As someone who works in a University and sets and marks coursework and exams, I can tell you that this is on our radar, but there are fairly easy mitigations to counteract it. The most obvious way is to alter the type of coursework you set. Give students a graph and ask them to interpret it for example, rather than factual recall or definition type questions, get them to give a presentation in person, give them a viva, produce and explain a poster in person etc. The essay as a piece of written work is particularly susceptible to AI, so we don't set an essay, we ask students to interpret data, or propose an experiment to explore a hypothesis and so on. We will certainly get to the point where AI can provide reasonable answers to these types of problems, but currently all programmes such as ChatGPT are doing is putting a shiny front-end on a google search. The single most obvious way to prevent AI cheating is to bring back face to face in person exams - quite hard to use AI to write an essay for you when all you have is a pen and a piece of paper!
To be honest, the shift in question style to avoid students google-fu skills has been going on since 6 months into the pandemic. We got 24 hours notice to shift all teaching online at the start of lockdown, that was also only a couple of weeks before exams were due. We typically set, quality control and confirm exams 3 months before they are due through a multi step process. It was not possible to re-write these for the first online iteration and the marks reflected the fact that students now had access to the internet, each other, lecture notes etc etc. The next set of exams 6 months later were very different, with question written to test understanding rather than factual recall. It is possible, the sector is still learning, we are getting there but are not there fully yet.
Going back to AI, this is largely an extension of essay mills, where students pay someone else to write an essay for them. This are illegal but the majority are not based in the UK. They are tempting for students but generally not very sophisticated, so relatively easy to spot under certain conditions. The classic example is the poor/failing student suddenly turning in a first class piece of work - it sticks out like a sore thumb. We also use plagiarism detection software for every piece of coursework and also match against different students to look for collusion (a fine line between copying and 'working together to discuss a problem' so definitely not an exact science). Could I spot a first class student using AI to produce a first class piece of work? Probably not in all honesty. Is AI clever enough to produce a mediocre piece of work for a mediocre student? Probably, and this is likely the real danger - excellent students are typically proud of their achievements and want to demonstrate their excellence. I am making a generalised statement here, but it is not the excellent students that are the target market for AI or essay mills. There is so much pressure on students to succeed, from themselves, peers, parents, society, financial reward of a good job etc that it is the weaker students who feel they need to do better that are most at risk. We have all sorts of checks and balances, QC and review within universities, many of which students do not know about for obvious reasons, that to get through an entire 3 year degree on the back of others or AI is pretty much impossible. To do it for specific modules or assessments however is more likely and it up to us as educators to recognise this and mitigate against it.
There is loads more I could write here, on student numbers, societal pressures and quality, but it would potentially veer into politics so I will avoid that avenue other than to say none of this is a simple problem that can be easily solved, and rose tinted spectacles of 'it was better in my day when only x% of students went to Uni' are not that helpful (or true!). I am also in the 'Sciences' where factual recall is perhaps more important than essay writing and therefore easier to mitigate against AI practices, but we are certainly not immune. Once AI becomes much more ubiquitous in society (think about what Alexa can answer now compared to 'Ask Jeeves' 20 years ago!) then as a population you will not want graduates with factual recall, you will want graduates with a different skill set - problem solving, empathy, compassion, teamwork etc that cannot be easily taught by AI. The biggest challenge is getting universities, and society, to recognise this and act now, rather than being 10 years behind the curve and scrambling to keep up. As a society we still equate knowledge with the ability to regurgitate facts, not solve problems. Knowledge is essential for this, but being knowledgeable is not the same as being a problem solver!
