1 of 76

Generative AI & Writing

Tamara Tate

Digital Learning Lab

This material is based on work supported by the National Science Foundation under Grant No. 23152984

2 of 76

Have you used GenAI?

Mentimeter: Menti.Com 6504 9325

3 of 76

How are you feeling about generative AI?

Digital Learning Lab

4 of 76

The foundation

Digital Learning Lab

5 of 76

What is generative AI?

ChatGPT and other large language models (LLMs) are a type of artificial intelligence that are trained to generate text similar to human-generated text. They are called “large” because they are typically trained on a corpus of text data ranging from millions to billions of words.

Digital Learning Lab

6 of 76

Digital Learning Lab

7 of 76

What is generative AI?

In spite of its startling capacities, generative AI’s serious flaws are also evident, including the risk of data privacy breaches, unclear intellectual property protection, algorithmic biases that replicate biases in the data, and ”hallucinations” when generative AI may give completely wrong answers.

Digital Learning Lab

8 of 76

Where to start?

  • Try something personal: “Suggest 5 recipes for dinner tonight using chicken and broccoli” Stir fry it is!
  • Try something for your class: “What are 5 different metaphors for explaining logarithms to undergraduate students in math class?” Hmmmm, maybe the password decoder?
  • Try out one of your assignments to see what generative AI would produce: “Write a 5 word poem on what you did this summer.” I’m a machine; I pondered.

Need help? See “Setting up ChatGPT

Digital Learning Lab

9 of 76

Other thoughts?

Mentimeter: Menti.Com 6504 9325

10 of 76

Where does generative AI fit into writing?

Digital Learning Lab

11 of 76

Where does generative AI fit in?

Generate prompt

Topic Selection

Summarizing Content

Review/Feedback

Outlining

Digital Learning Lab

12 of 76

But use of AI is filled with contradictions

  • The rich get richer
  • With or without
  • Imitation
  • June/July

Digital Learning Lab

13 of 76

The “rich get richer” contradiction

AI is incredibly powerful for assisting communication

To get the most out of it, you need to know how to prompt it well, critically evaluate its output, and edit and incorporate it into your work

Digital Learning Lab

Digital Learning Lab

14 of 76

The “with or without” contradiction

If students never learn to use AI, they will be at a disadvantage in their study and careers.

If they use AI too much and too early, they will also be at a disadvantage as they will be robbed of foundational skills necessary to use it well.

Digital Learning Lab

Digital Learning Lab

15 of 76

Imitation contradiction�

English learners and developing writers are constantly told they need to imitate native speakers’ or expert writers’ language use.

But when they borrow exact phrases from sources, they can be accused of plagiarism. Sometimes even paraphrasing is frowned upon.

Digital Learning Lab

Digital Learning Lab

16 of 76

June/July contradiction�

Digital Learning Lab

Digital Learning Lab

17 of 76

What other issues and concerns do you have?

  • Bias (GPT-4, 3.5 and 3 exhibit cultural values resembling English-speaking and Protestant European countries, Tao et al., 2023)
  • Privacy
  • Intellectual property
  • Environmental

…….

Mentimeter: Menti.Com 6504 9325

Digital Learning Lab

18 of 76

Disclaimer in every assignment that uses AI

What Students Need to Know

Students should have reviewed the Understand: How LLMs work curriculum and have an understanding of the Biases and Limitations of generative AI.

Remember:

  • The AI doesn’t think or understand. It does not have opinions and cannot make value judgements. It’s just predicting the next bit of text given what you told it and what it has been trained on.
  • Do NOT put sensitive personal information or confidential information into ChatGPT.
  • Hallucinations, biases, and other concerns: You are responsible for the accuracy and appropriateness of anything you use or incorporate into your own text.

Digital Learning Lab

19 of 76

Why should I bother?

  • Prepare students for the world they are living in
  • Ensure equitable access to quality education on using AI
  • Use as an opportunity to center critical thinking and corroboration/interrogation of information

Digital Learning Lab

20 of 76

Post-Secondary Students’ Use and Perspectives of ChatGPT

Clare Baek, Tamara Tate, & Mark Warschauer (In progress)

Survey of U.S. Students in Post-Secondary Education (N=1001)�September 2023

Use for Writing?

