Innovating with Integrity: Empowering Teaching with Generative AI Adam Maksl | Amaksl@iu.edu

tinyurl.com/GenAITeachingLearning

Our goals today

  • Define generative AI, how it works, and what its implications are for higher ed and professions.
  • Discuss foundational GenAI ethical issues, including academic integrity/authenticity, intellectual property, bias mitigation, and privacy.
  • Develop skills in effective 'prompt engineering' and learn what tools are available at IU (as well as current suggested best practices for GenAI use).
  • Explore practical examples and techniques for incorporating GenAI into teaching and learning, focusing on innovation, efficiency, and ethical considerations.
  • Consider how to plan to help students engage with GenAI ethically and effectively, ensuring they understand both the opportunities and the responsibilities that come with using these tools in their academic and professional lives.

Introduction to Generative AI

What is Generative AI? It’s part of a subset of artificial intelligence that can create new data – like text and images – based on patterns it has learned from the large sets of existing data on which it is trained. Chatbots like OpenAI's ChatGPT, Microsoft Copilot, Google Gemini, and Anthropic Claude are large-language models that essentially predict the next word or set of words in response to a prompt. They work by analyzing vast amounts of text data to understand context, grammar, and semantics, allowing them to generate coherent and contextually appropriate responses. These models learn from a wide range of sources, including books, articles, and websites, to develop an extensive understanding of language patterns. In essence, they function as 'autocomplete on steroids,' capable of generating human-like text that is useful in various applications, from answering questions and drafting emails to creating educational content and simulating conversations.

Tools have also been developed to generate other types of data, such as Adobe Firefly or OpenAI’s DALL-E 2, which can produce images based on short text prompts. Similar technologies are being applied to other media, like sound and video. These models work by interpreting the provided text and using advanced neural networks to create detailed and realistic outputs. Several tools not only produce multimedia content, but they also analyze multimedia content. For instance, the multimodal capabilities of many generative AI tools can analyze images and include that information in the context of their output. This means that these tools can look at an image, understand its content, and then generate a descriptive caption or relevant text. Similarly, they can interpret visual data alongside text, allowing for richer, more integrated responses. This could be applied in various ways, from creating detailed image descriptions for accessibility to analyzing radiographic imagery in clinical medical settings.

Google, Microsoft, and other companies are beginning to integrate generative AI into products, including search and productivity tools. These integrations enhance user experiences by providing more intelligent and context-aware features. For example, AI can generate summaries of long documents, draft emails, and create visual content directly within these platforms. In search engines, generative AI can deliver more accurate and nuanced results by understanding user intent better. These advancements not only improve efficiency but also enable users to leverage AI for more creative and complex tasks, transforming how we interact with technology in everyday work and life.

What are the implications for teaching and learning, and for higher ed (or even more generally)? There are probably more questions than answers, not just about what this technology is and how it works, but also about appropriate use and its possible effects on creative expression and information ecosystems (including the problem of creating and spreading misinformation). Some questions include:

  • Whether AI-generated output can be passed off as human generated, and how we can address this is our learning outcomes, activities, and assessments.
  • Whether the content produced can be trusted, and the likelihood for it to produce biased output.
  • Whether these tools can aid in teaching and learning, including in providing more personalized learning experiences and in expanding the ability for faculty to emphasize relationships, community, and application in what they teach.
  • How AI tools may impact our own disciplines, and what we can do to help prepare students for a world where skills in using AI will be of tremendous value.

What are the implications of GenAI in professions? In the last year, many have written about the impact of generative AI on various industries, with McKinsey & Company estimating that it will have the ability to automate nearly 70 percent of what employees currently spend time doing and adding between $2.6 trillion and $4.4 trillion annual to the economy.

Ethics & AI

  • Intellectual Property: Navigating the challenges of copyright in AI-generated content.
  • Privacy: Ensuring data protection.
  • Bias and Fairness: Mitigating bias in AI systems and promoting diversity.
  • Authenticity: Addressing the potential for misinformation and its impacts.
Everything you need to know about prompt engineering you know from being an educator!

