1 of 60

Preparing Students to Work with AI

Anna Mills, English Instructor at Cañada College

A presentation for Kentucky Community & Technical College System

October 26, 2023

Licensed CC BY NC 4.0

2 of 60

Welcome!

I like to approach this in a spirit of inquiry. I’ve had the chance to learn about the topic it for the last year, but I don’t claim to have all the answers. It feels best to explore the subject with educators and students.

Slides (open for commenting):

https://bit.ly/KCTCSAI

3 of 60

What to expect

  • What is AI and how will it affect our students’ working lives?
  • Microlessons on critical AI literacy we can use with students
  • Classroom uses of text generation that also promote critical scrutiny of AI

4 of 60

What do we mean by “AI”?

“AI” or “Artificial Intelligence” in general is used these days to refer to computer systems that essentially uncover patterns in big data and follow those patterns to make predictions or recommendations. They use what’s known as “machine learning” to train until their predictions are more or less reliable.

When we say AI in higher ed these days, we are often talking about generative AI, systems that generate text, images, and video based on a user request.

5 of 60

Other kinds of AI: systems that make recommendations or take action based on patterns in past data (machine learning).

Such AI systems can be designed to (try to)

  • Autocomplete search terms
  • Populate social media feeds and suggest targeted advertising
  • Recognize faces
  • Recommend hiring or loan decisions
  • Direct an autonomous vehicle

6 of 60

Some basics about AI text generators like ChatGPT

  • The software is designed to generate a series of words based on its statistical analysis of huge volumes of text.
  • It works by copying patterns and predicting likely next words. 
  • What would the humans write next? How would a human writer likely continue on from the prompt?

Presentation by Anna Mills, licensed CC BY NC 4.0.

7 of 60

ChatGPT Alternatives

ChatGPT is just one chatbot that runs on OpenAI’s large language models, GPT-3.5 (free version) or the more sophisticated GPT-4 (premium).

  • Anthropic’s Claude.ai (can process longer documents in prompts)
  • Meta’s Llama 
  • Google’s Bard
  • Microsoft's Bing combines search with GPT-4
  • Poe.com allows you to use a number of different models (free and subscriber options):

8 of 60

What does text generation look like? Say we need to write a memo on our organization’s new emergency plan. ChatGPT gives a version of such a memo (probably not copied from the internet), and we can refine it with our own specifics.

9 of 60

I’m going to bracket the questions about policy and academic integrity that have dominated the initial discourse in higher ed to give us a chance to look at how AI may change the kind of preparation students need.

I’ll just say that the prevailing strong recommendation is to make policies about AI explicit and detailed. For more on the questions below, see my resource list with the Writing Across the Curriculum Clearinghouse

  • How to craft institutional and course-specific policies
  • Whether to use unreliable and sometimes biased AI detection
  • How to design assignments to discourage misuse of AI.

Presentation by Anna Mills, licensed CC BY NC 4.0.

10 of 60

Will AI automate the jobs students are preparing for?

As I understand it, that is largely still to be determined, partly by the decisions we make as a society about regulation of AI.

Frank Pasquale makes a case for regulation in The New Laws of Robotics: Defending Human Expertise in the Age of AI, Harvard University Press, 2020:

“Robotic meat- cutters make sense; robotic day care gives us pause.”

“Too many CEOs tell a simple story about the future of work: if a machine can do what you do, your job will be automated. They envision everyone from doctors to soldiers rendered superfluous by ever-more-powerful AI. They offer stark alternatives: make robots or be replaced by them. Another story is possible. In virtually every walk of life, robotic systems can make labor more valuable, not less.”

11 of 60

What we know: professionals will be working with AI

“By 2030, activities that account for up to 30 percent of hours currently worked across the US economy could be automated—a trend accelerated by generative AI. However, we see generative AI enhancing the way STEM, creative, and business and legal professionals work rather than eliminating a significant number of jobs outright. Automation’s biggest effects are likely to hit other job categories. Office support, customer service, and food service employment could continue to decline.” –McKinsey Global Institute, “Generative AI and the future of work in America,” July 26, 2023

12 of 60

“Gen AI and other recent trends affecting the labor market point to the importance of upskilling, reskilling, and training to empower the global workforce.” - McKinsey Podcast, September 21, 2023

13 of 60

Chatbots may not be the future of text generation.

With the next generation of AI user interfaces, generative AI is present within familiar applications and might be used in myriad ways throughout a task.

