Recently, I've been thinking about an interesting question:
Now that AI has become so powerful and can solve competitive programming problems, how can we identify people who use AI to cheat in contests?
For example:
- Are there any recognizable patterns in AI-generated code?
- Can we distinguish AI-generated solutions from human-written ones?
- Are there any suspicious coding patterns or behaviors that contest organizers should look for?
- Is it even possible to reliably detect AI-assisted cheating?
What made me think about this?
Today, I had an exam with 24 other students in Group B, an IOI training group at Allameh Helli High School in Tehran(best school in IRAN), where students prepare for the International Olympiad in Informatics (IOI).
During the exam, something caught my attention.
One participant managed to solve Problem A, which was the hardest problem in the contest, surprisingly quickly.
What made this particularly interesting was that several other strong participants, including some who are usually very capable of solving difficult problems, couldn't solve it before the contest ended.
I don't want to jump to conclusions or accuse anyone of cheating. There could be many explanations for this result.
However, this situation made me wonder:
In an era where AI can solve programming problems, how can we distinguish an exceptionally talented programmer from someone who is simply using AI during a contest?
I'd especially like to hear your thoughts on the technical side of this question.
Have you noticed any patterns in AI-generated competitive programming code? Are there any reliable ways to identify it?
I'd really appreciate hearing your experiences and opinions!
You don't need to cut or other's line, you could just make yours' longer by practising even more than anyone discipline beats everything.
If this person really cheated, they might have taken the place of someone like you, who genuinely worked hard to earn it. That's what makes cheating so unfair.
In online contests, there is no way to detect one if he is cheating from his code. AIers can only ask AI to generate problems' solutions but not code, and they write code on their own. This way, you cannot see any AI signals from his code.
The only possible way is to ask contestants to provide the recording of their screens, and use a specialized client (and/or require another camera which monitors the contestant) to prevent them from using VMs, etc.
But then what about going to the bathroom, you can't record that! And they can use to cheat in there
Unless there's a camera in the bathroom...
Exactly! Today’s exam actually had a cheating incident like this. I even heard from one of my friends that he went to the bathroom and the person in the next stall was talking to ChatGPT using voice mode and asking it for ideas on how to solve the questions.