The most famous software of 2023 is ChatGPT, produced by Open AI. It is an artificial intelligence based software that can carry on conversations in natural language. This software has been tested in various fields and in some it has achieved satisfactory results but in others it has not. That is why I tried to see how ChatGPT behaves in the field of teaching Physics in Romanian language.
For this study, I used the ChatGPT Mar 23 Version, Free Research Preview.
At the beginning, I asked him some theoretical questions regarding the polarization of light, the interference of light on thin layers, Kirkhhoff’s laws, how a MOS-FET works, and movement on an inclined plane.
The first observation is that ChatGPT only gives us text responses. We have no images or interactive elements. Also, the text response limits the possibility of entering mathematical formulas. In most answers we do not have any formula.
The answers contain almost correct information, but they are vague, incomplete and lack the mathematical precision required for a physics lesson. Grammatical mistakes or scientific errors are also observed.
The answers given by ChatGPT resemble a series of quotes from different sources that sometimes do not even fit together.
ChatGPT is definitely not a useful tool for teaching Physics, but if a student gave such an answer to a test with subjects only from theory, he could get a passing grade (5 or 6).
But Physics is not limited to theory, it also involves solving problems. That’s why I tried to see how ChatGPT behaves in this area. Due to the fact that this chatbot is text-based, we could only give it very simple problems, without complex formulas or graphics. I only gave it only mechanics and thermodynamics problems because in other fields of physics even the simplest problems need electrical diagrams or other images.
Most of the time, ChatGPT did not even understand the requirements of the problem and calculated something else. Even though I rephrased the problems in a clearer way, it was not able to solve any problem until the end.
Occasionally he managed to find the answer to a sub-point of the problem, but it was insufficient to obtain a passing grade. If ChatGPT did not manage to solve even simple problems, it is clear that it will not be able to solve more complex problems, such as those that are given in exams.
The first conclusion of this test is that we are still far from the moment when artificial intelligence will be able to replace a physics teacher.
The second conclusion is that if you want to be sure that students do not cheat on tests with the help of ChatGPT, it is better to give them problems than theory, because ChatGPT is not able to solve the problems.
Bibliography
1. chat.openai.com/
2. A. Shaji George , A.S. Hovan George , A.S.Gabrio Martin, (2023) “A Review of ChatGPT AI’s Impact on Several Business Sectors”, Partners Universal International Innovation Journal, Volume: 01 Issue: 01 January-February 2023
3. H. Holden Thorp (2023) “ChatGPT is fun, but not an author”, Science, 379 (6630)