What is generative AI?

An image of a robot in the style of Gustav Klimt generated by MidJourney

Generative AI is a type of Artificial Intelligence that creates new content, including text, images, videos, audio, and computer code. It is trained to identify patterns, relationships, and characteristics of existing data, and then mimic those patterns and relationships when creating new content. It is called “generative” because it generates new content based on these patterns. For example, the image on the right shows a new piece of artwork in the style of Gustav Klimt. Below, is an AI Generated poem in the style of Shel Silverstein generated in March 2025 using Google Gemini. As discussed in the Ethics section, although generative AI technologies are quite good at mimicking style, the ability to mimic writing and art styles of specific authors and artists has led to numerous controversies around intellectual property and copyright.

A conversation with Gemini 2.0.

Where does generative AI fit into the AI landscape?

Generative AI is one subset of Artificial Intelligence.

Artificial Intelligence (AI) are technologies that can simulate human intelligence by performing tasks that require the ability to reason, learn, and act independently. In popular media, AI is often represented as nearly indistinguishable from humans (for example the Cylons in Battlestar Galactica or the Replicants in Bladerunner ). In the present, AI systems are not quite this advanced, but they are becoming more advanced and able to tackle increasingly complex tasks.

Generative AI models are a specialized type of Artificial Intelligence built using Deep Learning techniques to create new content. Deep Learning systems are modelled on the neural networks in the human brain, which allows them to perform very complex tasks like image and speech recognition and generation.

Some generative AI models are Large Language Models (LLMs). LLMs are specifically trained for natural language processing and production tasks. They are pre-trained on large amounts of text, and from this text they learn patterns of syntax and semantics in human language. Large Language Models are used in both discriminative and generative AI systems, meaning that they can be used to both classify new input (for example, to decide if a new email message is spam or not) and to generate new content (for example, to write a new email message asking for an extension on a piece of work).

LLMs are typically general-purposed, meaning that they are trained to solve common language problems. They can be used for:

• Translation
• Text classification
• Text prediction / completion
• Text generation
• Text revision

Examples of Large Language Models include GPT-4, Claude, Gemini, and LlaMA.

Pre-trained LLMs can be further trained on a smaller task or domain specific datasets to allow them to achieve better results or perform specialized tasks. This process is called fine-tuning. For some LLMs, fine-tuning can be done by the end user.

For example, imagine that we wanted to create a TA ChatBot for a course on Information Ethics. An existing LLM would have the foundational understanding of language and perhaps some knowledge on the topic but may not have the specialized knowledge of all of the content covered in the course. This model could be fine-tuned using all of the course readings, lectures, and other course content. This would increase its ability to respond accurately to specific course questions.

Diffusion Models are another type of deep learning that can be used to learn and replicate patterns in visual data. Many image or video generating tools use Diffusion Models. Examples of image-generating diffusion models include Adobe Firefly, MidJourney, DALL-E, and Stable Diffusion.

Image generated January 2025 using Adobe Firefly in response to prompt “The future of higher education with an emphasis on emerging technologies.”

Most of the examples and activities in this resource will focus on LLMs.

What are common applications of generative AI technologies?

ChatBots are one example of a modern user interface that has made access to generative AI models much easier for the general public. ChatBots are designed to simulate human conversation by accepting natural language prompts or inputs and producing responses in natural language. They can be built on LLMs, allowing them to provide sophisticated responses to prompts. Note that not all chatbots are generative; many are rule-based, meaning they have a set of pre-defined responses to prompts and do not generate unique or original text.

Popular generative AI ChatBots include:

• Microsoft Copilot (accessible with your UWO credentials)
• ChatGPT
• Google Gemini 
• Claude 

The functionality of these tools is constantly changing; however, currently many of these tools are able to accept multiple types of input (e.g. text, images, files) and produce multiple types of output.

