ChatGPT

“Why should I test it if it threatens my job?” asked one of the AITAD developers when the proposal to use ChatGPT to create program code was discussed.

Coding is possible, but traceability suffers

From a technical point of view, ChatGPT is not yet a major threat, as it uses software libraries that are no more than two years old. Current frameworks and libraries are not yet taken into account by the current version. However, this will of course change with future versions. Compared to the salary of a developer, the costs for the premium version of the chatbot are not an obstacle.

Viacheslav Gromov, CEO of AITAD, sees the issues of transparency and traceability of the program code as more problematic. “It is conceivable that the chatbot not only writes program code, but also documents it and writes the test program for it.” However, it is to be expected that the program code will vary depending on the formulation of the task with different (synonymous) wordings. The style of the code, for which there are standardizations, is also volatile, adds Gromov.

The use of chatbots as programmers harbors dangers

It also remains questionable whether human developers would work more efficiently if they first had to understand the code generated by the AI, check it for errors and add to it. “This is already a challenging task with human-generated code,” says Gromov, who initially sees the use of chatbots for specifically defined programming tasks. Last but not least, it is debatable whether customers would agree to their development tasks being sent to the chatbot on the internet.

As with all comprehensive automation projects, the use of chatbots reduces differentiation and therefore competitiveness, as the corresponding AI frameworks (chatbots) could be used by any market participant. Only the company with more and better data and more extensive computing resources to train the chatbots will have the better end product. “This can also quickly lead to very fast-moving and global competition between countries that set the framework conditions through data protection regulations and the form of government.”

The chatbot as a helper

An AI that programs AI itself is no longer a (nightmare) dream, but has the potential for faster, more extensive and more complex development, despite a lack of personnel. “This should be tapped into. However, it should be noted that the most likely output is not always the best. This is because there is no compelling correlation between probability and quality. It’s a bit like tearing up yesterday’s newspaper and putting together the best fitting snippets of words one after the other without any overall understanding,” concludes Gromov.