14 of 60

Google now offers beta text generation within Google Docs

Nowhere in version history does it indicate which text is auto-generated and which is written by the user.

15 of 60

Let’s take a look at what it’s like to work with AI in Google Docs.

  • “Help me write”
  • I ask it for a memo on an updated emergency plan.
  • I ask it to elaborate, change the tone, tailor it.
  • I insert the text into the document.
  • Then, anytime, I select text and request AI revision.

On the right is a screenshot of options for auto-revising the AI text.

16 of 60

Microsoft is releasing Copilot, “Your everyday AI companion”

“Microsoft 365 Chat combs across your entire universe of data at work, including emails, meetings, chats, documents and more, plus the web. Like an assistant, it has a deep understanding of you, your job, your priorities and your organization. It goes far beyond simple questions and answers to give you a head start on some of your most complex or tedious tasks — whether that’s writing a strategy document, booking a business trip, or catching up on emails.”

17 of 60

So how do we prepare students now to work with AI in future?

How do you work with generative AI?

  • You ask it for what you want.
  • Then you question what it gives you. You revise, reject, add, start over, tweak.

How do you know how to change what it gives you?

  • You develop critical thinking, reading, and writing skills.
  • You develop subject-matter expertise.

18 of 60

Critical thinking, reading, and writing skills?

Subject-matter expertise?

Hmm, I think that’s what we do in higher ed…

So let’s keep teaching those things! They will be needed in new work contexts.

19 of 60

Should we be teaching “prompt engineering”--techniques for getting the most out of current systems like ChatGPT?

User interfaces are changing so quickly that any current technical aspects of prompting may go away.

“Despite the buzz surrounding it, the prominence of prompt engineering may be fleeting. A more enduring and adaptable skill will keep enabling us to harness the potential of generative AI? It is called problem formulation — the ability to identify, analyze, and delineate problems.” Aguz A. Acar, “AI Prompt Engineering Isn’t the Future,” Harvard Business Review, June 6, 2023

“Identify, analyze, and delineate problems”? We’re back to critical thinking, no?

20 of 60

“Prompt engineering” may not be essential in the long run.

But our guidance is still important

Students do need to understand some basic things about AI.

21 of 60

We don’t know through what interface we’ll interact with AI in a year. But we can be sure students need to know that they need to question what AI gives them to “add value.”

  • What language models are and aren’t (not sentient), 
  • How to identify the problems in AI outputs 
  • How they/we/the public might help shape future policy on AI. 

Digital literacy needs to include critical AI literacy

22 of 60

Goals around AI in higher ed

  • Boost student ability to recognize the flaws in text generation AI (for practical workplace reasons and also for their own information literacy and civic participation)  
  • Boost student confidence and sense of the value of their voice and human judgment.

23 of 60

What (if anything) do you tell students about AI? Is there anything you think to yourself about AI that you don't tell them?

�Scan the QR code 

Or, go to https://www.menti.com/al7nmcnexgqv (in the chat)�Or, go to Menti.com and enter 43 89 24 1

24 of 60

Critical AI Literacy Microlessons

Teach one or more microlessons on

  • What AI is and isn’t
  • How it can mess up

25 of 60

AI Microlessons  Start building a foundation of critical AI Literacy before we invite students to use generative AI

26 of 60

How do we start? Baby steps?

Watch a very short video?�

27 of 60

Perhaps assign an interactive reading?

Let Us Show You How GPT Works — Using Jane Austen from The New York Times gives readers the choice of seeing rough language model generations in the style of Harry Potter, Star Trek: The Next Generation, Shakespeare, or Moby Dick.

Sample: “Hermione raised her wand. "Professor Dumbledore never mimmed Harry."

He looked back at the room, but they didn't seem pretend to blame Umbridge in the Ministry. He had taken a human homework, who was glad he had not been in a nightmare bad cloak. Her face looked over her closely past her and saw Harry crossed the grip, looked down at the wall. "Come off!" she said tentatively, with a crumpled note into his own hand on her book. "That’s beating attacks how we’ve got detentions or not to realize how she did the Maps worse doesn’t want.”

28 of 60

Assign a short reading like the introduction to Elements of AI, a set of free online course materials from the University of Helsinki.

29 of 60

Artificial Intelligence: Last Week Tonight with John Oliver is well researched.  Caveat: there’s a bit of swearing.