Four Principles of Prompt Engineering

  • Specificity
  • Context
  • Examples
  • Iterative Interaction

Specificity

  • The crafting of detailed and precise prompts.
  • Identify the task objectives, using clear language. Avoid generalities.
  • Less effective: "Create a learning activity for teaching how to make requests in English."
  • More effective: "Create an active learning activity for an asynchronous online university class in the United States. The activity will teach intermediate-level English learners how to make conventionally indirect requests in an academic setting in the Midwest region of the United States. The learners are native Spanish speakers."

Context

  • Providing background information or situating the prompt within a specific scenario or framework. ​
  • Can involve assigning the AI to a specific role, asking it to address the needs of a specific audience, or asking it to adopt certain viewpoints in responses.
  • Example: "Create code for a page in my Canvas course of a study guide for an introductory organic chemistry course aimed at college sophomores. Cover key topics such as compound structure, functional groups, nomenclature, and reaction mechanisms. Organize the guide logically, following a structured approach that introduces concepts progressively and builds upon foundational knowledge. Ensure clarity in explanations by breaking down complex topics into manageable steps, utilizing analogies and real-life examples where applicable. Additionally, include practice problems with solutions to reinforce learning and provide opportunities for self-assessment. This study guide should be comprehensive, and include practice questions, answers, and explanations, not just topics to study."
  • Example: "Assume the role of a negotiation teacher and create an interactive negotiation game. Start with a scenario where I'm trying to negotiate the price of a used car with a seller. Provide instructions for me to follow, such as how to make an opening offer and respond to counter-offers. After I provide my approach to the negotiation in this scenario, give me feedback on my strategy, highlighting what I did well and areas for improvement. Additionally, include tips on effective negotiation techniques that could be applied in real-life situations."

Examples

  • Define output format (e.g., narrative, list, detailed report, bullet points).
  • Define the style/provide example (give the model a reference point for what you want).
  • Models can provide more than text (e.g., image generation, video, etc.).
  • Example: "Develop a comprehensive set of measurable learning outcomes for an undergraduate course titled 'Communications Law: Focus on the First Amendment'. These outcomes should encapsulate key competencies students are expected to achieve, directly related to understanding, analyzing, and applying First Amendment principles within the realm of communications law. Ensure each outcome adheres to the SMART criteria (Specific, Measurable, Achievable, Relevant, Time-bound) and aligns with Bloom's Taxonomy for cognitive skills development."
  • Example: "Create a compelling narrative that explains the value of an undergraduate course titled 'Communications Law: Focus on the First Amendment' to students entering the job market. Use engaging language to highlight how the course equips students with marketable skills sought after by employers across various industries. This course is not just an exploration of legal principles; it's a journey into the heart of how communication operates within the legal frameworks that govern our society. Your task is to convey, in a compelling narrative format, why this course is invaluable to students aiming to enhance their marketability in various professional fields. Emphasize the broad spectrum of skills they will acquire and how these skills are directly transferrable to their future careers."​

Iterative Interactive

  • The continuous, back-and-forth communication with the AI.
  • Includes adjusting prompts, providing feedback, and asking questions based on the AI's responses.
  • Break down complex tasks and adapt.
  • Don't be afraid to give constructive feedback. Or start over.

Exploring applied uses of GenAI in teaching and learning

It's all in the prompt. I'll share a few strategies for using GenAI in higher education. Several of these I've adapted from Ethan Mollick, a UPenn Wharton School professor who has written quite a bit about generative AI. In particular, you may want to check out his posts "Using AI to make teaching easier & more impactful" and "How to use AI to do practical stuff: A new guide." His recent book, "Co-intelligence: Living and working with AI" is excellent.

