Tag Archives: AI

Media Literacy in the Age of AI, Part II: A Review of Verified

Posted on by | Leave a reply

For the first part of this post, please see Media Literacy in the Age of AI, Part I: “You Will Need to Check It All.”

Just how, exactly, we’re supposed to follow Ethan Mollick’s caution to “check it all” happens to be the subject of a lively, forthcoming collaboration from two education researchers who have been following the intersection of new media and misinformation for decades.

In Verified: How to Think Straight, Get Duped Less, and Make Better Decisions about What to Believe Online (University of Chicago Press, November 2023), Mike Caulfield and Sam Wineburg provide a kind of user’s manual to the modern internet. The authors’ central concern is that students—and, by extension, their teachers—have been going about the process of verifying online claims and sources all wrong—usually by applying the same rhetorical skills activated in reading a deep-dive on Elon Musk or Yevgeny Prigozhin, to borrow from last month’s headlines. Academic readers, that is, traditionally keep their attention fixed on the text—applying comprehension strategies such as prior knowledge, persisting through moments of confusion, and analyzing the narrative and its various claims about technological innovation or armed rebellion in discipline-specific ways.

The Problem with Checklists

Now, anyone who has tried to hold a dialogue on more than a few pages of assigned reading at the college level knows that sustained focus and critical thinking can be challenging, even for experienced readers. (A majority of high school seniors are not prepared for reading in college, according to 2019 data.) And so instructors, partnering with librarians, have long championed checklists as one antidote to passive consumption, first among them the CRAAP test, which stands for currency, relevance, authority, accuracy, and purpose. (Flashbacks to English 101, anyone?) The problem with checklists, argue Caulfield and Wineburg, is that in today’s media landscape—awash in questionable sources—they’re a waste of time. Such routines might easily keep a reader focused on critically evaluating “gameable signals of credibility” such as functional hyperlinks, a well-designed homepage, airtight prose, digital badges, and other supposedly telling markers of authority that can be manufactured with minimal effort or purchased at little expense, right down to the blue checkmark made infamous by Musk’s platform-formerly-known-as-Twitter.

Three Contexts for Lateral Reading

One of the delights in reading Verified is drawing back the curtains on a parade of little-known hoaxes, rumors, actors, and half-truths at work in the shadows of the information age—ranging from a sugar industry front group posing as a scientific think tank to headlines in mid-2022 warning that clouds of “palm-sized flying spiders” were about to descend on the East Coast. In the face of such wild ideas, Caulfield and Wineburg offer a helpful, three-point heuristic for navigating the web—and a sharp rejoinder to the source-specific checklists of the early aughts. (You will have to read the book to fact-check the spider story, or as the authors encourage, you can do it yourself after reading, say, the first chapter!) “The first task when confronted with the unfamiliar is not analysis. It is the gathering of context” (p. 10). More specifically:

  • The context of the source — What’s the reputation of the source of information that you arrive at, whether through a social feed, a shared link, or a Google search result?
  • The context of the claim — What have others said about the claim? If it’s a story, what’s the larger story? If a statistic, what’s the larger context?
  • Finally, the context of you — What is your level of expertise in the area? What is your interest in the claim? What makes such a claim or source compelling to you, and what could change that?
“The Three Contexts” from Verified (2023)

At a regional conference of librarians in May, Wineburg shared video clips from his scenario-based research, juxtaposing student sleuths with professional fact checkers. His conclusion? By simply trying to gather the necessary context, learners with supposedly low media literacy can be quickly transformed into “strong critical thinkers, without any additional training in logic or analysis” (Caulfield and Wineburg, p. 10). What does this look like in practice? Wineburg describes a shift from “vertical” to “lateral reading” or “using the web to read the web” (p. 81). To investigate a source like a pro, readers must first leave the source, often by opening new browser tabs, running nuanced searches about its contents, and pausing to reflect on the results. Again, such findings hold significant implications for how we train students in verification and, more broadly, in media literacy. Successful information gathering, in other words, depends not only on keywords and critical perspective but also on the ability to engage in metacognitive conversations with the web and its architecture. Or, channeling our eight-legged friends again: “If you wanted to understand how spiders catch their prey, you wouldn’t just look at a single strand” (p. 87).

SIFT graphic by Mike Caulfield with icons for stop, investigate the source, find better coverage, and trace claims, quotes, and media to the original context.

