AI and "research"

[pjclinch]

this morning's email had a note from Academia.edu (a platform for sharing research) that "an AI created a podcast of your paper"...

I'm not in the business of writing papers (I'm a secondary author of a couple of things on surgical skill evaluation I wrote software for in the 90s), in this case it was a letter I had published in the JRSM about 20 years ago, and it had been spun out to sound as if it was a research paper, spoken as if by the site editor, taking quite a lot longer than reading the original in full. So utter mince, in other words. I rated it 1 star out of 5 and said why.

So a learning point here, of beware "helpful" AI summaries where you're not sure what the inputs are. In this case I was very familiar with the inputs, and major extrapolations beyond them had been made (not least in describing it as a research paper!).

I suspect a case of:
"why are you using AI to do this on a letter?"
"because we're using it on everything without much thought or quality control; because we can"

Pete.

[Manc33]

It's wrong more than it's right from what I have found. Also, there's a sort of "anti-AI" thing going on where AI is being made to detect whether something was created using AI. Ironic isn't it.

I asked Chat GPT to make the above few sentences as formal as possible:

Based on my observations, it appears to be incorrect more often than it is accurate. Furthermore, there seems to be an underlying sentiment of resistance toward artificial intelligence, exemplified by efforts to develop AI systems capable of detecting AI-generated content—an endeavor that is, arguably, somewhat ironic.



Honestly that's a lot better than my effort. It's used the United States spelling of "endeavor" though and this is what I mean about how it's being caught out by other AI, in fact in this case simply by my browser underlining "endeavor" because it's meant to say "endeavour". Also notice the long dash Chat GPT uses:

— instead of -

Another giveaway. I don't even know where that is on my keyboard, or if it exists.

So then you have to start every Chat GPT session by saying "Use UK spelling and don't use a long dash" and whatever else it's going to put in that you wouldn't.

[Jdsk]

AI tools have a lot to offer to evidence-based research and to evidence-based policy making.

Cochrane has just published:
"How is Cochrane advancing responsible AI for evidence synthesis?"
https://www.cochrane.org/about-us/news/ ... -synthesis

It's a very good snapshot of what's happening.

Jonathan

[tim-b]

It also makes stuff up

Problem is, we didn’t actually have that feature. We’ve never supported ASCII tab; ChatGPT was outright lying to people. And making us look bad in the process, setting false expectations about our service.

So what did Soundslice do?
They added the feature, of course

To my knowledge, this is the first case of a company developing a feature because ChatGPT is incorrectly telling people it exists.https://www.holovaty.com/writing/chatgpt-fake-feature/

In other realms AI has been banned from some US Courts for making caselaw up, you couldn't write it (well, apparently AI can)
https://news.bloomberglaw.com/bloomberg ... ing-orders

[fastpedaller]

My Daughter marks students' work for a Uni - Lots of students are using AI (the great majority in the subject she marks) . Because she considered many of the students had very little real knowledge, she was adamant they should be failed for the assignment. She expressed her concerns to a more senior colleague, who said (this is my recollection from what my daughter said) "The thing is, we can't prove they've used AI, so it's difficult to justify failing them.
As an example..... One student wrote "I am behooved to present my findings to you". My Daughter had to google it to find the meaning, and we also had no idea Perhaps the student went to school in the 1800's?
So in future it could be the students who don't use AI but achieve 60% overall get failed, and the cheats (no other word for it) sail through and pass with distinction! What are we heading for (Daughter works for the NHS)

[Cugel]

My Daughter marks students' work for a Uni - Lots of students are using AI (the great majority in the subject she marks) . Because she considered many of the students had very little real knowledge, she was adamant they should be failed for the assignment. She expressed her concerns to a more senior colleague, who said (this is my recollection from what my daughter said) "The thing is, we can't prove they've used AI, so it's difficult to justify failing them.
As an example..... One student wrote "I am behooved to present my findings to you". My Daughter had to google it to find the meaning, and we also had no idea Perhaps the student went to school in the 1800's?
So in future it could be the students who don't use AI but achieve 60% overall get failed, and the cheats (no other word for it) sail through and pass with distinction! What are we heading for (Daughter works for the NHS)

Much of what's offered by universities in Blightedland is not an education but a certificate-of-education product. The university is no longer a grove of academe but a business like any other.

To make a profit it must sell popular and fashionable goods, in this case a cert of ed for ..... anything sounding clever. It can't afford to fail the buyers as this would be tantamount to selling faulty goods or even theft. ("I paid them the £NN,000 fees but they wouldn't give me the goods"). Failing the dunces and cheaters in large swathes will soon see their market shrink.

**********
No doubt some groves remain in academe in which a genuine education can still be had. But most seem to come away with a "degree" that's of little or no worth, either as a traditional education that improves the student mind or as a means to an interesting (or any) employment.

More blight in this benighted land.

[al_yrpal]

I have never tried AI before but this prompted me to try it. I asked ChatGtp to write a modern Biggles story and it did. 'Biggles and tbe drone war!' Brilliant, very entertaining....

Al

[Jdsk]

My Daughter marks students' work for a Uni - Lots of students are using AI (the great majority in the subject she marks) . Because she considered many of the students had very little real knowledge, she was adamant they should be failed for the assignment. She expressed her concerns to a more senior colleague, who said (this is my recollection from what my daughter said) "The thing is, we can't prove they've used AI, so it's difficult to justify failing them.
As an example..... One student wrote "I am behooved to present my findings to you". My Daughter had to google it to find the meaning, and we also had no idea Perhaps the student went to school in the 1800's?
So in future it could be the students who don't use AI but achieve 60% overall get failed, and the cheats (no other word for it) sail through and pass with distinction! What are we heading for (Daughter works for the NHS)

Appropriate and inappropriate use of AI tools by students on courses is very different from research. One of the biggest differences being that it's clear who is responsible for sorting it out.

Most institutions have approached this by building on the policies and procedures that were already in place for plagiarism. And of course those had recently undergone massive changes with the appearance of the Web.

An example from Oxford:`
https://www.ox.ac.uk/students/academic/ ... plagiarism
which includes:
"Please note that artificial intelligence (AI) can only be used within assessments where specific prior authorisation has been given, or when technology that uses AI has been agreed as reasonable adjustment for a student’s disability (such as voice recognition software for transcriptions, or spelling and grammar checkers)."

Jonathan

[Stevek76]

As an example..... One student wrote "I am behooved to present my findings to you". My Daughter had to google it to find the meaning, and we also had no idea Perhaps the student went to school in the 1800's? )

hmm, not sure behoved is a word an LLM is likely to chuck out unless it's been asked to role play as jane austen or something

Use of LLMs is also affecting the peer review process:

https://www.theguardian.com/technology/ ... er-reviews