1 of 48

How Teachers Can Harness AI in Our Work

A workshop for the Center for Learning and Teaching at the American University in Cairo and Equity Unbound

Anna Mills, English Instructor at Cañada College

October 24, 2023

Licensed CC BY NC 4.0

2 of 48

Welcome!

Please share your name, college, and

  • 1-2 words on how your day has gone/is going.
  • 1-2 words on how you currently feel about using AI yourself (all feelings welcome)

Slides, open for commenting:

https://bit.ly/HarnessAIAUC

3 of 48

What to expect

  • Some ground rules for using AI text generators like ChatGPT
  • Prompting: how to get the most out of AI
  • Ways teachers can use these systems
  • Time to experiment and ask questions
  • Assorted other uses for AI

4 of 48

5 of 48

Is that AI hype? Let’s see.�Before we use a language model, some caveats and ground rules

  • Disclosure
  • Privacy
  • Verification
  • Critique

Presentation by Anna Mills, licensed CC BY NC 4.0.

6 of 48

Disclosure��Rule of thumb: If you would credit a human for similar assistance, credit ChatGPT (and remember the unknown human authors it was trained on)

Refer to MLA guidelines on citing generative AI

Image by Monstera on Pexels.com

7 of 48

Sample ChatGPT text in quotes with an asterisk, integrated into my textbook section on template phrases for critiquing AI outputs

8 of 48

Sample disclosure from my textbook

9 of 48

Privacy: Don’t share sensitive or private information with ChatGPT, especially not student writing or data

From the OpenAI Privacy Policy

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

10 of 48

Verification: Check all ChatGPT outputs

  • Only use it when you have enough expertise to check
  • From OpenAI’s message to educators: “Verifying AI recommendations often requires a high degree of expertise.”
  • If you don’t have the expertise, do you have another way to independently verify the output?
  • Do you have time to evaluate the output?

Image by OpenClipart-Vectors on Pixabay

11 of 48

Remember, the algorithm is designed to play a guessing game. ��What would the humans write next? How would a human writer likely continue on from the prompt?

Statistical prediction based on human data

Researcher and “AI humorist” Janelle Shane Twitter profile photo

Presentation by Anna Mills, licensed CC BY NC 4.0.

12 of 48

Check for bias as well as inaccuracy

  • These systems are trained to copy patterns in all the human-written text they were trained on.
  • Companies claim they needed such huge volumes of text that they couldn’t exclude biased text or try to make the corpus more representative than the internet.
  • Biases can be mitigated but are still baked in, impossible to remove.

a

Presentation by Anna Mills, licensed CC BY NC 4.0.

13 of 48

Wait, should we even be using ChatGPT or other AI text generators?

14 of 48

A host of other concerns surround the training and use of language models (See Teaching AI Ethics by Leon Furze)

Presentation by Anna Mills, licensed CC BY NC 4.0.

15 of 48

I continue to wrestle with these questions, but I am using it for certain things because

  • I’m fascinated
  • It’s useful
  • I want to understand what my students have access to
  • I want to understand how the writing environment is changing as professionals use language models more commonly

16 of 48

Using ChatGPT isn’t necessarily “phoning it in.” We can use it to help us think and to push our thinking, not to skip thinking

I first saw articulated this approach articulated by Kyle Booten, an English professor at University of Connecticut at AI FUTURES: An Interdisciplinary Conversation on LLMs and the Future of Human Writing

Presentation by Anna Mills, licensed CC BY NC 4.0.

17 of 48

Prompt it to reflect back and draw out our ideas

  • Ask it to ask us questions to help us think something through
  • Ask it to give feedback

18 of 48

Or ask it for help with the parts of the work that aren’t about developing core ideas, for example,matching specific formats and styles

  • Give it an example or template
  • Give it raw ideas and ask it to translate them�
  • Course outlines
  • Lesson plans
  • Rubrics
  • Grant proposals
  • Reports

19 of 48

Techniques for getting the most out of ChatGPT

Image by kaboompics on Pixabay

20 of 48

“Prompt engineering”? It doesn’t have to be technical.�Just try.. No arcane knowledge needed!

  • When we want to get somewhere in conversation, we try things
  • We think about how to draw out our interlocutor
  • We consider scaffolding to help the student
  • We have these skills as educators, colleagues, and managers!
  • Let’s be daring and playful 

Image courtesy of Shafin_Protik on Pixabay.com

21 of 48

Simple, but maybe not obvious

  • Try again, or ask it for multiple answers so you can choose.
  • Tell it what it gets wrong; ask it to redo.
  • Give it what you have to go on (bullet points, sources, text to respond to, examples)
  • Take it step by step through a process. Ask it to explain its reasoning for each step.

Photo by Wes Hicks on Unspliash

22 of 48

You’re directing; it’s the actor

  • Give it an example of the style you want:�“Can you make that more friendly?”�
  • Tell it to play a particular role such as a coach, mentor teacher, administrator, or editor (see Rob Lennon’s post about this): �“You are a supportive editor.”�
  • Tell it to be unbiased (this doesn’t always work but sometimes helps): �“Make sure to give an unbiased response.”

Image by OpenClipart-Vectors on Pixabay

23 of 48

Get ChatGPT to help you get the most out of it

  • Ask it to check or give feedback on its own outputs: �Are there any flaws in X?”�
  • “How can you help me write X? What information do you need?” (Ethan Mollick’s suggestion)

Presentation by Anna Mills, licensed CC BY NC 4.0.