Image 2: Mike Caulfield’s “four moves”

Reconstructing Context

Much of Verified is devoted to unpacking how to gain such perspective while also building self-awareness of our relationships with the information we seek. As a companion to Wineburg’s research on lateral reading, Caulfield has refined a series of higher-order tasks for vetting sources called SIFT, or “The Four Moves” (see Image 2). By (1) Stopping to take a breath and get a look around, (2) Investigating the source and its reputation, (3) Finding better sources of journalism or research, and (4) Tracing surprising claims or other rhetorical artifacts back to their origins, readers can more quickly make decisions about how to manage their time online. You can learn more about the why behind “reconstructing context” at Caulfield’s blog, Hapgood, and as part of the OSU Libraries’ guide to media literacy. (Full disclosure: Mike is a former colleague from Washington State University Vancouver.)

If I have one complaint about Caulfield and Wineburg’s book, it’s that it dwells at length on the particulars of analyzing Google search results, which fill pages of accompanying figures and a whole chapter on the search engine as “the bestie you thought you knew” (p. 49). To be sure, Google still occupies a large share of the time students and faculty spend online. But as in my quest for learning norms protocols, readers are already turning to large language model tools for help in deciding what to believe online. In that respect, I find other chapters in Verified (on scholarly sources, the rise of Wikipedia, deceptive videos, and so-called native advertising) more useful. And if you go there, don’t miss the author’s final take on the power of emotion in finding the truth—a line that sounds counterintuitive, but in context adds another, rather moving dimension to the case against checklists.

Given the acceleration of machine learning, will lateral reading and SIFTing hold up in the age of AI? Caulfield and Wineburg certainly think so. Building out context becomes all the more necessary, they write in a postscript on the future of verification, “when the prose on the other side is crafted by a convincing machine” (p. 221). On that note, I invite you and your students to try out some of these moves on your favorite chatbot.

Another Postscript

The other day, I gave Microsoft’s AI-powered search engine a few versions of the same prompt I had put to ChatGPT. In “balanced” mode, Bing dutifully recommended resources from Stanford, Cornell, and Harvard on introducing norms for learning in online college classes. Over in “creative” mode, Bing’s synthesis was slightly more offbeat—including an early-pandemic blog post on setting norms for middle school faculty meetings in rural Vermont. More importantly, the bot wasn’t hallucinating. Most of the sources it suggested seemed worth investigating. Pausing before each rabbit hole, I took a deep breath.

Related Resource

Oregon State Ecampus recently rolled out its own AI toolkit for faculty, based on an emerging consensus that developing capacities for using this technology will be necessary in many areas of life. Of particular relevance to this post is a section on AI literacy, conceptualized as “a broad set of skills that is not confined to technical disciplines.” As with Verified, I find the toolkit’s frameworks and recommendations on teaching AI literacy particularly helpful. For instance, if students are allowed to use ChatGPT or Bing to brainstorm and evaluate possible topics for a writing assignment, “faculty might provide an effective example of how to ask an AI tool to help, ideally situating explanation in the context of what would be appropriate and ethical in that discipline or profession.”

References

Caulfield, M., & Wineburg, S. (2023). Verified: How to think straight, get duped less, and make better decisions about what to believe online. University of Chicago Press.

Mollick, E. (2023, July 15). How to use AI to do stuff: An opinionated guide. One Useful Thing.

Oregon State Ecampus. (2023). Artificial Intelligence Tools.

Media Literacy in the Age of AI, Part I: “You Will Need to Check It All”

Posted on by | Leave a reply

Have you found yourself worried or overwhelmed in thinking about the implications of artificial intelligence for your discipline? Whether, for example, your department’s approaches to teaching basic skills such as library research and source evaluation still hold up? You’re not alone. As we enter another school year, many educators continue to think deeply about questions of truth and misinformation, creativity, and how large language model (LLM) tools such as chatbots are reshaping higher education. Along with our students, faculty (oh, and instructional designers) must consider new paradigms for our collective media literacy.

Here’s a quick backstory for this two-part post. In late spring, shortly after the “stable release” of ChatGPT to iOS, I started chatting with bot model GPT-3.5, which innovator Ethan Mollick describes as “very fast and pretty solid at writing and coding tasks,” if a bit lacking in personality. Other, internet-connected models, such as Bing, have made headlines for their resourcefulness and darker, erratic tendencies. But so far, access to GPT-4 remains limited, and I wanted to better understand the more popular engine’s capabilities. At the time, I was preparing a workshop for a creative writing conference. So, I asked ChatGPT to write a short story in the modern style of George Saunders, based in part on historical events. The chatbot’s response, a brief burst of prose it titled “Language Unleashed,” read almost nothing like Saunders. Still, it got my participants talking about questions of authorship, originality, representation, etc. Check, check, check.

The next time I sat down with the GPT-3.5, things went a little more off-script.

One faculty developer working with Ecampus had asked our team about establishing learning norms in a 200-level course dealing with sensitive subject matter. As a writing instructor, I had bookmarked a few resources in this vein, including strategies from the University of Colorado Boulder. So, I asked ChatGPT to create a bibliographic citation of Creating Collaborative Classroom Norms, which it did with the usual lightning speed. Then I got curious about what else this AI model could do, as my colleagues Philip Chambers and Nadia Jaramillo Cherrez have been exploring. Could ChatGPT point me to some good resources for faculty on setting norms for learning in online college classes?