30 of 60

Consider a microlesson on one particular aspect of critical AI literacy

  • Privacy
  • Fabrication/hallucination
  • Bias
  • Lack of understanding (not sentient)

Consider also 

Labor, AI colonialism, intellectual property, environmental impact

31 of 60

A microlesson on privacy

(Inspired by Autumn Caines’ blog on annotating the privacy policies)

Share highlights from the OpenAI Privacy Policy (or another model's policy) with students

“We may use Personal Information for the following purposes:

  • To provide, administer, maintain and/or analyze the Services;
  • To improve our Services and conduct research;
  • To communicate with you;
  • To develop new programs and services…”

Image by GDJ on Pixabay.com

32 of 60

A microlesson on fabrication (also called hallucination)

  • Share a chat session where the chatbot made something up
  • Show a credible source with the accurate information
  • Remind students it is producing plausible text, not checking for truth.

This Photo by Unknown author is licensed under CC BY-SA.

33 of 60

Here's a session where ChatGPT attributes a quote to Mark Twain that is actually a popular quote supposedly by Abraham Lincoln. It’s a paraphrase.

“Whenever I hear anyone arguing for slavery, I feel a strong impulse to see it tried on him personally."--popularly attributed to Lincoln online.

The actual Lincoln quote: “While I have often said that all men ought to be free, yet would I allow those colored persons to be slaves who want to be, and next to them those white people who argue in favor of making other people slaves. I am in favor of giving an appointment to such white men to try it on for these slaves.” (UCSB American Presidency Project)

34 of 60

A microlesson on AI bias

Show a video that gives an example of racism and sexism compounded by AI

    • Algorithms of Oppression video with Safiya Umoja Noble (USC)
    • Gender Shades with Joy Buolomwini (MIT)

35 of 60

Another microlesson on bias: Invite students to read “How AI reduces the world to stereotypes” by Victoria Turk in restofworld.org

36 of 60

Biased Barbies

“THEY GAVE SOUTH SUDAN BARBIE A GUN”

Business Insider reporting on the Buzzfeed article “A list of AI-generated Barbies from 'every country' gets blasted on Twitter for blatant racism and endless cultural inaccuracies”

37 of 60

“They turned Thailand Barbie Russian”--@sighyam

38 of 60

“A few Middle Eastern Barbies wore a ghutra, a traditional headdress for men.”--Business Insider

39 of 60

A microlesson on AI'ls lack of understanding: share a chat session that suggests ChatGPT isn’t thinking, such as this session where it counts wrong

40 of 60

Another way to demonstrate ChatGPT’s lack of real understanding: “correct” its right answer and watch it agree. Here I told it it had rounded the decimal incorrectly for ⅓.

41 of 60

What approach have you taken or might you take to introducing text generation AI and/or key elements of critical AI literacy? 

�Scan the QR code 

Or, go to https://www.menti.com/alirvxz51kke (in the chat)�Or, go to Menti.com and enter 5628 5048

42 of 60

Options for incorporating AI into our pedagogy

43 of 60

Combine AI literacy and discipline skills

Plan learning activities involving AI that

    • support existing learning goals
    • give students a chance to practice identifying AI limitations, risks, ethical concerns.

44 of 60

One approach: critique AI performance on specific skills we want to teach

This could involve students 

  • turning in their reflections, 
  • posting them in a discussion, or 
  • annotating a chat session transcript.

We might 

  • give them the AI output or
  • invite them to generate it.

45 of 60

Clarity

  • This sounds plausible because ______________, but it doesn't really make sense because ______________.
  • This sounds good, but it doesn't really fit the purpose. What we are looking for is ______________.

Accuracy

  • This is inaccurate because ______________.
  • The AI seems to have misinterpreted ______________.*

46 of 60

Juan Pablo Pardo-Guerra at University of California San Diego has documented his approach in TextGenEd: Teaching with Text Generation Technologies from the Writing Across the Curriculum Clearinghouse

"...[S]tudents are provided with an AI-generated text relevant to a course’s topics and focus and then asked to comment, review, and expand on it using a feature such as "track changes."  In engaging with the AI-generated text, students review their knowledge, offer critiques, modify theoretical and empirical claims, and provide concrete examples that illustrate or disprove the provided answer.... additionally, they gain some AI literacy in evaluating the AI-generated text."