AI technologies are embedded in many other technological tools and processes. Examples of other places you may encounter AI on a daily basis include:

• Digital voice assistants (Siri, Alexis, Cortana) which use AI for voice recognition, to understand and process requests, to generate responses to requests and for a variety of other tasks.
• Text Editors (Grammarly, Microsoft Office) which use AI to generate grammar, style, and word choice suggestions
• Auto-captioning tools (Zoom, Otter.ai) which use AI for voice recognition and word recognition; they may also use AI training models to learn specific voices and improve their captioning.
• Search Engines (Google, Bing) which use AI to understand and process requests, to optimize search results, to analyze images for image searches, to generate responses, and for a variety of other tasks.
• Grading Tools (Gradescope) which use AI for text recognition and to group similar responses together to aid marking tasks.

What do we know about how large language models work?

Training

The model is then released for use and can be accessed by users

Generation

LLMs generate text in response to a user-provided prompt.

A diagram showing the cycle of interacting with a Large Language Model.

Prompt: Tell me a joke about higher education.

Token Breakdown: ["Tell", "me", "a", "joke", "about", "higher", 
"education", "."]

Prediction 

The LLM analyses the vectors and, based on the patterns and other information learned in training, the AI model will start to predict a response to the prompt based on the probability of response tokens appearing in sequence.

For example, in responding to our request to generate a joke, the most likely starting token may be “Why”, “What”, “Here’s” etc.

 

Probability chart Generated by DALL-E via ChatGPT 4o in March 2025 for demonstrative purposes; not an accurate representation of probabilities.

Although Large Language Models do generate output based on probability, they do this using millions or billions of parameters, so often the process of generating is so complex that creators and users don’t really know how they work. This opaqueness is why AI systems are often referred to as a black box, making generative AI systems vulnerability to unseen biases, vulnerabilities, and other problems.

For more a more detailed introduction to generative AI, see this video from Google:

Introduction to Generative AI

For a more in-depth look at how LLMs function, see this article from the Financial Times:

Generative AI exists because of the transformer

What are the Limitations of Large Language Models?

Generative AI is evolving quickly but still has certain limitations. Large Language Models (LLMs) are constrained by the data upon which they are trained and the methods through which they are trained. It’s important to be aware of the limitations of the tools that you’re using, especially if currency or accuracy is important for the tasks that you’re using generative AI to complete.

• We do not fully understand how LLMs work, which presents issues for safety, reliability and accuracy.
• LLMs are susceptible to hallucinations or the creation of nonsensical words, phrases, or ideas. This can also result in the generation of non-existent references .
• Many LLMs are pre-trained and have knowledge cut-off dates, meaning that data may be out of date or inaccurate. However, increasingly generative AI tools are able to access and process information in real time. This is called Retrieval Augmented Generation (RAG).
• There is a trade-off between processing speed and accuracy with LLMs. Many basic models do not fact-check, meaning that the information that they share is not guaranteed to be accurate or logical. These models produce much faster results at the risk of lower accuracy. Reasoning models have increased accuracy because they break tasks down into micro-steps, apply logic, and evaluate possible results. However, they have longer processing times and require significantly more resources. They are also not immune to making mistakes.
• Standard LLMs produce output based on averages or probabilities of patterns, so they are susceptible to reproducing biases found in their data sets, including but not limited to biases based on human biases that may be embedded in historical records, cultures, patterns of research, societal norms, and any other elements reflected in the text data used for their training. This will be discussed more in the Ethics section.