24 of 48

Some recommend complex, multilayered prompts that give detailed instructions and context

Rob Lennon’s popular thread Anatomy of a ChatGPT MegaPrompt

25 of 48

More on getting the most out of ChatGPT and other AI systems

Presentation by Anna Mills, licensed CC BY NC 4.0.

26 of 48

But don’t get hung up on making the fanciest prompt! Many use simpler prompts successfully.

27 of 48

A dialogue with the chatbot can help us clarify and articulate what we are after

28 of 48

Even a general request for suggestions might be fruitful. In this case, one of the five suggestions resonated with me (using bullet points).

29 of 48

Let’s see this in action:

Some uses for writing and language teachers

30 of 48

Let’s see some of this in action

  • Course planning
  • Lesson planning
  • Assignment design
  • Explanations
  • Examples

31 of 48

Language instruction: explanations (a Poe chat excerpt)

32 of 48

Language instruction: examples tailored to topics of interest to students (a Poe chat excerpt)

33 of 48

I’m going to talk through some highlights of the chat transcripts below. Then I’ll invite reflections on what stands out to you either in how I prompted the system or in the responses I got.

34 of 48

What stands out to you in these chat sessions?

What questions or comments do you have?

  • Go to https://www.menti.com/albsgkpmaxep (in the chat)
  • Or, go to Menti.com and enter 6614 8432 to rate the options

35 of 48

A few other ways we might use language models

36 of 48

Ask it to help you identify discrepancies, like an auditor (see this sample prompt for a comparison of a state and local outline for the same course )

Presentation by Anna Mills, licensed CC BY NC 4.0.

37 of 48

Ask it for editing and style feedback

Sample prompt: “Does this email come across as friendly? List sentences that might be read as negative, cold, or hostile. For each, suggest a more friendly alternative.”

See an example of ChatGPT’s suggestions

  • Ask it to list and explain its suggestions rather than just making changes
  • Give it a sample of the tone and style you are aiming for�

Don’t let it take control. Consider each suggestion.

38 of 48

�For example, I asked ChatGPT for suggestions on how to reword the category descriptions in my resource list�

My prompt: “Below are the categories in a resource list on AI text generators and writing in higher education. For each category, could you list the original version, two alternate titles, and a discussion of why each alternate title might be preferable?”

Sample from the output:� “Original: Scholarship

Alternate 1: Academic Research on AI Text Generators

Alternate 2: Scholarly Works on AI-based Text Generation”

See more of ChatGPT’s suggestions

Presentation by Anna Mills, licensed CC BY NC 4.0.

39 of 48

AI for Learning Designers by Dr. Philippa Hardman

  • #1: Use AI to Optimise Your Feedback”�Hardman uses AI to revise feedback on students work to be more positive and motivating.

  • #2: Use AI to Optimise Your Learning Objectives”�She asks ChatGPT to use Bloom’s taxonomy and include “so I can…” statements with each learning objective.

40 of 48

Uses for Research: These apps build on language models combined with search to help find and summarize sources

41 of 48

Poe.com’s web search did fairly well in this trial on Gaza aid (Reuters, Al Jazeera, The Guardian)

42 of 48

Cautions on AI research tools

  • Sometimes the summaries of sources are wrong.
  • Sometimes the system ”misunderstands” the question.
  • Here Bing answers a different question than the one I asked. The links it gives are not relevant to my question.

Image by cripi on Pixabay

43 of 48

ChatGPT is designed to mimic patterns. Automate the purely formal…but check!

  • Have it convert APA to MLA citations
  • Ask it to clean up a transcribed talk or meeting notes
  • Ask it to check formatting

Presentation by Anna Mills, licensed CC BY NC 4.0.

44 of 48

Technical assistance? It might have very good suggestions. And again, it might be wrong.

I now turn to ChatGPT or Claude when I’m looking to figure out a technical problem though I take what it says as unreliable. See this sample chat.

45 of 48

Here I explain what I want to know about Google Sheets data and ask for a formula

Presentation by Anna Mills, licensed CC BY NC 4.0.

46 of 48

What are some tasks you have coming up? Would you want to use a language model to help with any?

  • Rubric
  • Assignments
  • Lesson plans
  • Peer or self evaluations
  • Email
  • Funding requests
  • Job descriptions
  • Program reporting

Presentation by Anna Mills, licensed CC BY NC 4.0.

47 of 48

Now we all try? Share at

https://bit.ly/GenAINotes

Choose one kind of assistance you would like from a language model, make an account if you need to, and try out a prompt. Share your notes and takeaways on how it went! https://bit.ly/GenAINotes

  • ChatGPT (chat.openai.com)
  • Poe.com (try a range of different models and compare)
  • Anthropic’s Claude.ai (can process longer documents in prompts)
  • Google’s Bard (bard.google.com)
  • Microsoft's Bing (Bing.com, choose chat)

48 of 48

Let’s share–what have you tried? What has worked? What hasn’t? Tips? Questions?

�Anna Mills�annarmills.com

X/Twitter: @EnglishOER

LinkedIn: anna-mills-oer

Slides: https://bit.ly/HarnessAIAUC�This presentation is shared under a CC BY NC 4.0 license.