AI as a Sparring Partner

How critical thinking improves collaboration with LLMs

Large language models such as ChatGPT can make knowledge and ideas readily available – but good results do not come about automatically. A recent study in the renowned 'Journal of Management Information Systems' examines how people can improve the quality of their collaboration with large language models (LLMs) through conscious reflection. We spoke to the author, Prof. Dr Martin Adam, about the key findings. He holds the Chair of Smart Services at the Faculty of Business and Economics.

Mr Adam, what was the central question of your study?

My colleagues and I were interested in how people can work with LLMs without either blindly trusting them or hastily rejecting their assistance. Both extremes are problematic. That is why we regard reflection as a key mechanism: users should consciously compare what they themselves know about a topic with the LLM’s responses and actively question any contradictions or new perspectives.

What form of collaboration leads to the best results?

The combination of dialogical thinking and adversarial interaction with the LLM is particularly effective. Dialogical thinking means not only being able to make minor adjustments to one’s own assumptions, but also to question them fundamentally. At the same time, the LLM should not simply agree, but should provide counter-arguments, highlight weaknesses and offer alternative perspectives. In our findings, it is precisely this combination that leads to a higher factual quality of the work outputs.

What mistakes do users often make when using generative AI?

A common problem is that convincingly worded answers are accepted too readily. As a result, people hand over part of their own cognitive responsibility to the system. We are talking here about the importance of cognitive primacy: people should retain mental control. An LLM can provide new information and perspectives, but the task of evaluating and contextualising them should remain with humans.

What does this mean in practical terms for businesses and universities?

We should view LLMs less as all-knowing assistants and more as sparring partners. This can also be put into practice quite easily: rather than simply asking for a solution, one can ask the model to critique one’s own arguments, take opposing viewpoints, or identify potential errors.

For organisations, this means not only providing access to generative AI, but also building the skills needed to use it in a reflective manner. The aim should not be to replace human thinking with AI, but to enhance it through productive challenge.

What is your most important recommendation for working with LLMs?

Make the most of AI’s strengths, but keep yourself in the intellectual driving seat. Effective human-AI collaboration does not arise simply because the system agrees with us as often as possible. It arises when humans and AI challenge one another.

Thank you very much for the interview!

Publication: "How Reflection Enhances Task Factuality in the Use of Large Language Models"