Given the rate of AI advancement, what does homework look like in 5 years? Do teachers just rely on in-person, handwritten exams and in-class essays as the only way to prove comprehension?
Ideally it brings on a wave of project based learning and public speaking.
Really hard question! Testing is very hard. The optimist in me is thinking that maybe we can use AI "defensively" as much as "offensively". Maybe there's a way that Professors use AI to create unique-and-hard-to-cheat-with-AI-for homework?
If a student can get an AI to write an essay that even the teacher can't tell is from an AI, IMO the student should get an A š "Writing is just AI tuning" is the new "coding is just Googling"
The cynicist in me says the same. The optimist says styles that slant away from written and more to "Proof of Work", so verbal presentations, debates, re-enactments, etc
I bet there will end up being some kind of Benfordās Law equivalent here eventually. Where the weights of popular models kick out a kind of statistical thumbprint that teachers can track.
presentations - project based learning - moving more towards individual creation vs answering templated questions
Great question. Was just chatting to my son about Lex - he has a writing disability and thought this might be perfect for him. Is it cheating? If professional writers are using it to get through blocks why can't a high school student?
i dont have the answer but i recently shared Lex with a college student and 1) he loves it 2) he has been sharing with all of his friends Perhaps it also shifts way more to project based work ā its hard to fake comprehension in a dialogue
an amazing opportunity to re-think homework from the ground up humans will be more needed than ever
embrace it--teachers used to turn noses up at calculators, many still do at wikipedia frame working with AI as a real-time workshop, then it's about growth & critical thinking & not being "lazy" ran an early exp in the spring: https://twitter.com/courtlandleer/status/1518663013724610560?s=20&t=QKx5hkGK2m-s473DJYrg0A
I always wondered why teachers forced students to limit in-class tooling that they would have easy access to in the real world (like calculators). Maybe the fundamental "schooling" shifts from a tactical problem solver to a creative prompt designer?
I'd argue we will likely see an increase in apprenticeship learning vs. traditional university learning similar to other non-US countries. Especially for things outside of traditional tradecrafts like electrical, plumbing, construction. Less about "tell me you can do it" and more about "show me".
Moving away from models that assume constant information scarcity and focus on rote memorization is a good start.
I think it's already fairly easy to cheat for students that want to with current tools. Maybe homework moves towards completion based (a lower % of your total grade in a class) & teachers will put more weight on presentations/exams/in-class essays
There's probably a future word processor that offers an AI copilot that produces an essay together with the student, and it could be possible that teachers simply require written work to be done in that tool. There's another possibility that essays become hybrid interactive multimedia projects that are harder to fake.
Shift towards soft skills? I wonder if there'll be AI that detects AI-aided or AI-created work. Not sure if that's even possible or could be accurate.
Hopefully creates a shift towards in-person, no playbook, problem solving required by real life (especially company building). Unfortunately that problem solving requires a certain āfloorā of knowledge that AI could certainly smoke screen for students. Hard problemā¦
All the charts I look at don't show AI coming close to human levels of intel for a long long time. Sure, when you compare AI now to AI in 2000, it's come a long way. But even though Mauna Loa is gaining on Everest it's still nowhere near Olympus Mons.
I think a better question is what do schools look like in five years with AI. Many schools have phased out homework already. Good schools are moving more toward authentic assessments where, yes, work is hard to fake and done in class⦠easier to plagiarize a paper than give a speech or work through a case study.
homework and exams are like GANs, an adversarial game, an arms race between students and teachers exams cause advancements in bypassing exams new ways to bypass exams causes new kinds of exams
hopefully AI drives homework to extinction If you need to prove comprehension, teacher can talk to you. If you need to scale that, talk to an AI
Iād argue even if little changes, thatās kind of ok. Current AI tools donāt bring insight. They bring boilerplate text. If a student has an understanding of the material it can provide amazing scaffolding. But if they donāt, itās only going to give them a mediocre composition.
Confession, I had zero interest in learning more about biomedical ethics, so had GPT-3 write my arguments and reverse engineered my paper.
Maybe this offers the chance to reconceive homework. AI is commoditizing online content for SEO purposes so differentiating requires a unique perspective / a human touch beyond parroting info. If homework prioritizes thought process, AI can automate other parts so focus is on creative thinking vs parroting.
No more homework. Best way to prove comprehension is to prove expertise. Project based learning could substitute for homework, but it lacks structure because you could be doing the same case studies on concepts you already mastered. I think a knowledge graph protocol could help here.
Is there a tool that a teacher could use that'd detect AI usage? That seems more likely to be applied than all teachers everywhere changing how they've done homework for decades, unfortunately
No, the homework is to come up with the prompt that solves the task best. Only half joking hereā¦
What sort of metadata is there with AI created content? Is there a flag to indicate Author really - Iām very intrigued by this. Then, from a litigation/authentication standpoint, how do we know if the author is really the author and not AI? But to answer your question, my vote is for interpretive dance.