“Certainly!” came the cheery reply, along with a summary of five sources that would provide me with “valuable information and guidance” (see Image 1). Noting OpenAI’s fine-print caveat (“ChatGPT may produce inaccurate information about people, places, or facts”), I began opening each link, expecting to be teleported to university teaching centers across the country. Except none of the tabs would load properly.

“Sorry we can’t find what you’re looking for,” reported Inside Higher Ed. “Try these resources instead,” suggested Stanford’s Teaching Commons. A closer look with Internet Archive’s Wayback Machine confirmed that the five sources in question were, like “Language Unleashed,” entirely fictitious.

An early chat with ChatGPT-3.5, asking whether the chatbot can point the author to some good resources for faculty on setting classroom norms for learning in online college classes. "Certainly," replies ChatGPT, in recommending five sources that "should provide you with valuable information and guidance."

Image 1: An early, hallucinatory chat with ChatGPT-3.5

As Mollick would explain months later: “it is very easy for the AI to ‘hallucinate’ and generate plausible facts. It can generate entirely false content that is utterly convincing. Let me emphasize that: AI lies continuously and well. Every fact or piece of information it tells you may be incorrect. You will need to check it all.”

The fabrications and limitations of chatbots lacking real-time access to the ever-expanding web have by now been well-documented. But as an early adopter, the speed and confidence ChatGPT brought to the task of inventing and describing fake sources felt unnerving. And without better guideposts for verification, I expect students less familiar with the evolution of AI will continue to experience confusion, or worse. As the Post recently reported, chatbots can easily say offensive things and act in culturally-biased ways—”a reminder that they’ve ingested some of the ugliest material the internet has to offer, and they lack the independent judgment to filter that out.”

Just how, exactly, we’re supposed to “check it all” happens to be the subject of a lively, forthcoming collaboration from two education researchers who have been following the intersection of new media and misinformation for decades.

Stay tuned for an upcoming post with the second installment of “Media Literacy in the Age of AI,” a review of Verified: How to Think Straight, Get Duped Less, and Make Better Decisions about What to Believe Online by Mike Caulfield and Sam Wineburg (University of Chicago Press, November 2023).

References

Mollick, E. (2023, July 15). How to use AI to do stuff: An opinionated guide. One Useful Thing.

Wroe, T., & Volckens, J. (2022, January). Creating collaborative classroom norms. Office of Faculty Affairs, University of Colorado Boulder.

Yu Chen, S., Tenjarla, R., Oremus , W., & Harris, T. (2023, August 31). How to talk to an AI chatbot. The Washington Post.

You won’t believe where AI could replace you next! (But you’re probably going to like it).

Posted on by | Leave a reply

Introduction

We’ve all heard by now of ChatGPT, the large language model-based chat bot that can seemingly answer most any question you present it. What if there were a way to provide this functionality to students on their learning management system, and it could answer questions they had about course content? Sure, this would not completely replace the instructor, nor would it be intended to. Instead, for quick course content questions, a chatbot with access to all course materials could provide students with speedy feedback and clarifications in far less time than the standard turnaround required through the usual channels. Of course, more involved questions about assignments and course content questions outside of the scope of course materials would be more suited to the instructor, and the exact usage of a tool like this would need to be explained, as with anything.

Such a tool could be a useful addition to an online course because not only could it potentially save a lot of time, but it could also keep students on the learning platform instead of using a 3rd-party solution to answer questions as is the suspected case right now with currently available chatbots.

To find out what this would look like, I researched a bit on potential LLM chatbot candidates, and came up with a plan to integrate one into a Canvas page.

Disclaimer!
This is simply a proof of concept, and is not in production due to certain unknowns such as origin of the initial training data, CPU-bound performance, and pedagogical implications. See the Limitations and Considerations section for more details.

How it works

The main powerhouse behind this is an open source, Large Language Model (LLM) called privateGPT. privateGPT is designed to let you “ask questions to your documents” offline, with privacy as the goal. It therefore seemed like the best way to test this concept out. The owner of the privateGPT repository, Iván Martínez, notes that privacy is prioritized over accuracy. To quote the ReadMe file from GitHub:

100% private, no data leaves your execution environment at any point. You can ingest documents and ask questions without an internet connection!

privateGPT, GitHub Site

privateGPT, at the time of writing, was licensed under the Apache-2.0 license, but during this test, no modifications were made to the privateGPT code. Initially, when you run privateGPT, train it on your documents, and ask it questions, you are doing all of this locally through a command line interface in a terminal window. This obviously will not do if we want to integrate it into something like Canvas, so additional tools needed to be built to bridge the gap.

