Brilliant analysis from a history professor of why student use of ChatGPT should be unacceptable

[highplainsdem]

Ran across this via a link in a Vice article about ChatGPT thar I just posted about at https://www.democraticunderground.com/100217693429 .

Bret Devereaux's blog post about ChatGPT at https://acoup.blog/2023/02/17/collections-on-chatgpt/ is a very long read, but well worth reading in its entirety, and sharing.

What one can immediately see is that a student who simply uses ChatGPT to write their essay for them has simply cheated themselves out of the opportunity to learn (and also wasted my time in providing comments and grades). As we’ve seen above, ChatGPT cannot effectively replace the actual core tasks we are training for, so this is not a case where the existence of spinning jennies renders most training at hand spinning obsolete. And it certainly doesn’t fulfill the purpose of the assignment.

To which some boosters of the technology respond that what I should really be doing is training students on how to most effectively use ChatGPT as a tool. But it is not clear to me that ChatGPT functions well as a tool for any part of this process. One suggestion is to write an outline and then feed that into ChatGPT to generate a paper, but that fails to train the essential communication component of the assignment and in any case, ChatGPT is actually pretty bad at the nuts of and bolts of writing paragraphs. Its tendency in particular to invent facts or invent non-existent sources to cite makes it an enormous liability here; it is a very bad research tool because it is unreliable. Alternately the suggestion is that students could use ChatGPT to produce an essay they edit to fit or an outline they fill in; both problems run into the issue that the student is now trying to offload the most important part of the task for them to learn: the actual thinking and analysis. And the crucial thing to note is that the skill that is not being trained in both cases is a skill that current large language models like ChatGPT cannot perform or perform very poorly.12

I suspect this argument looks plausible to people because they are not thinking in terms of being trained to think about novel problems, but in terms of the assignment itself; they are thinking about the most efficient way to produce ‘one unit of essay.’ But what we’re actually doing is practicing a non-novel problem (by treating it as a novel problem for the purpose of the assignment), so that when we run into novel problems, we’ll be able to apply the same skills. Consequently they imagine that ChatGPT, trained as it is on what seems to be an awful lot of mediocre student essays (it mimics the form of a bad student essay with remarkable accuracy), can perform the actual final task in question, but it cannot.

And he has harsh words for teachers and schools that accept ChatGPT-aided cheating:

Now will this disrupt some classrooms? Yes. Overworked or indifferent graders will probably be fooled by these papers or more correctly they will not care who wrote the paper because those instructors or graders are either not very much invested in learning outcomes or not given the time and resources to invest however much they might wish to. I think schools are going to need to think particularly about the workload on adjuncts and TAs who are sometimes asked to grade through absurdly high amounts of papers in relatively little time and thus will simply lack the time read carefully enough. Of course given how much students are paying for this, one would assume that resources could be made available to allow for the bare minimum of scrutiny these assignments deserve. Schools may also need to rethink the tradeoffs of hiring indifferent teachers ‘for their research’ or for the prestige of their PhD institutions because the gap between good, dedicated teachers and bad, indifferent ones is going to grow wider as a result of this technology.

[Redleg]

1. We were told by our provost and dean to "just embrace it"

I almost suffered a stroke from the effort it took me to bite my tongue.

[Earth-shine]

2. "Its tendency in particular to invent facts or invent non-existent sources"

That problem is going to be solved very fast. I'd say within two years.

All the tech companies are focused on creating their own AI systems. Some will be more accurate and factual than others, but all will be improved fast.

The "Chat" is out of the bag. Get used to it.

[Renew Deal]

3. It doesn't take a brilliant analysis from a professor to figure this out.

The benefits to research and writing outweigh the benefits of ChatGPT for students. At the same time, students should be prepared to responsibly use these types of tools in the real world.