47 of 60

Critique AI summaries

Professional Writing for Healthcare: Writing & Revising Research Summaries with Artificial Intelligence by Heidi A. McKee in TextGenEd

Translating a Policy Document into Plain English by Timothy Laquintano, Lafayette College in TextGenEd

PerplexityAI integrates chatbot and search features and doesn't require sign-in. Like Bing Chat and Google's Bard, it will provide brief summaries of sources it links to. Ask students to read the sources and find inaccuracies in the summaries.

48 of 60

A bit of fun: critique summaries generated in tailored styles

In their TextGenEd contribution, Daniel Hutchinson and Erin Jensen of

Belmont Abbey College invite students to ask a language model to explain a difficult passage or concept using an example that incorporates the student’s pop culture reference of choice

Then they ask students to evaluate how well the model’s example illustrates the meaning of the passage. 

49 of 60

Critique an AI critical assessment

Students reflect on the differences between a ChatGPT critical assessment and a human-written one about the same New York Times article.

50 of 60

What did ChatGPT miss? What did its output get right?

How do those observations match what we learned about how language models work?

How might the student essay have turned out if the student had started with the ChatGPT output and revised from there?

What lessons do we draw from this comparison?

51 of 60

Critique AI "research" 

Have students fact-check AI-generated claims and references to sources as a way to help them learn about academic research and peer review.

In my first-year writing class, we just did this through collaborative annotation of a chat session.

52 of 60

I asked students to check if sources were real and accurately described

ChatGPT output from a session on surprising AI facts:

AI Can Decode Ancient Scripts:

  • Fact: Researchers used AI to decipher the ancient script known as Linear B.
  • Source: Jucha, M. A., Arjovsky, M., & Bengio, Y. (2017). Decipherment of the Linear B script with deep learning. arXiv preprint arXiv:1708.08731.
  • Credibility: arXiv is a repository for electronic preprints, and while not peer-reviewed, many significant findings are first reported here. Yann Bengio, one of the authors, is a Turing Award winner.

No such paper. No such author. (Yann Bengio mixes Yann LeCun and Yoshua Bengio, two famous AI scientists.)

53 of 60

Critical reflection on supplementary AI feedback 

Using LLMs as Peer Reviewers for Revising Essays by Antonio Byrd, University of Missouri-Kansas City in TextGenEd:

"Revise your essay using a large language model (LLM) such as ChatGPT.  First, revise your essay in response to peer review comments from myself and your classmates. 

Cast careful judgment on the responses from the LLM, as the analysis may include misinformation or show that the LLM did not understand the intent of your prompt command."

54 of 60

I am advising on an app I am still getting permission to pilot:

MyEssayFeedback.ai

55 of 60

Perhaps AI can supplement teacher, tutor and peer feedback without replacing them

  • Peer and instructor feedback are still essential: they represent a human audience receiving the communication. Without that, what’s the point?
  • But trying out different feedback requests might help students reflect on how they feel, what kind of help they want, and what they want to do with their writing.

56 of 60

Guardrails and guidance are needed: language model feedback can sometimes be bad while sounding authoritative.��With guidance, though, students can practice recognizing points they disagree with in the AI feedback and points that don’t represent their essays accurately.  

57 of 60

Guardrails and nudges

  • System instructions prohibit rewriting student text or suggesting new text.
  • Reminders are built in to question the AI suggestions and check in with peers, instructor, tutors.
  • Links to Template Phrases for Reflecting on AI Feedback from my textbook.

58 of 60

Sample template phrases for reflecting on AI feedback (From How Arguments Work)

Reflect on ways the feedback doesn't fit our meaning or purpose

  • The AI feedback suggests _____________, but I'm not sure this is what I want to do because _____________.
  • The AI feedback seems to assume that I am claiming _____________ when in fact I meant to say _____________.
  • The AI feedback suggests a misunderstanding of the section of my essay about _____________. I was aiming to convey _____________.*

59 of 60

What first step(s) have you taken or might you take toward preparing students to work with AI? ​

Click the right arrow if you're in the Mentimeter already�Scan the QR code 

Or, go to https://www.menti.com/al7nmcnexgqv (in the chat)�Or, go to Menti.com and enter 43 89 24 1

60 of 60

Here’s to open discussion about AI among educators!

Resource list:https://bit.ly/AITextEdu

�Get in touch:

annarmills.com

Twitter/X: @EnglishOER

LinkedIn: anna-mills-oer

Slides open for commenting: https://bit.ly/KCTCSAI

�This presentation is shared under a CC BY NC 4.0 license.