Native Speakers

English learners

At least weekly

19.2%

39.5%

At least daily

4.5%

13.2%

Several times a day

2.3%

5.3%

21 of 76

Best available human standard

Would

  • the best available AI
  • in a particular moment,
  • in a particular place,

do a better job solving a problem than the

  • best available human that is actually able to help
  • in a particular situation?

--Ethan Mollick, Oct. 22

Digital Learning Lab

22 of 76

Other thoughts?

Mentimeter: Menti.Com 6504 9325

23 of 76

What do we know?

  • Prior research on digital technology integration into schools
    • Laptops/1-1 devices
    • Automated writing evaluation
    • Online learning

BUT: Prior technology not as powerful or rapidly diffused

Digital Learning Lab

24 of 76

Understand

  • Only the basics, BUT enough to know
    • It’s built on prediction & language
    • It doesn’t think
    • It fabricates things rather than say “I don’t know”
    • It was trained on a specific set of texts & by humans, so very real biases
  • Impacts prompting
  • Creates bias
  • Need for corroboration

Digital Learning Lab

25 of 76

Understand

Digital Learning Lab

26 of 76

Access & navigate

  • Literally, able to connect to the tool

{shout out to the IT team!}

  • Know which tool to use for what
  • Know how to navigate the tool
  • Know how to protect the privacy of information when using the tool

Digital Learning Lab

27 of 76

3 lessons learned (so far)

Digital Learning Lab

28 of 76

Prompt

  • Technical, prompting-specific knowledge
    • Personas
    • Emotion
    • Details
  • Content knowledge
    • What question to ask
    • Key words and phrases
    • (Also helps recognize fabrications)

Critical new skill or flash in the pan?

Digital Learning Lab

29 of 76

Corroborate

  • AKA sourcing (history), good research skills (librarians)
  • Critical habit to get into in the current environment, not all “facts” are equal

FYI: Quality of AI feedback is very close to the quality of human feedback. It appears to be similarly good at scoring writing on a rubric (in process research).

Digital Learning Lab

30 of 76

Incorporate

  • Transparency
  • Citation
  • Academic integrity
    • No reliable detector
      • Not even humans
    • Easy to evade detectors w/ prompting
    • Especially unreliable for non-native English speakers
      • Presumably dialects would also be problematic

Digital Learning Lab

31 of 76

GPT Detectors are Biased against Non-Native English writers (Liang et al, 2023)

Used 7 GPT Detectors on

  • Real TOEFL Essays
  • Real US 8th-grade essays

(All pre ChatGPT release)

32 of 76

How to talk about generative AI to your class, team, department ….

  • Providing support and managing risk
  • Evaluating learning objectives and assessment
  • Addressing ethical concerns

Adapted from Daniel Stanford

Digital Learning Lab

33 of 76

How to talk (& think) about generative AI

  • Providing support and managing risk
    • What support do you need to learn more about AI and feel empowered?
    • What obstacles need to be removed?
    • What might backfire?
    • What risks would you be taking on?

Adapted from Daniel Stanford

Digital Learning Lab

34 of 76

Disability considerations

  • You need to have a clear, transparent expectation on student use of generative AI in your class.
  • Make sure it is covered in your syllabus.
  • Discuss regularly.
  • For help and examples, see my blog on this.

Digital Learning Lab

35 of 76

How to talk (& think) about generative AI

  • Evaluate learning objectives and assessments
    • What skills might be lost if students rely too heavily on AI?
    • What skills might be essential for students in the future?
    • Should course/program learning objectives be updated? What might that process/timeline look like?
    • Which assessments are solid as is? Which might need minor adjustments or a major redesign?
    • What strengths and weaknesses can we identify for specific assessment methods?

Adapted from Daniel Stanford

Digital Learning Lab

36 of 76

And that is the really hard part

  • When?
  • Who?
  • How much?
  • In what ways?

While balancing all the contradictions

Digital Learning Lab

37 of 76

And now the really hard part

  • When?
  • Who?
  • How much?
  • In what ways?

