New York City has also banned AI for younger students as concerns over classroom use grow
Another school district has banned AI from being used in the classroom by students across all grades. While this isn’t the first ban of its kind that we’ve seen, what’s interesting is that it was introduced just weeks after the district started piloting Google Gemini for older students. Frederick County became the latest school district to ban AI use in the classroom this week, voting 6-1 to restrict the technology from pre-K through 12th grade. Initially, the board considered only banning AI for younger students, but eventually extended the ban through high school.
AI is anathema to teaching and learning.
Easier isn’t better for learning, it’s actually worse. Until you have solid foundations in something, AI should not be used.
Arguably, unless you can use a more ethical AI, it should be discouraged.
Your statement on its face value, sure I agree, but I can see a path were AI is a fantastic tool for teaching and learning critical thinking. If the teacher is able to alter the “temperature” setting of the LLM, they could crank it all the way up to 1 (meaning full creativity) which will likely induce AI hallucination.
The AI LLM can create output which can speak to a topic but even those with only a little bit of knowledge can easily see the flaws in the AI response. It would give the student a chance to see how rational sounding output can be factually wrong. It would teach students to evaluate the statements individually and tease out logical fallacies or even outright manufactured falsehoods.
Further, if the teacher can again set the temperature lower (perhaps to .7) which reduces the creativity then the LLM output would be more factual with possible logical or factual errors. This would increase the difficulty on the students to be able to inspect what they are given as LLM output and explore the tools and resources available to provide a human check on not only AI, but anything that another human says too.
It can become an excellent tool for teaching critical thinking.
EDIT: The more I’m thinking about this, the better I see this as a teaching tool. Tests students take to cover the material can be teacher created questions accompanied with a completely AI generated answer and the question to the student is “Is this true or bullshit? If bullshit, which parts are wrong?”. Every question on the entire test could use that format. It would be a great test of the student because they’d have to know the material well enough to know where the LLM produced a “right sounding answer” but was in fact, wrong. Even better, if a student tried to cheat using their own AI to answer the “bullshit or not?” question there’s an even higher chance the students cheating LLM would give a wrong answer because of the repeated Error Cascade.
We don’t need AI to teach critical thinking. It doesn’t do anything better or in an innovative way, especially when accounting for the costs associated with AI. It is unnecessary to make a brand new round hole for this square peg.
It does it in a very applicable way in the sense that these students will be encountering AI “in the wild” in the years to come. The ways and methods AI is wrong isn’t really identical to the way humans deceive. As an example, most humans won’t straight up manufacture a fake citation to prove a point where this is a commonplace action of factually wrong LLMs.
In this way using AI to produce the content that students are assigned to criticise is much more useful and less costly than human generated content. I’m not saying that LLMs should be used exclusively to create the target content, but it would be equally wrong to use human generated target content exclusively because humans are wrong in different ways (and that content is much more costly to create).
Are you talking about monetary costs or something more philosophical (ethical, environmental, sociological)? The monetary costs would be extremely cheap. Probably less than $10 in tokens for the entire curriculum for the AI generated content. The necessary human contributions would be significantly more expensive, likely tens of thousands of dollars. The human component is still absolutely necessary, but it would be the far more monetarily expensive part.
AI content students (when graduated into adults) have the evaluate is a new peg with a new shape. We can pretend its the same shape as the previous one and attempt to use the prior solution, but it will fail the education system and the students going through it arriving unprepared in the adult world rife with AI content that they will need to evaluate in their daily lives.
So I don’t disagree that AI can be used as a teaching aid.
There used to be tons of really cool Java applets (before the whole security implications of bundling applets in the browser reared their heads) that made learning interactive.
AI is great for generating interactive content or diagrams for teaching. It can be a tutor to help explore questions when a student doesn’t understand something, or needs additional explanation.
Your example could also help — when I was a student a teacher brought in published papers and asked us to find issues with them, so it’s an established method with a new take. Though, I have had a lot of issues with temperature on newer LLMs, there a lot more stable than they used to be, but you could load up an older Qwen for this on consumer hardware.
My point is more that AI inside of a task is problematic. AI in Google Docs is going to hurt learning, the friction of writing and choosing formatting and punctuation builds mental pathways. Passing homework to ChatGPT removes all the banging your head against a problem that mates you remember how to solve it. Taking a photo of a math problem means you’ll never write 4 pages of balancing an equation just to get it back to the original formulation instead of the desired proof, and so you’ll never build the intuition that keeps you moving forward.
Can I ask what software/device are you using to post to Lemmy?
Mlem, why?
You used an em dash in your post response. Its extremely rare for a human to do that, but extremely common for LLMs to do so.
I just think they’re neat.
I don’t want to give up perfectly good grammatical options to the machines. On iOS and macOS two dashes get switched to em dash, or alt+hyphen does, so they’re easy to type. You can also do en dashes, but I don’t really know when to use those.
I probably make too many typos to be a bot, but I guess that could be programmed in…