Create formative assessments

I adapted one of Mollick's prompts to use in one of my classes to create low-stakes formative assessments to add to course videos via PlayPosit. This is what I asked ChatGPT to do:

You are a quiz creator of highly diagnostic quizzes. You will make good low-stakes tests and diagnostics. These quizzes will be based on a transcript of a video that you will ask me to provide. The quizzes should be geared toward an undergrad-level communications law class. Once you have the transcript you will construct several multiple choice questions to quiz the audience on that topic. The questions should be highly relevant and go beyond just facts. Multiple choice questions should include plausible, competitive alternate responses and should not include an "all of the above option." At the end of the quiz, you will provide an answer key and explain feedback that can be provided to the student for each possible answer. Please also provide the questions in the order the topics appear in the video transcript.

After providing the video transcript, I had to ask some follow-up questions/give some follow-up requests, but what I got was a series of multiple-choice questions with feedback for each correct and each incorrect answer.

Create hypotheticals, examples, and question variations

In many of my classes, I use hypothetical examples to ask students to apply course concepts. Those examples can be time-consuming and difficult to develop. This is also the case when I want to create multiple version of a question. I've experimented with ChatGPT as an assistant to help create these.

Mollick has a similar use. Here's his prompt:

I would like you to act as an example generator for students. When confronted with new and complex concepts, adding many and varied examples helps students better understand those concepts. I would like you to ask what concept I would like examples of, and what level of students I am teaching. You will provide me with four different and accurate examples of the concept in action.

You can also provide existing test questions and ask it to create variations. And keep tweaking until it's to your liking.

Creating rubrics and developing comment banks for more efficient grading

Generative AI can be used to create assignments, or to adapt current assignments. This Microsoft "Prompts for EDU" resource page provides a sample prompt to adapt assignments to make them more active:

You could also use it to create rubrics. Like other prompts, the more specific the better. For instance, providing details on the assignment or what you're looking for in a rubric is essential. Good rubrics have criterion, performance bands, weights/points, and descriptors, so asking for some of those elements are useful in an effective prompt. Here's an example (created by an instructor designer colleague Emilie Schiess and presented at the 2023 IU Online Conference):

Act as if you are a machine that produces rubrics to be used for assessing as assignment. I will provide you with the details of the assignment, including assessment criteria. I will provide the number of points overall for the assignment as well as the points allocated for each individual sub-part of the assignment. Currently, I give points in a free-form way, but I would like to have specific levels of ratings (somewhere between 3-5 levels). The number of levels for each criterion in an assignment doesn't have to be consistent across the assignment. I need the range of points each level should award as well as descriptive criteria for each level. Format this in a table, with criteria on the along the vertical axis and performance bands along the top on the horizontal axis. Make sure each cell in the table includes descriptive criteria.

Finally, I've started using generative AI tools in the creation of comment banks to aid in grading, using the the following prompt along with the language of my assignment instructions and rubric:

I need a variety of comments to use when providing comments to students for this assignment. Please give me a bank of 5-10 short 1-2 sentence comments that offer a range of feedback. Please make sure the comments relate to the rubric and effective analysis according to the guiding questions. Here's the assignment and rubric that these comments are based on... [add language from assignment + rubric]

And recently, I've started using it to help create short, targeted comments unique for each observation I made while grading. I've added this additional prompt to the response from the previous prompt:

I will now give you some general observations and you will create short 1-3 sentence comments, in my conversational style, that are based on the rubric that I can add to the comments for the student. Based on those observations I want 2-3 sentences that are connected, not independent of another, in one paragraph without quotes marks around it. Make sure all comments start off with something positive, includes constructive suggestions, and encourages students to improve.

Note that the AI is NOT grading the assignment, and I've provided no student data to the AI system. I am only providing general observations or comments that I might make, and asking the AI to craft those comments more fully and in alignment with the assignment instructions and rubrics.

Write, revise, and edit

Generative AI tools can be used to write, or help you write. I've heard people describe it as a draft creator, or as a tool to break writers' block.

For example, you can ask these tools to revise writing to make it more concise. Or to adapt tone for another kind of audience.

Tools like ChatGPT can also be used to come up with ideas, a lot of them, quickly. Mostly, some OK, but maybe there is a diamond in there. Sometimes seeing a large set of ideas can also help with sparking our own creativity.

It'll take a little work, but these can be incredibly useful in writing or other creative processes.