Financial Times Visual, Data & Podcasts. (2023, September 11). Generative AI exists because of the transformer [Interactive graphic]. Financial Times. https://ig.ft.com/generative-ai/

Schmid, P. (2025, June 30). The new skill in AI is not prompting, it’s context engineering. Retrieved August 21, 2025, from https://www.philschmid.de/context-engineering

Stryker, C. (n.d.). What is agentic AI? IBM. Retrieved August 21, 2025, from https://www.ibm.com/think/topics/agentic-ai

University of Waterloo Library. (2025, May 22). ChatGPT and generative artificial intelligence (AI): Incorrect bibliographic references [Web page]. University of Waterloo. https://subjectguides.uwaterloo.ca/chatgpt_generative_ai/incorrectbibreferences

Wikipedia contributors. (n.d.). Cylons. In Wikipedia… the free encyclopedia. Retrieved August 17, 2025, from https://en.wikipedia.org/wiki/Cylons

Wikipedia contributors. (n.d.). Replicant. In Wikipedia… the free encyclopedia. Retrieved August 17, 2025, from https://en.wikipedia.org/wiki/Replicant

YouTube. (2023, May 8). Introduction to generative AI [Video]. YouTube. https://www.youtube.com/watch?v=G2fqAlgmoPo

Privacy, Intellectual Property & Copyright

A key ethical issue around the training and use of generative AI tools is understanding how data is being used. Privacy, intellectual property and copyright all need to be considered at all stages of generative AI development and use, including text and data mining (TDM), prompting, and the output of generative AI tools.

Text and Data Mining (TDM)

As we learned in the first section, generative AI is trained on massive datasets of texts and images. These datasets may include copyrighted or otherwise protected data or works.

In fact, multiple lawsuits have been filed against generative AI companies because of suspected unethical use of copyrighted content for training.

• The New York Times has sued OpenAI and Microsoft for using New York Times articles to train their Large Language Models 
• A collective of artists has filed a class-action lawsuit against Stability AI, DeviantArt and Midjourney for using the work of millions of artists as training data 
• Another lawsuit was filed against GitHub Copilot on behalf of open-source programmers who claim that using their code to train the Copilot AI was in violation of the open licenses 

Ongoing discussions in Canada and abroad have focused on whether AI training and TDM require permission from copyright holders or should be an exception to copyright law (Government of Canada, 2025 ; WIPO, 2024 ). These discussions are ongoing as of early 2025.

Prompting

Another question of Intellectual Property is the information and data that is entered into generative AI tools by users. Different tools have different policies around the collection and use of personal data, including the data required to set up an account and any information included in prompts. Users may inadvertently provide personal or sensitive data in their prompts or queries to AI systems, many of which are saved, stored, and used by companies for training or other purposes.

Data is typically grouped into three risk levels:

• Low risk: data is non-sensitive and has minimal impact on privacy. Examples include email addresses and usernames.
• Medium high risk: includes personally identifiable information and may present some risk for privacy. Examples include names, addresses, phone number, and birthdays. Note that as more medium risk data is shared simultaneously, it can increase the risk levels (for example, providing a name, birthday, and address is riskier than just providing a name).
• High risk data: includes information that has significant implications for privacy risks. Examples include financial information and identification numbers (such as social insurance numbers).

The Freedom of Information and Protection of Privacy Act (FIPPA)  states that you can only use personal information for the purpose it was collection and prohibits the use of an individual’s personal information for other purposes without their consent. The answers in completed assignments, exercises, exams, etc., are considered to be the personal information of the student. (See Western’s statement on FIPPA ).

Before adopting new technologies, it’s important for instructors to review and fully understanding the privacy, security and data management policies and practices of the tool and encourage students to do the same. It’s also important to understand the settings and features of each tool. For example, some tools may have settings that allow you to exclude your prompts from training data or create a temporary chat.

ChatGPT is one example of a tool that offers a temporary chat and different settings for chat history and data privacy.

See Western’s Cyber Security Awareness Training for more information on safe digital practices regarding privacy.

Intellectual Property is also an important consideration when using generative AI. Any information included in your prompts is subject to existing copyright laws. It’s important to ensure that you have permission to use any content (text, images, data, etc.) that you are including in your prompts.

When considering using generative AI tools to support teaching and learning, it’s especially important to keep in mind that any work created by a student as part of their academic studies is their Intellectual Property and subject to the same copyright laws. (see Policy 7.16 – Intellectual Property ).

Consider the cases below to examine how privacy, copyright and intellectual property might impact the use of generative AI tools.

