The Dunning-Kruger effect is a cognitive phenomenon in which people with little knowledge or low competence in a particular area tend to overestimate their abilities. At the same time, more competent people tend to underestimate their abilities. This effect was named after the psychologists David Dunning and Justin Kruger, who described the phenomenon in 1999.

The basic assumption is that people with little experience do not have the necessary knowledge to assess their abilities realistically. They lack the competence to recognise their mistakes, leading them to overestimate their performance. Conversely, competent people often have a better understanding of the task and tend to relativise or underestimate their abilities compared to others because they understand the complexity of the subject better.

The Dunning-Kruger effect often manifests itself in everyday situations or a work context, especially when people believe they have a complete command of a topic even though they only have a superficial understanding. The phenomenon emphasises the importance of self-reflection and feedback in developing a realistic picture of one’s own abilities.

The Dunning-Kruger effect can easily be transferred to the use of generative AI tools because many users can misjudge their abilities or the potential and limits of these technologies can be misjudged.

They may believe that ChatGPT can fully understand the emotional context of a conversation and provide appropriate responses, even if the technology is not yet mature enough to do so reliably.

On the other hand, some people are sceptical about ChatGPT and dismiss it as a gimmick or toy. They may assume that the technology is only useful for trivial tasks, such as answering quiz questions or providing weather information, and is not capable of conducting more complex or meaningful conversations.

Here are some typical examples:

1. Overconfidence with little experience: Novices who have just started working with generative AI tools may overestimate the quality and accuracy of the results. Because AI models can often generate impressively ‘human’-sounding content, these users may assume that the AI will always provide correct or useful answers. They often underestimate the need to scrutinise and critically question the results.
2. Underestimating complexity: A common phenomenon is that new users underestimate the complexity and expertise that is often required to use such tools effectively. They believe that AI can simply answer or solve ‘everything’ and that no further technical or specialised knowledge is required. The limitations of AI models or ethical implications often remain unconsidered.
3. Self-critical handling through experience: Advanced users and experts who work with generative AI tools usually develop a more realistic assessment of this technology’s possibilities and limitations.
   They know that generative AI is primarily based on large amounts of data and can be flawed or biased in certain situations. They pay attention to how they guide the AI, scrutinise the results and adapt them to their requirements. Over time, users will develop a better sense of when and how to use generative AI tools effectively.

Constant learning curve and adaptation: As generative AI models are constantly improved and updated, users need to continuously educate themselves to stay up to date. Overconfidence can inhibit the willingness to evolve and adapt to new features and limitations.

To counteract the Dunning-Kruger effect, it is essential to learn the necessary AI skills and how AI tools work. It is important to promote a culture of reflection and critical questioning. As generative AI models are constantly being improved and updated, constant learning is required.