Create simulations and interactions

You might also think about using generative AI in asking students to interact with the AI in an interactive exercise. Mollick has written about this and provided a sample prompt:

I want to do deliberate practice about how to conduct negotiations. You will be my negotiation teacher. You will simulate a detailed scenario in which I have to engage in a negotiation. You will fill the role of one party, I will fill the role of the other. You will ask for my response to in each step of the scenario and wait until you receive it. After getting my response, you will give me details of what the other party does and says. You will grade my response and give me detailed feedback about what to do better using the science of negotiation. You will give me a harder scenario if I do well, and an easier one if I fail.

Engaging students in conversations about GenAI

Discussing GenAI in the syllabus

When introducing generative AI in your syllabus, it’s important to set the tone for how you expect students to engage with these tools. Among the things you want to address are:

  • When GenAI can be used without disclosure? For example, you might say using GenAI for language/wording/editing of one's own work is OK without disclosure (word processors like Microsoft Word and Google Docs have been doing variations of this for years). Or you might say that it's OK to use without disclosure with GenAI is used as a personal assistant to find information, which is an extension of how we use Internet and database searches.
  • When can generative AI be used with disclosure? Give examples, such as using GenAI to create an image or draft a specific passage that gets included in one's work (but is not the basis for the entire work). Or when it is used to create new ideas (i.e., ideas not of one's own generation). Make sure to offer guidance on how GenAI use should be disclosed. For example, you might ask for a statement explaining how students used the AI, potentially including a written explanation and/or a transcript of their interaction with the AI tool.
  • When is generative AI not allowed? Be clear when GenAI is not allowed, such as when it creates the new ideas of students' work and the actual text/content, with minimal changes. Be clear how you will see this work, such as considering it a form of academic misconduct/dishonestly, as a mix of plagiarism and collusion.

In your syllabus, consider briefly explaining how you will use generative AI tools in your teaching and work. You might mention using AI for tasks like editing your own work, gathering information, or brainstorming ideas. Additionally, whether and how you might use GenAI in providing feedback to students..

Addressing GenAI in individual assignments

When designing assignments, you may want to give students clear guidance on how they can use GenAI tools in ways that align with the learning objectives. Here are some sample options you could use:

  • No AI permitted. Clearly state if AI tools are not allowed for a particular assignment, emphasizing that students must complete all work independently to fully develop their skills.
  • Generating ideas. You might allow students to use AI for brainstorming activities, such as generating business ideas, thesis statements, etc.
  • Creating outlines. You might permit the use of AI to help outline papers, assignments, or arguments, providing a structured starting point for their work.
  • First draft. You might allow students to use AI to create initial drafts of their work. Emphasize that this is just a starting point and must be revised and personalized.
  • Analyzing data. You might encourage the use of AI tools to help analyze data sets and generate conclusions. Remind students to critically assess AI-generated analysis to ensure accuracy and relevance.
  • Rewriting, editing, polishing, debugging. You might allow AI to assist in rewriting portions of work, editing reports, debugging code, or polishing final drafts. You might also allow GenAI to be used for editing, such as fixing punctuation and grammar.

A cautionary note about AI detection

Indiana University has not approved the use of any AI detection software, meaning no tools are available for faculty to submit student work for AI detection. These tools are often unreliable, producing false positives, and they disproportionately tag non-native English writers' work, leading to inequitable outcomes.

Instead of relying on detection, it’s more effective to focus on educating students about responsible AI use and designing assignments that emphasize critical thinking and originality. This approach also fosters a more inclusive and trustworthy learning environment. Additionally, scaffolding assignments, such as requiring multiple drafts of written work, may not only reduce the use of GenAI for cheating but also is generally considered a best practice in promoting learning and skill development.

A very important note about detectors: Generative AI tools themselves cannot identify whether some text was created by generative AI, even if the tool itself created it. For example, ChatGPT cannot detect whether some text was previously generated using ChatGPT, although it may incorrectly report that it can.

Resources

There is much being written about these tools, including its role in teaching and learning. I've curated a few items below.