Works Generated by AI

Another question of Intellectual Property & Copyright arises with the output generated by generative AI models. The question that has been raised is whether or not works produced by non-human tools can be copyrighted. Currently, the discussions in Canada suggest that in order to be copyrighted, a work must have sufficient human input  (Government of Canada, 2025 ). Works produced with generative AI may also raise concerns about copyright infringement if they are too similar to existing works.

Access

Any technology presents issues of equitable access related to cost. Even tools that are free often offer a more feature-rich or advanced version at a cost. When allowing or encouraging students to use AI tools, it’s important to be aware that many students may not be able to afford access to certain tools or to premium licenses. When designing assignments or activities, ensure that this won’t unfairly advantage students who are able to access the paid versions of these tools.

Paid accounts can vary in cost, with very high costs for the most sophisticated license options.

Accessibility

Another important consideration when adopting any new tool for teaching is the accessibility of the tool. This means ensuring that every learner is able to access and use the tool.

Basic AODA standards require web-based tools to allow for keyboard navigation and for assistive technology compatibility. A review of the accessibility of AI interfaces conducted by Langara College suggests that many AI tools are not in compliance with AODA requirements or present other accessibility barriers. When evaluating a new technology, like generative AI, review its documentation for information about built-in accessibility features to ensure that they comply with these basic requirements.

It is important to recognize that even technologies compliant with AODA standards may still present barriers to access for some learners. You may need to provide an alternative tool or an alternate way of completing an activity to ensure all students are able to participate.

For more information, see Western’s Policy on Accessibility and the Accessibility Western site .

Environmental Impact of AI

Image generated using MidJourney in 2023. Prompt unrecorded.

The rapid growth of AI technologies (and cloud-based technologies in general) has sparked a lot of concern around the environmental impact of technological advancement.

There are environmental considerations at each stage of the AI development process:

Hardware: the physical resources required for generative AI hardware and infrastructure involves extensive mining and extraction of minerals, which can lead to deforestation and increased soil and water pollution. The production of hardware, like Graphic Processing Units, can also consumed large amounts of energy and water. (Hosseini et. al., 2025 ). The rapid growth of generative AI technologies will also contribute to the global increase in ewaste, which when not properly disposed of can also contribute to air, water, and soil pollution. A report from 2022 indicates that only 22% of ewaste is properly recycled (Crownhart, 2024).

Training:  Training generative AI models requires significant amounts of energy. For example, it has been estimated that creating GPT-3 resulted in carbon dioxide emissions equivalent to the amount produced by 123 gasoline powered vehicles driven for a year (Saenko, 2023 ).

Usage: using generative AI also has a substantial water footprint and significant carbon emissions. It is estimated that a ChatGPT dialogue with 20-50 prompts uses approximately 500ml of water (McLean, 2023 ). Estimates suggest that by 2027, use of AI technologies globally will account for water withdrawal equivalent to 4-6 times that of Denmark or half that of the United Kingdom (Li, Islam & Ren, 2023). However, in contrast, an analysis conducted in 2024 suggests that the carbon emissions of content creation (text and images) may actually be lower for generative AI produced content than human-produced content (Tomlinson, Black, Patterson, & Torrance, 2024). 

Data centres, though not limited to generative AI technologies, are becoming one of the largest consumers of energy, currently accounting for 3% of global energy consumption (Cohen, 2024 ). They also require large amount of water for cooling. There are more environmentally sustainable approaches to cooling down data centres, but these are substantially more expensive, which could be seen as another example of values-friction (sustainability VS profitability), as discussed in section 3.2, (Ammachchi, 2025 ).

Often, the largest environmental impact of technological development occurs in already disadvantaged communities, perpetuating existing inequities. For example, a study in the US shows that high-pollutant data centres were more likely to be built in racialized communities (Booker, 2025 ).