I therefore set about making two additional pieces of software: a web-interface chat box that would later be embedded into a Canvas page, and a small application to connect what the student would type in the chat box to privateGPT, then strip irrelevant data from its response (such as redundant words like “answer” or listing the source documents for the answer) and push that back to the chat box.

A diagram showing how the front-end of the system (what the user sees) interacts with the back-end of the system (what the user does not see). Self-creation.

Once created, the web interface portion, running locally, allows us to plug it into a Canvas page, like so:

A screenshot showing regular Canvas text on the left, and the chat box interface on the right, connected to the LLM.

Testing how it works

To begin, I let the LLM ‘ingest’ the Ecampus Essentials document provided to course developers on the Ecampus website. Then I asked some questions to test it out, one of which was: “What are the Ecampus Essentials?”

I am not sure what I expected here, as it is quite an open ended question, only that it would scan its trained model data and the ingested files looking for an answer. After a while (edited for time) the bot responded:

A video showing the result of asking the bot “What are the Ecampus Essentials?”

A successful result! It has indeed pulled text from the Ecampus Essentials document, but interestingly has also paraphrased certain parts of it as well. Perhaps this is down to the amount of text it is capable of generating, along with the model that was initially selected.

A longer text example

So what happens if you give it a longer text, such as an OpenStax textbook? Would it be able to answer questions students might have about course content inside the book?

To find out, I gave the chatbot the OpenStax textbook Calculus 1, which you can download for free at the OpenStax website. No modifications were made to this text.

Then I asked the chatbot some calculus questions to see what it came up with:

Asking two questions about certain topics in the OpenStax Calculus 1 book.

It would appear that if students had any questions about mathematical theory, they could get a nice (and potentially accurate) summary from a chatbot such as this. Though this brings up some pedagogical considerations such as: would this make students less likely to read textbooks? Would this be able to search for answers to quiz questions and/or assignment problems? It is already common to ask ChatGPT to provide summaries and discussion board replies, would this bot function in much the same way?

Asking the chatbot to calculate things, however, is where one would run into the current limitations of the program, as it is not designed for that. Simple sums such as “1 + 1” return the correct answer, as this is part of the training data or otherwise common knowledge. Asking it to do something like calculate the hypotenuse of a triangle using Pythagorus’ theorem will not be successful (even using a textbook example of 32 + 42 = c2). The bot will attempt to give an answer, but its accuracy will vary wildly based on the data given to it. I could not get it to give me the correct response, but that was expected as this was not in the ingested documentation.

Limitations and Considerations

OK, so it’s not all perfect – far from it, in fact! The version of privateGPT I was using, while impressive, had some interesting quirks in certain responses. Responses were never identical either, but perhaps that is to be expected from a generative LLM. Still, this would require further investigation and testing in a production-ready model.

How regular and substantive interaction (RSI) might be affected is an important point to consider, as a more capable chatbot could impact the student-instructor Q&A discussion board side of things without prior planning on intended usage.

A major technical issue was that I was limited to using the central processing unit (CPU) instead of the much faster graphics processing unit (GPU) used in other LLMs and generative AI tools. This meant that the time between the question being sent and the answer being generated was far higher than desired. As of writing, there appears to be a way to switch privateGPT to GPU instead, which would greatly increase performance on systems with a modern GPU. The processing power required for a chatbot that more than one user would interact with simultaneously would be substantial.

Additionally, the incorporation of a chatbot like this has some other pedagogical implications, such as how the program would respond to questions related to assignment answers, which would need to be researched.

We also need to consider the technical skill required to create and upkeep a chatbot. Despite going through all of this, I am no Artificial Intelligence or Machine Learning expert; a dedicated team would be required to maintain the chatbot’s functionality to a high-enough standard.

Conclusion

In the end, the purpose of this little project was to test if this could be a tool students might find useful and could help them with content questions faster than contacting the instructor. From the small number of tests I conducted, it is very promising, and perhaps a properly built version could be used as a private alternative to ChatGPT, which is already being used by students for this very purpose. A major limitation was running the program from a single computer with consumer components made 3 years ago. With modern hardware and software – perhaps a first-party integrated version built directly into a learning management system like Canvas – students could be provided with their own course- or platform-specific chatbot for course documents and texts.

If you can see any additional uses, or potential benefits or downsides to something like this, leave a comment!

Notes

  1. Martínez Toro, I., Gallego Vico, D., & Orgaz, P. (2023). PrivateGPT [Computer software]. https://github.com/imartinez/privateGPT.
  2. “Calculus 1” is copyrighted by Rice University and licensed under an Attribution-NonCommercial-Sharealike 4.0 International License (CC BY-NC-SA).