Exposing Illusions – The Limits of AI by the Example of ChatGPT

The article critically analyzes modern developments in the field of artificial intelligence using the example of the ChatGPT program created by OpenAI. The idea of creating AI was expressed already in 1950 by Alan Turing who also proposed a test, the passing of which would allow us to assert that an AI was created. Defining the concept of AI faces difficulties. According to the point of view adopted here, the so-called intellectual activities allowed Homo sapiens to stand out against the surrounding animal world. With intellectual activity one no longer relies on strength and speed of movement alone. Pattern recognition, self-learning, and purposefulness of activity are not characteristic features of intelligence. The main type of human activity that is specific to humans and which – when added to pattern recognition, self-learning and purposeful activity – makes them intelligent, is conceptual thinking. namely the ability to represent things in language and use them in reasoning. Historically, there have been two main competing approaches to AI – logical and neural networks. One of the serious flaws of the neural network approach is its inability to explain the course of reasoning that leads to a particular conclusion, which makes it difficult to verify its correctness. Specific examples show that ChatGPT is not able to correctly model the simplest conceptual reasoning. The reason for this lies in fundamental limitations of the underlying large language model that cannot be corrected by additional training. Another disadvantage of ChatGPT is its susceptibility to neurohacking – forcing the user to make the necessary decisions during the dialogue. This is a serious threat to the widespread use of neural networks for decision-making in the field of management. The paper is based on research conducted in the summer of 2023.