Digital Learning Lab

38 of 76

  • Researcher
  • Instructor
  • V.P. Teaching & Learning
  • Researcher

39 of 76

Implementing generative AI in a course

Process:

  1. Identifying key takeaways for students to learn about GenAI and understanding what GenAI can and can’t do

Researchers came up with types of AI activities based on what we had learned about course, goals, AI

Digital Learning Lab

40 of 76

Implementing generative AI in a course

Process:

  1. Identifying key takeaways for students to learn about GenAI; understanding what GenAI can and can’t do
  2. Connecting with existing course learning objectives; understand course assignments, tone, content

Then team aligned AI w/ appropriate assignments where the use supported learning goals

Pesky Professor; executive summary

Project 1 and 2

Pesky Professor; executive summary

Digital Learning Lab

41 of 76

We always start with the course goals

Digital Learning Lab

42 of 76

What this looks like: Sample mini lessons

Digital Learning Lab

43 of 76

What this looks like: Sample mini lessons

Digital Learning Lab

44 of 76

UCSD Slide Deck

  • Evaluating learning objectives and assessment

Image source: UCSD

Digital Learning Lab

45 of 76

How to talk (& think) about generative AI

  • Address ethical concerns
    • How can we support AI literacy in a way that is ethical and equitable?
    • When might AI use be forbidden, discouraged, encouraged, or required?
      • Why/how might this vary for specific courses, assignments, students, or instructors?
    • How might we address concerns around data privacy, unpaid labor, and equal access?

Adapted from Daniel Stanford

Digital Learning Lab

46 of 76

Interdisciplinary uses for Instructors

  • Idea generation, refinement
  • Generate metaphors, culturally diverse examples of concepts
  • Drafts of lesson plans, quiz questions, writing prompts
  • Creation of models: e.g., an essay that uses mostly passive voice
  • Draft emails, administrative reports
  • Feedback on writing

Consider whether your use aligns with the uses allowed for your students. Why or why not? Are you transparent about this?

Digital Learning Lab

47 of 76

Interdisciplinary uses for students

  • Studying: create self-testing, plan for studying, practice questions
  • Feedback: use like peer feedback for suggestions
  • Generate metaphors, examples to explain difficult concepts
  • Generative AI as a thinking partner (example of what it can do with prompting, simulated middle school student)
  • Topic generation, refinement

Feedback: Our research shows within 1 point of humans

Digital Learning Lab

48 of 76

Research in process

Generative AI -- Holistic Scoring of Essays

Research Questions

  1. How internally consistent is ChatGPT scoring?
  2. How does ChatGPT scoring’s agreement with human raters compare to reliability between human raters on the same data?
  3. Do scores systematically vary by English language status differently for ChatGPT ratings than for human ratings?

Holistic Scoring

Holistic Scoring/ Grading

Digital Learning Lab

49 of 76

Research in process

Highlights

  • Multiple scores by AI were more consistent than multiple scores by humans
  • Two human scores exactly agreed 49%; within 1 point 74% in Sample 1
  • AI-human exact agreement was 19-42%; within 1 point 76-89%
  • AI and human scoring of English learners was not significantly different

Holistic Scoring

Holistic Scoring/ Grading

Digital Learning Lab

50 of 76

Research in process

Generative AI -- Holistic Scoring of Essays

Agreement within 1 point: Close enough?

Holistic Scoring

Note: Sample 3 was done on a newer version of GPT 3.5

All

Non English Learner

English Learner

Sample 1: Human v. Human

74%

73%

80%

Sample 1: Human v. AI

76%

77%

75%

Sample 2: Human v. AI

83%

NA

NA

Sample 3: Human v. AI

89%

89%

92%

Holistic Scoring/ Grading

EL better

Digital Learning Lab

51 of 76

Research in process

Generative AI -- Feedback on Essays

  • Formative feedback improves writing
  • Secondary student source-based, argumentative history essays (n=200), with a focus on including sufficient numbers of English learners to analyze heterogeneous differences
  • Already received human feedback from well-trained human experts on these essays

Feedback

Digital Learning Lab

52 of 76

Research in process

Generative AI -- Feedback on Essays

  • Deductively coding human and AI feedback,
  • Across research-based discrete components
    • Criteria based--referencing or alluding to rubric components
    • Clarity--specific, actionable, clearly described
    • Accuracy--looking at the essay, is it true?
    • Developmentally appropriate--is this the next important thing to work on?
    • Supportive tone
  • Heterogeneity by essay quality and EL status of the student writers

Feedback

Digital Learning Lab

53 of 76

Pretend you are a secondary school teacher. Provide 2-3 pieces of specific, actionable feedback on each of the following essays written to this rubric that highlight what the student has done well and what they could improve on. Use a friendly and encouraging tone. If needed, provide examples of how the student could improve the essay.