Many of these environmental concerns are not new, but generative AI has brought a renewed focus on the environmental impact of digital technologies and rapid technological advancement. The environmental impact is a prime example of the complexity of the generative AI conversation as it highlights tensions in values and priorities, social inequities, and the affective nature of these conversations.

See AI’s impact on energy and water usage for a review of recent research on the environment impact of generative AI technologies.

Bias-In, Bias-Out

Image generated using MidJourney in 2023 in response to prompt: “The incoming class of Western University students for 2023”

Many AI models are trained on data where social biases are present. These biases are then encoded into the patterns, relationships, rules, and decision-making processes of the AI and have a direct impact on the output.

Biased data can be easy to spot, such as in this AI generated image which shows a predominantly white class of 2023 at Western, but it can also be more invisible. AI-generated text will reflect dominant ideologies, discourses, language, values, and knowledge structures of the datasets they were trained on. For example, Large Language Models may be more likely to reproduce certain dominant forms of English, underrepresenting regional, cultural, racial, or class differences (D’Agostino, 2023 ).

The ethical issue is twofold: first, the information generated by generative AI is more likely to reflect dominant social identities, meaning that students who use AI will not be exposed to certain worldviews or perspectives, and some students may not feel that their experiences and identities are reflected in the output. Second, the use of generative AI to produce knowledge will continue to reinforce the dominance of these ideologies, values, and knowledge structures, contributing to further inequities in representation.

As an instructor, it’s important to be aware of this limitation of AI tools. If you ask your students to use these tools, it’s also important to teach them critical AI literacies to similarly be able to identify and reflect on these issues of representation, bias and equity.

Some generative AI companies have taken steps to correct for biases in the training data by establishing content policies or other guardrails to prevent generating biased or discriminatory output. However, these guardrails are inconsistent and can be subject to the ethical standards of each generative AI company.

Conversation with ChatGPT in January 2025.

Bias Case Study

Misinformation, Disinformation, & Mal-information (MDM)

Because of the ability to generate plausible information at scale, generative AI technologies have exacerbated the potential harms of misinformation, disinformation and mal-information in a time of information abundance. The generation or dissemination of fake, inaccurate, or misleading information through generative AI could be either unintentional or deliberate.

Misinformation refers to inaccurate or false information that is shared without intending to create harm. This could occur as a result of generative AI users not verifying generative AI outputs before sharing them.

Mal-information refers to information that may be rooted in truth or fact, but removed from context or distorted in ways that can mislead. When using generative AI, this might be the result of inaccurate outputs or hallucinations. Generative AI could also be used by malicious individuals to distort information in a way that is plausible.

Disinformation refers to inaccurate or false information that is shared with malicious intend, to mislead recipients or manipulate decision-making or perspectives. Generative AI could be used to generate fake news stories, fake datasets, or otherwise employed in attempts to deceive at large scales.

(Canadian Centre for Cyber Security, 2024; Jaidka et. al., 2025)

One of example of this is the use of text-to-image and text-to-video generative AI tools to produce visual media for the purposes of (malicious or not) deception. A deepfake is the product of generative AI that creates a believable but fake video, audio, or image. They often feature real people saying or doing something that they didn’t really say or do. Deepfakes do have potential benefits for the arts, for social advocacy, for education and for other purposes, but they do present ethical issues because often permission has not been received to use the person’s likeness and because it has the potential to spread misinformation or to mislead people.

Ammachchi, N. (2025, August 19). Water-guzzling data centers spark outrage across Latin America. Nearshore Americas. https://nearshoreamericas.com/water-guzzling-data-centers-spark-outrage-across-latin-america/

Booker, Mario DeSean. (2025). Digital redlining: AI Infrastructure and Environmental Racism in Contemporary America. World Journal of Advanced Research and Reviews. 10.30574/wjarr.2025.27.1.2602.

