Dmitry Soshnikov, MAI Associate Professor: "ChatGPT could score a three or a four for exam"

Artificial intelligence has already replaced humans in many areas – even in those where the benefits of such assistance are still under discussion. For instance, ChatGPT, that wrote a diploma instead of a university graduate, is now on the agenda: a ban on the use of neural network models in research and education activities has already been put up for discussion.

Dmitry Soshnikov, associate professor of MAI Institute No. 8 "Computer Science and Applied Mathematics", shared his opinion on whether ChatGPT is useful or harmful.

– ChatGPT will make teachers’ life more complicated, and that is for sure! My colleagues and I conducted an experiment to see if ChatGPT could pass an exam in one of the subjects, logic programming, in MAI. It turned out that the chatbot successfully solves simple tasks as far as one of quite exotic programming languages, Prolog, is concerned, and more or less OK passes tests with multiple answer options. Moreover, the chatbot during the dialogue told us where to apply Prolog and it even also drew an example of a program that could be used in bioinformatics. In general, artificial intelligence could score a three or even a four on the exam, says Dmitry Soshnikov.

A chatbot can become a good tool for improving the efficiency of creative work, if used correctly. For example, it might write large texts, leaving the most important thing to the author – the search of the very idea of it.

– I don't see anything wrong with the fact that a student or a research scientist creates a significant part of his work using generative tools – as long as he is fully responsible for the content, – says the associate professor. – A neural network can very easily write meaningless text, and the task of a person is to ensure that the result corresponds to the original goals. If a diploma written with the help of ChatGPT contains the student's original thoughts, is scientifically sound, and the student demonstrates knowledge of the work he has done during the defense, I do not see anything wrong about it.

Dmitry Soshnikov emphasizes that the discussion about ChatGPT resembles the debate about whether it is possible to give students free use of reference materials during the exam. By prohibiting the use, it is possible to test students' knowledge via relatively typical tasks. At the same time, if you open access to information, you will have to come up with more creative tasks, but such an exam will be as close as possible to real life, in which you need to be able to apply creative skills while solving non-standard cases. By the way, the generation of such tasks can also might be assigned to neural networks.

For more information on the experiments with ChatGPT, including examples of dialogues, visit Dmitry Soshnikov's blog.