54 of 76

Great job on providing a brief background on the Montgomery Bus Boycott and highlighting the key players involved. You have also provided some reasons why the boycott succeeded, such as the loss of revenue for the bus company and the strong leadership of King, MIA, NAACP, WPC, and churches. To improve your essay, consider providing more specific examples to support your points. For instance, you could mention how much revenue the bus company lost during the boycott or provide specific actions taken by the leaders and organizations involved. Additionally, you could expand on the impact of the boycott beyond Montgomery and how it influenced the Civil Rights Movement as a whole.

You did a wonderful job explaining what the Montgomery Bus Boycott was historically and you have a strong claim toward the end of your first paragraph.: “The Bus Boycott has succeeded because most of the people who rode Montgomery's buses were African Americans and this made the company lose a lot of business and revenue from the boycott so this left them with no easy solutions to fix.”

Next time, I would love to see some evidence from the sources to help back up your claim. This will show your readers that your claims are supported! Keep on practicing your writing and you’ll keep improving.

55 of 76

Great job on providing a brief background on the Montgomery Bus Boycott and highlighting the key players involved. You have also provided some reasons why the boycott succeeded, such as the loss of revenue for the bus company and the strong leadership of King, MIA, NAACP, WPC, and churches. To improve your essay, consider providing more specific examples to support your points. For instance, you could mention how much revenue the bus company lost during the boycott or provide specific actions taken by the leaders and organizations involved. Additionally, you could expand on the impact of the boycott beyond Montgomery and how it influenced the Civil Rights Movement as a whole.

You did a wonderful job explaining what the Montgomery Bus Boycott was historically and you have a strong claim toward the end of your first paragraph.: “The Bus Boycott has succeeded because most of the people who rode Montgomery's buses were African Americans and this made the company lose a lot of business and revenue from the boycott so this left them with no easy solutions to fix.”

Next time, I would love to see some evidence from the sources to help back up your claim. This will show your readers that your claims are supported! Keep on practicing your writing and you’ll keep improving.

Mentimeter: Menti.Com 6504 9325

56 of 76

Great job on providing a brief background on the Montgomery Bus Boycott and highlighting the key players involved. You have also provided some reasons why the boycott succeeded, such as the loss of revenue for the bus company and the strong leadership of King, MIA, NAACP, WPC, and churches. To improve your essay, consider providing more specific examples to support your points. For instance, you could mention how much revenue the bus company lost during the boycott or provide specific actions taken by the leaders and organizations involved. Additionally, you could expand on the impact of the boycott beyond Montgomery and how it influenced the Civil Rights Movement as a whole.

ChatGPT

You did a wonderful job explaining what the Montgomery Bus Boycott was historically and you have a strong claim toward the end of your first paragraph.: “The Bus Boycott has succeeded because most of the people who rode Montgomery's buses were African Americans and this made the company lose a lot of business and revenue from the boycott so this left them with no easy solutions to fix.”

Next time, I would love to see some evidence from the sources to help back up your claim. This will show your readers that your claims are supported! Keep on practicing your writing and you’ll keep improving.

Human

57 of 76

Scoring the Feedback

  • Experts blindly coded all human and AI feedback
    • Criteria based
    • Clarity
    • Accuracy
    • Developmentally appropriate
    • Supportive tone

58 of 76

All were within 1 point

59 of 76

What would a feedback assignment look like?

Let’s take a look at the assignment and the reflection after use.

Digital Learning Lab

60 of 76

Want to learn more?

If you only read a few things on generative AI in education, start with these:

Digital Learning Lab

61 of 76

Need curricular resources?

Foundational Resources

  • Do you need materials to support students' understanding of generative AI and how it works? This instructor resource has suggested reading and a video you could assign to students or you can use this student handout (and related quiz questions). Want to learn even more? This detailed explanation by Stephen Wolfram is a great technical resource. This statement by the MLA-CCCC Joint Task Force on Writing and AI provides additional context for writing instructors. This UNESCO guide is a useful general reference.
  • Would you like to provide students with materials on generative AI’s inherent limitations and biases to make them better users of AI in your class and in the world? Here is an instructor guide with suggested texts, videos, and deeper dives. This student hand out covers limitations and biases (and here are some quiz questions related to the readings and videos). For more information on hallucinations, you can reference this article and video.

