Artificial Intelligence, Academic Integrity, and the Role of the Library - PowerPoint Presentation

Slide1

Artificial Intelligence, Academic Integrity, and the Role of the Library

Slide2

Outline

The Threat: AI Generated Text

Time for 1-2 Questions

The Problems for Classroom Faculty and for Higher Education

Time for 3-4 Questions

Suggestions for Higher Education

Solutions for Classroom Faculty

Questions

Slide3

The Threat: What is AI generated text?

A bot is trained, fed?, various sources and it learns content and language. Parts of the program learn grammar, other parts learn “facts”, and other parts learn how to put it together.

Currently these AI programs are trained offline; they are not “fed” or “seeded” other large data dumps once they are live and online.

“Garbage in, garbage out”. That explains why most AI is accused of being racist, sexist, and rude. They are typically seeded with a lot of freely available social media, like Twitter and Reddit.

Slide4

AI (Artificial Intelligence) bots (programs) have been writing for a few years now. Many clickbait websites and social media posts are generated by AI. Many bots are used for chat purposes / automated customer service; better than a static “help” button on a webpage.

The bot making the news most recently is

ChatGPT, based on GPT-3. It is a third generation AI. It was trained on Wikipedia, Reddit, Twitter, and a lot of eBooks. GPT-4, released 3/14/2023, appears to have the same “training” materials. GPT-4 is the basis for Bing Chat. Jasper AI is based on GPT-3.AI can produce original memes, social media posts, emails, code for websites, poetry of various types, comic strips, paintings, recipes, and essays.

Slide5

The Problems of AI for Classroom Faculty and Higher Education

Slide6

The Problems with AI for Classroom Faculty

The work will “pass” plagiarism checkers – It creates original work, so it will pass any plagiarism checker (Turnitin.com, Grammarly, or whatever).

It writes in one voice, albeit somewhat awkwardly – It writes like a student.

Some

of the work is good and coherent. It already writes passable 5 paragraph essays. GPT-4 can pass the LSAT and a Bar exam.

It makes up real looking citations – It knows how to create citations, so it creates properly formatted citations. BUT they are fabrications (for now).

The existing AI detectors are just algorithms based on “coherence” and repetition. It is still a judgement call, not 100% accurate

Slide7

The Problems with AI for Higher Education

Students will pass / graduate without being personally responsible for work turned in.

Students will pass / graduate without developing critical thinking skills needed for life and work.

Current plagiarism statements are not strong or broad enough.

Turning in an AI paper is a form of fraud or cheating

Plagiarism definition doesn’t fit (“theft” or “work” AND “other”)

Slide8

Why “plagiarism” doesn’t seem to be sufficient“Plagiarism” definitions sometimes mention “theft” – this is not theft as the bot is giving what was asked of it. No theft is involved.Plagiarism is often described as the “work of another” – It can be argued that it is not the “work” of just the bot, but collaboration with the bot to produce the final product. The student “works” to get the AI to write. It takes time and effort to produce something with AI.

Lastly, plagiarism is often tied to the idea of the work of “someone” else. I will not assign personhood to the bot. Nothing has passed a modified Turing test yet, but still…

Slide9

What we really haveWhat is produced is a collaboration, a back and forth, between the AI and the student. We end up with a paper written by two authors. The student provides prompts and moves the creative AI in the right direction, but together they write the paper.The amount of student involvement is not much, but there is some.

What we have is a form of fraud. The student falsely claiming something as their own in order to gain something, a grade. The problem being that the student gains no knowledge or critical thinking by turning in the work.

Slide10

Additional comments about collaborationSince 2011, and the Leahy-Smith America Invents Act, the “inventor” is defined as “an individual” or “individuals”. “Individual” was defined by the US Supreme Court, in 2012, in

Mohamad v. Palestinian Auth., 566 U.S 449, 454 as “a human being, a person.”Thaler v. Vidal

– a patent lawsuit that was dismissed because the “inventor” was not a “natural person”, a “human being” as intended / interpreted by the Patent Act. A Federal Appeals court agreed that personhood is essential to applying for patents. [This is why I focus on the “human being” aspect and struggle to apply “personhood” to any AI]Thaler v. Perlmutter (same Thaler, but now against the head of the copyright office) the Copyright Office is defending the idea that a work must have human authorship.

