AI in Education

[AngelaHornung]

I'm currently in university studying Computer Science. My region allows those with Bachelors to teach in public education (after a two year licensing program after university) for those with technical degrees. My plan is to become an educator in my community which is a bit on the rougher side but still lovely. 

 

I've lately been utilizing Microsoft Copilot to aid in my studies. It will create large problem breakdowns for me in subjects I'm struggling in without needing to nag and wait for the professor or search over forums across several web pages which may include contradictory or misleading information. While not perfect it does an overall great job and while I utilize it 9/10 for pure productivity aids, sometimes if tired I've caught myself slipping and just tossing the answer in (I know I'm cheating myself when I do this). 

 

This got me thinking, this is going to be a HUGE problem soon isn't it? Its already made some splashes, but like if these language models become so sophisticated we can't tell genuine from generated, how will teachers of the future be expected to teach students? In an ironic twist the same silicon which allowed online schooling seems to be making it obsolete due to the need to guarantee or at least attempt to guarantee no cheating. Students will need to be actively observed while completing ALL assignments. 

 

On the bright side this could spell the end of homework as there will be no point in sending assignments home with students if you can't determine what is genuine or generated on a student's assignment. 

[Kilrah]

You're about a year late, shortly after LLMs got unleashed it did indeed become a problem and pretty reliable ML-based tools to detect LLM outputs have been made.

[AngelaHornung]

4 minutes ago, Kilrah said:

You're about a year late, shortly after LLMs got unleashed it did indeed become a problem and pretty reliable ML-based tools to detect LLM outputs have been made.

Yes, hence my mention that it has made some splashes, but what happens when the differentiation between genuine and generated can't be differentiated? As these models continue to improve their responses become much more human. 

[Kilrah]

Detection tools will evolve just the same in parallel since they use the same tech...

And then even if the output really gets that good it'll be easy to find out the opposite way, your typical kid/student is not able to write pristine output, so it'll be found out by being "too good", just like previously "there's no way they could have written that on their own, has to be lifted from a book/reference as is".