Curricular Resources

Digital Learning Lab

62 of 76

Let’s try it

Topic

Act as a kind, helpful tutor with an understanding of the applicable subject matter area and good research and writing skills, who is helping a student begin their research; provide useful, specific advice. I’m writing a research paper for [course topic] and I need help coming up with a topic. I’m interested in topics related to [subject]. [Consider stating additional suggestions, requirements of the assignment, things you are interested in and areas to not include.] Please give me a list of 10 topic ideas.

  1. Want some keywords for your library research? You can use this prompt:

My research question is going to be this: [input topic or question]. Please list some keywords I can use when searching library databases.

and even:

Please construct a few Boolean search strings I can use when researching this topic in library databases.

and:

Now I'd like you to recommend 2 or 3 library databases that would be good to search for this topic.

Digital Learning Lab

63 of 76

What did you notice?

Mentimeter: Menti.Com 6504 9325

64 of 76

What would this assignment look like?

Digital Learning Lab

65 of 76

Mark Warschauer

Tamara Tate

Clare Baek

Sharin Jacob

Daniel Ritchie

Jacob Steiss

Beth Harnick-Shapiro

Michael Dennin

Waverly Tseng

Sabrina Look

66 of 76

Questions?

67 of 76

Thank you!

Tamara Tate

This material is based upon work supported by the National Science Foundation under Grant No. 23152984.

© 2023 The Regents of the University of California

tptate.com

68 of 76

69 of 76

Instructional Framework

Digital Learning Lab

70 of 76

GenAI for professionals

Noy & Zhang, 2023

  • Substantial increases in average productivity
    • Less time taken
    • Higher output quality
    • Through substituting worker effort
  • Decreased inequality between workers (boosting lower-ability)
  • Workflow restructured to focus on idea-generation, editing
  • After 2 weeks, increased job satisfaction + self-efficacy

Dell’Acqua et al., 2023

  • Substantial increases in average productivity
    • Less time taken
    • Higher output quality
    • More tasks completed
    • Generally enhanced for AI+overview
  • For tasks more difficult for AI, decreased correctness of output, but still increased quality
  • Workflow: Separate (“Centaurs”) vs. Integrated (“Cyborgs”)

Digital Learning Lab

71 of 76

Patterns: GenAI for professionals

  • Can improve productivity + quality in professional settings
    • Decreasing time taken + increasing amount completed
    • Higher output quality with tasks within frontier
  • Can help decrease inequality between workers
    • Boosting lower-performers
  • Can effect changes to workflow
    • Focus on idea generation, editing (human agency + decision)
    • Either division of labor or integration of tool into the process
    • Possible over-reliance on GenAI output (copying, correctness)

Digital Learning Lab

72 of 76

What types of things is generative AI good at?

  • Role playing, Personas
  • Making connections between related things
  • Smoothing out language to match “average” voice
  • Changing genres, voices, emulating some well known voices
  • Note: not good at word count (e.g. 100 word paragraph) because it runs on TOKENS not WORDS
  • Note: Not good at citations still (for now) even with internet access (e.g., Bard)

Digital Learning Lab

73 of 76

Interdisciplinary uses for Instructors

  • Idea generation, refinement
  • Generate metaphors, culturally diverse examples of concepts
  • Drafts of lesson plans, quiz questions, writing prompts
  • Creation of models: e.g., an essay that uses mostly passive voice
  • Draft emails, administrative reports
  • Feedback on writing

Consider whether your use aligns with the uses allowed for your students. Why or why not? Are you transparent about this?

Digital Learning Lab

74 of 76

Interdisciplinary uses for students

  • Studying: create self-testing, plan for studying, practice questions
  • Feedback: use like peer feedback for suggestions
  • Generate metaphors, examples to explain difficult concepts
  • Generative AI as a thinking partner (example of what it can do with prompting, simulated middle school student)
  • Topic generation, refinement

From Tim Brodsky:

Tutor for AP Macro

Digital Learning Lab

75 of 76

Research in process

Feedback

Digital Learning Lab

76 of 76