Canadian Centre for Cyber Security. (2024, May). How to identify misinformation, disinformation and malinformation (ITSAP.00.300). Communications Security Establishment. Retrieved from https://www.cyber.gc.ca/en/guidance/how-identify-misinformation-disinformation-and-malinformation-itsap00300

Chen, M. (2023, January 24). Artists and illustrators are suing three A.I. art generators for scraping and ‘collaging’ their work without consent. Artnet News. Retrieved August 17, 2025, from https://news.artnet.com/art-world/class-action-lawsuit-ai-generators-deviantart-midjourney-stable-diffusion-2246770

Cohen, A. (2024, May 23). AI is pushing the world toward an energy crisis. Forbes. https://www.forbes.com/sites/arielcohen/2024/05/23/ai-is-pushing-the-world-towards-an-energy-crisis/

Crownhart, C. (2024, October 28). AI will add to the e-waste problem. MIT Technology Review. https://www.technologyreview.com/2024/10/28/1106316/ai-e-waste/

D’Agostino, S. (2023, June 5). How AI tools both help and hinder equity. Inside Higher Ed. https://www.insidehighered.com/news/tech-innovation/artificial-intelligence/2023/06/05/how-ai-tools-both-help-and-hinder-equity

Earth.Org. (2023, April 28). The environmental impact of ChatGPT [Web article]. Earth.Org. https://earth.org/environmental-impact-chatgpt/

Hosseini, M., Gao, P., & Vivas-Valencia, C. (2024, December 15). A social-environmental impact perspective of generative artificial intelligence. Environmental Science and Ecotechnology, 23, Article 100520. https://doi.org/10.1016/j.ese.2024.100520

Ippolito, J. (n.d.). 9 takeaways about AI energy and water usage [Web page]. Learning With AI. Version 1.9. Retrieved August 17, 2025, from https://ai-impact-risk.com/ai_energy_water_impact.html

Jaidka, K., Chen, T., Chesterman, S., Hsu, W., Kan, M.-Y., Kankanhalli, M., Lee, M. L., Seres, G., Sim, T., Taeihagh, A., Tung, A., Xiao, X., & Yue, A. (2025). Misinformation, Disinformation, and Generative AI: Implications for Perception and Policy. Digital Government (New York, N.Y. Online), 6(1), Article 11. https://doi.org/10.1145/3689372

Joseph Saveri Law Firm, LLP. (n.d.). GitHub Copilot intellectual property litigation [Web page]. Joseph Saveri Law Firm, LLP. Retrieved August 17, 2025, from https://www.saverilawfirm.com/our-cases/github-copilot-intellectual-property-litigation

Langara College Educational Technology. (n.d.). Accessibility of AI interfaces [Web page]. Langara College. Retrieved August 17, 2025, from https://students.langara.ca/about-langara/academics/edtech/AI-Accessibility.html

Li, P., Yang, J., Islam, M. A., & Ren, S. (2023). Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models. arXiv. https://doi.org/10.48550/arXiv.2304.03271

Ontario. (n.d.). Freedom of Information and Protection of Privacy Act (R.S.O. 1990, c. F.31) [Statute]. e-Laws, Government of Ontario. Retrieved August 17, 2025, from https://www.ontario.ca/laws/statute/90f31

Pope, A. (2024, April 10). NYT v. OpenAI: The Times’s about-face [Blog post]. Harvard Law Review. https://harvardlawreview.org/blog/2024/04/nyt-v-openai-the-timess-about-face/

Saenko, K. (2023, May 25). A computer scientist breaks down generative AI’s hefty carbon footprint. Scientific American. Reprinted from The Conversation US. https://www.scientificamerican.com/article/a-computer-scientist-breaks-down-generative-ais-hefty-carbon-footprint/

Tomlinson, B., Black, R.W., Patterson, D.J. et al. The carbon emissions of writing and illustrating are lower for AI than for humans. Sci Rep 14, 3732 (2024). https://doi.org/10.1038/s41598-024-54271-x

Western Technology Services. (n.d.). Learn It. [Web page]. Western University. Retrieved August 17, 2025, from https://cybersmart.uwo.ca/for_western_community/learn_it/index.html