Both Patent Office and Copyright Office seem to be working with the idea that the AI can be a collaborator, but not listed as the responsible party. The AI can be listed as “inventor” or “co-author”, but not as the owner of the patent or copyright.

Slide11

Suggestions for Higher Education

Slide12

Suggestion 1: Change of focus of the University (or at least one marketing effort)

Focus on critical thinking

Focus on personal responsibility

The student is focused on getting through classes and school, just to get the piece of paper at the end. They want a degree just to get a high paying job. Reinforced during Obama Administration with the creation of the “College Scorecard”.

We need to change that script. It is

not “go to school, get paper, and earn money”, but to “go to school, get an education, gain critical thinking, knowledge, and skills, demonstrate responsibility, earn respect, get better jobs, and earn money.”

The goal of a University education is (should be) not a degree, the paper, but an education. They are very different things with very different goals. If we focus on education, then we can focus on intellectual pursuits and that a student that cheats, isn’t cheating a system, but themselves. Yes, personal responsibility. The intellectual contribution of the student is what we are supposed to be grading, what we should be checking for, and what we should be promoting. A student that plagiarizes, or buys a paper, or turns in a paper produced by AI is cheating themselves and is not getting an education.

Slide13

Suggestion 2: Collaboration Statement

Our policies need to emphasize personal responsibility. That the individual student is responsible for what gets turned in. It is their composition, their thoughts, in their words, that may include the words, thoughts, and data from others,

but

properly attributed to them. If the student wants their name on the transcripts, and on the degree, then the work must be theirs. Their writing, their research is about how they contribute to the scholarly conversation. That’s a major goal of a university education. The student learns from others and then contributes; they join the conversation. Notice I’m throwing in the language from the

ACRL

Framework. That’s because Academic Integrity is intimately related to IL.

Slide14

A collaboration statement should encompass the ideas:Grades should be based on the quantity and quality of the work done by each student / author.

Each part of a collaboration (down to the sentence?) should be identifiable as to whose ideas are whose and who wrote what. (Google Docs and MS Word track changes or as “comments”)The amount of work allowed by the AI will be up to the instructor and the nature of the work.

Slide15

Suggestion 3: AI Statement

For now, AI is being treated like Wikipedia was in the beginning; forbid it until we figure out if and how we can use it. Eventually, AI will be a tool that may be used in the classroom. Some have already figured out ways to incorporate it in assignments. We just need to figure out how to use it ethically and in ways that advance the educational purposes of the course or the University.

Slide16

An AI statement should include (in my pro-AI opinion):AI should be allowed and encouraged in courses where the instructor choses to use it. This is based upon the academic freedom of classroom instructors.

AI is only allowed when stated explicitly in the instructions by the faculty member. When in doubt, the answer is no.This is a temporary statement, as a few years from now, AI will be viewed more as a tool (I hope). This must be revisited, often, for the next decade or so.Possibly allowed as a computer aid for accessibility purposes.

Slide17

Solutions for Classroom Faculty

Which for the sake of time, I will rush through and point out a website to view them more in depth later.

Slide18

Solutions: The list

Require rough draft turned in early

Require reference page turned in early

Require more formats of information

Require search strategy turned in as part of assignment

Lower-division option – footnotes or annotated reference page

Upper-division/grad option – appendix including evaluation of method

Require resources come from library’s licensed database

Require evaluation of resources (form, annotations, footnotes, or appendix)

Require notes turned in with paper

Use one of the AI detectors out there

Slide19

Require rough draft turned in early

You require students to turn in a rough draft before the paper is due.

It requires that you read the two side by side, so that you can check if the “voice” changes; the voice of the student in the rough will not match the voice of the bot in the final paper.

This can work, if you can tell the difference between the voice of two authors. It helps if you have a large enough sample of the student’s writing.

Slide20

Require reference page turned in early

You require that the student turn in their reference page early.

