting the best ideas decreases significantly.
— A brilliant idea that is not selected or properly implemented will not bring any benefits to the organization – adds the co-author of the study.
When cooperation with AI does not harm
However, the picture is not clear. Although the overall conclusions of the study suggest that the “further” assistant model is safer for the quality of decisionsDr. Ciechanowski indicates specific situations in which AI as a proactive partner can bring benefits or lose its negative impact.
It is important originality of ideas. When participants developed very original concepts, the negative effect of close cooperation with AI disappeared. Even with the partnership model, the selection stage then ran smoothly and implementation problems were much less severe.
The reason is simple: the more original the idea, the more it engages a person – forcing him to leave the autopilot mode and return to deep thinking. In such situations, the focus is again on the content and not on the interaction with the system itself.
The second element is how to use the tool. When users treated AI only as a source of information and analytical support, and not as a “co-decision maker”, the quality of implementation plans was higher regardless of the cooperation model, both in the partner and assistant versions.
Read also: I was one of the first victims of the AI mush tsunami. Here's my reaction
Technology is not to blame
The effectiveness of cooperation with AI depends not on the technology itself, but on the user's characteristics and attitudes.
— My observations – both at work and with students – show that AI does not equalize competence differences, and sometimes even deepens them. – emphasizes Dr. Ciechanowski. — People who coped poorly without chatbots often become even worse when using them. In turn, those who were ambitious and had a solid foundation were able to use these tools to achieve even better results.
— The difference is in the way of working. Some treat AI as a quick source of ready-made answers, “copy-paste-forget”. Others use it as a thinking tool: they analyze, try to understand, modify. As a result, their works – whether diploma or professional – are of better quality, and they learn more and faster. In this sense, chatbots are simply tools. You can use them well or badly – a bit like using a knife – he adds.
They are key here criticism and resistance to “cognitive laziness” — consciously resisting the temptation to give all the thinking effort to the machine. Humility towards technology is also important – understanding that artificial intelligence can be fallible, hallucinates and can generate erroneous code, which requires systematic verification of results.
Read also: AI is conquering the Polish IT sector. Salaries are rising at a dizzying pace
How to use AI wisely? Conclusions for organizations and employees
It is better to treat AI as an assistant than as a partner. — A chatbot can be great at supporting us in simple, repetitive tasks, but it is not yet a good partner for co-deciding on key implementations, the researcher concludes.
It's worth it consciously design “AI-free” moments. Research shows that when AI support suddenly disappears, users accustomed to instant help give up more quickly. Therefore, it is worth leaving people space for independent thinking – design sessions without a chatbottime to review ideas without prompts or development tasks that deliberately need to be completed without technological support.
Finally – and this may be uncomfortable but necessary – we need to talk openly about cognitive laziness. Our brains naturally strive to minimize effort, and AI makes this much easier.
Dr. Leon Ciechanowski emphasizes that this is only the beginning of research on how various configurations of cooperation with AI affect the innovation process. This is a good moment to ask yourself – as an employee, manager, organization – a simple question: why do we need this technology and how to use it wisely enough so as not to get used to the fact that for every more difficult task there is always “some prompt”
