Dhe week started with dinner in town. The guest is a top-class mathematician and computer scientist, a luminary. The topics are bubbling up, the exchange is lively as always, and at some point he is asked whether he can explain how Chat GPT’s artificial intelligence works. This mixture of chat system, search engine and language model is currently doing the rounds. She can answer questions formulated in natural language, write short texts and even program. The system is trained with data from websites, Wikipedia entries and books, and the answers seem to be amazingly intelligent. The question is already being raised in public as to how one can protect oneself from cheating and misinformation in view of the impressive performance.
* * *
If Chat GPT had been used in Germany to create the property tax return, the system would certainly have zipped through the data from the municipalities, land register and tax offices and would have collected all the information that the state already has but instructs its citizens to collect again and to present. For the fun of it.
* * *
Chat GPT was developed by the American company Open AI, in which Microsoft has already invested more than a billion dollars. Everyone is saying that artificial intelligence has reached a new level, self-appointed experts are overflowing with praise and fantasizing about the future. Chat GPT seems too good to be true. The author of these lines would have liked to know how it works that evening. However, the guest’s answer is disappointing: Hardly anyone is able to explain how it works, let alone understand which data and sources were included when and how in the respective answer of the software.
* * *
The week ends with brooding when seasoned experts in their field give such information. On the other hand, the self-confidence of Chat GPT is impressive when it proudly delivers answers to extremely complicated questions within a few seconds. But is that all correct? The mathematician Edmund Weitz wanted to know and gave Chat GPT a series of mathematical tasks. His conclusion after checking the answers: The system has “no serious computing capabilities”.