Western University. (2009, December 1). Policy 1.47 – Accessibility at Western [PDF]. Manual of Administrative Policies and Procedures. University Secretariat. Retrieved August 17, 2025, from https://www.uwo.ca/univsec/pdf/policies_procedures/section1/mapp147.pdf

Western University. (2018, April 26). Policy 7.16 – Intellectual Property [PDF]. Manual of Administrative Policies and Procedures. University Secretariat. Retrieved August 17, 2025, from https://www.uwo.ca/univsec/pdf/policies_procedures/section7/mapp716.pdf

Western University, Office of the Vice-President (Operations & Finance), Legal Counsel. (n.d.). FIPPA – Some basics for faculty and staff [Web page]. Western University. Retrieved August 17, 2025, from https://www.uwo.ca/vpfinance/legalcounsel/privacy/fippa.html#protection

World Intellectual Property Organization. (2024). Generative AI: Navigating intellectual property [Factsheet]. World Intellectual Property Organization. Retrieved August 17, 2025, from https://www.wipo.int/documents/d/frontier-technologies/docs-en-pdf-generative-ai-factsheet.pdf

Our values are our fundamental beliefs about what is important. Our values motivate action and impact who we are and how we exist in the world. All individuals  hold multiple values simultaneously, but attribute different levels of importance to each value (Schwartz, 2012). The values that an educator holds will impact their pedagogical practices and curricular decisions (e.g. what they teach, how they teach, how they interact with learners, etc.). In particular, your values will impact how you choose to use/not use generative AI tools in your teaching and course design and how you talk about generative AI with your learners. Similarly, the values that a learner holds will impact their decisions around if/how they use generative AI as part of their learning. Recognizing and naming your values may help you better navigate your beliefs and emotional response to generative AI.

Values Friction

Although values do impact decisions and actions, the link between a value and an action or decision is not always direct or obvious. There may be many factors that impact decision making, including different understanding of the value proposition of a decision. For example, one instructor who values inclusivity may encourage the use of generative AI technologies to help support student access needs, reduce language barriers, or provide additional scaffolds; another instructor who values inclusivity may discourage or ban the use of generative AI because not all student may have the resources to pay for a subscription which can result in inequal access for students.

Values friction (conflicting values) may also impact decision-making. Values friction can occur both within an individual and in interactions between an individual and others or between an individual and organizations. When one individual holds multiple values. they can come into conflict with each other (e.g. honesty and compassion; individualism and belonging). Typically, an individual will place different levels of importance on different values, which will affect which values have the greatest impact on decision making or action. However, individuals may also be constrained by the values friction that exists between their own values and peers’ values or institutional values and norms (Jensen, Schott, & Steen, 2021).

Consider the following scenarios and how values may impact decision-making.

Values are intrinsically linked to affect or emotion (Schwartz, 2012). When our values are threatened or questioned, it can activate a negative emotional response. Similarly, when our values are realized, it can activate a positive response. For more information on how to navigate the affective nature of generative AI, see the section on Emotional Intelligence.

Fundamental Values of Academic Integrity 

Ignoring AI Use

Ignoring generative AI and continuing on as usual may seem like the easiest option for instructors. However, doing so fails to address several fundamental values of Academic Integrity: trust, fairness, responsibility, and courage.

The absence of an AI policy or practices will not prevent its use. You will still encounter student submissions that you suspect (or know) to be generated with the assistance of AI, as demonstrated in the below image. When this happens, you will need to decide how to address it. Without establishing clear expectations around AI use, this process lacks transparency (a core characteristic of trust) and will impede the ability to apply policies equitably (fairness).

Responsible faculty acknowledge the possibility of academic misconduct and create and enforce clear policies around it. Academic Integrity also requires a willingness to take risks and deal with discomfort (courage).

 

Example of a paper with AI generated text.

Prohibiting AI Use

It’s clear that institutions and instructors need to take some action in response to generative AI. The immediate reaction of many institutions and instructors may be to implement a blanket ban on AI use.