It requires that you evaluate the reference page to be authentic works (books, journals, and such) from real publishers and real authors. You may have to click hyperlinks or search online for some of the works.

This can work because currently, the bots are just making up resources. If the resources aren’t found online, you know they are made up. AND once the paper is turned in you can see if the works are incorporated into the paper. You can require a certain number make it into the final paper. Again, if the student just adds random quotations and citations, you should detect a change in voice.

Slide21

Check a Publication; is it real?WMS, and I’m sure other OPACs

, have publication finders, article finders, or the ability to just look up a book by its title. Advertise this (send links) to faculty so that they can quickly copy and paste from a student’s reference page.At Hope InternationalHere is a link to HIU’s

publication finder. You search for the title of a journal and see if we have it. https://hiu.on.worldcat.org/atoztitles#journalHere is a link to HIU’s

article finder. You search for the article and see if we have it. https://hiu.on.worldcat.org/atoztitles#article

Slide22

Require more formats of information

You require that the student’s paper or project use multiple formats for resources (dictionaries, encyclopedias, atlases, etc., as well as newspapers, books/monographs, conference papers, blogs, etc.)

Even, or especially, interviews, surveys, and such that require human contact/interaction

It requires that you are generally aware of what formats and titles would be appropriate for your subject.

This can work because the AI will not have access to all these resources. And the AI should not be able to make up people for interviews.

Slide23

Require search strategy turned in as part of assignment (lower-division)

You require that the student write out a detailed search strategy for all electronic resources used in a footnote or in an annotated reference page; what database, what keywords, and what limiters did they use.

It requires that you read and understand the process and that you know which databases we have.

[this also leads to wonderful discussions about finding good resources and a great opportunity for the librarian to come and walk students through this research process] ;)

This thwarts the use of a bot paper because those resources must be included into the paper. The bot cannot do that (yet).

Slide24

Require search strategy turned in as part of assignment (upper-division or grad)

You require that the student write out a detailed search strategy for some electronic resources used in an appendix; what database, what keywords, and what limiters did they use. AND they evaluate the efficacy of that strategy; what worked and what could have improved their search.

It requires you to read and evaluate their thoughts on their process.

Again, this beats the bot because the bot doesn’t have access to databases.

Slide25

Require resources come from library’s licensed database

You require that all (some) of the electronic resources come from the licensed resources in the library. (see the Database A to Z page)

It requires that you look at the URLs or trust the stated search strategy for each item.

As of now, the chatbots can only use what they “know” or have been taught. They will have access to Google Scholar (eventually), BUT those articles are free, open access, or pre-prints, NOT the version in a licensed database. In other words, the bot will not have access to these resources. If your paper has nothing from our database, it may be a bot (or bought) paper.

Slide26

Require evaluation of resources (form, annotations, footnotes, or appendix)

You require that the students evaluate each resource in footnotes or annotated reference page on some of the following elements: Currency, Relevancy, Authority, Accuracy, and Purpose or Persuasion.

It requires you to verify their resources and understand the

CRAAP

ideas.

Evaluation of the resources implies that the resources should be included in the paper. Again, the chatbot will make up resources, so nothing to evaluate.

Slide27

Require notes turned in with paper

You require that the student turn in notes of some sort with the final paper. Require notes taken from research, either a document or a scan of their highlighted articles, turned in as backmatter for the paper.

It requires you to look at and evaluate if they were really reading or just marking an article to make it look real.

This can work because it, again, makes the student work and to be aware that they are being graded on their work process, not just a final product.

Slide28

Use one of the AI detectors out there

No requirement for the students.

It requires that you turn in the paper / project to one of the AI detectors that exist:

Article that lists 7 current AI detectors

https://www.digippl.com/2022/12/ai-content-checker.html

https://originality.ai/can-gpt-3-5-chatgpt-be-detected/

http://gltr.io/dist/index.html

https://www.zerogpt.com/

It works by checking the paper for coherence (algorithm to look for proper order of parts of speech) and repetition.

Slide29

Questions?

Slide30

Link to this presentationhttps://sites.google.com/view/practical-information-literacy/beating-ai

Website also has the instructions written out.Email: sejung

@hiu.edu