However, completely prohibiting the use of AI could also been seen to fail to address multiple fundamental values of Academic Integrity. It is also unrealistic, as generative AI tools are embedded in many other applications (such as Microsoft Word and web browsers).

This approach lacks: respect, responsibility, trust, courage, honesty, and fairness.

A blanket ban will contribute to a culture of mistrust as it could be seen to be built on the assumption that students’ most likely use of generative AI tools would be to commit scholastic offenses. It also ignores the opportunities generative AI offers to support and deepen learning.

Some students will absolutely use generative AI in inappropriate ways.  However, many other students may benefit from it as a tool to enhance and improve learning and provide supports for diverse learners.

Conversation with CoPilot from January 2025.

Students will also encounter AI tools in their future careers and studies, so by choosing to not consider AI in our teaching, fields, and course designs, we may be failing to prepare students for the future.

Our goal as instructors is to demonstrate respect for the motivations and goals of all learners while still being able to hold individuals responsible for their actions.

A blanket ban of generative AI lacks Trust in students’ ability to use generative AI responsibly and denies them the opportunity to develop important AI literacies. It also lacks the courage to explore if/how AI can enhance our disciplines and teaching and learning practices. Of course, there may be times when it is not appropriate for students to use generative AI tools, particularly if it interferes with students demonstrating the learning outcomes of the course. The key thing to consider is whether it is necessary to prohibit tools and, if it is, that you can clearly explain to students why.

Detecting generative AI

This screenshot was taken from the TurnItIn tool in OWL, but the limitations of AI detection tools are not specific to this tool.

Another limitation of blanket bans is that it means more time must be spent detecting student use of generative AI. Yet, there is currently no reliable way to identify AI generated content. Current tools for doing so are unreliable and biased.

For example, one study shows that AI detection tools flagged more than half of the submitted essays from non-native speakers as AI-generated (Liang et. al., 2023 ).

Because of these limitations, the use of AI detection tools innately does not provide true evidence of academic misconduct (thus lacking honesty) and will not process all students’ work equitably (thus lacking fairness).

 

Crafting the Syllabus Statement 

Syllabus Statement Example 1

Syllabus Statement Example 2

Emotional Intelligence

In this section, we will consider 4 key components of emotional intelligence (adapted from Goleman, 2020):

1. Self-Awareness: The ability to recognize your emotions and name them. Recognize the emotional response that you have to generative AI and how this reaction may change over time in different scenarios or as your knowledge of AI changes.
2. Self-Regulation: The ability to manage your emotions and prevent them from controlling your behaviour and response to situations. Recognize how your emotional response to generative AI may impact your desire to use it in your teaching or other work and your response to others using it in their work.
3. Empathy: The ability to recognize the emotions of others. Recognize that different people will have different emotional reactions to AI technologies and use, and that this will impact their adoption of or resistance to AI tools.
4. Social Skills: The ability to communicate clearly in a way that acknowledges your own and others’ emotions.

The following section will provide multiple case studies exploring different ways in which discussion and use of generative AI may evoke emotional responses and require Emotional Intelligence. For each case study, you will be asked to reflection on the 4 key components of EI.

If you do integrate generative AI technologies in your teaching, learning activities, or assessments, you will also be introducing a need for learners to increase their emotional intelligence with regards to how these tools are being used.  The following case provides an example of EI considerations from a student perspective.

• Goleman, D. (2020). Emotional intelligence (25th anniversary edition). Bantam Books.

Context Engineering

Many AI interfaces allow users to communicate using natural language, rather than programming or machine language. This makes it easier for general audiences to use generative AI technologies; however, as with human-to-human communication, it’s important to use clear and specific language to get the desired results. The process of designing effective prompts has been referred to as prompt engineering. 

Effective prompt writing can help you more quickly achieve the results that you’re seeking, minimizing the number of queries that you need to make.

The following tips can help you write effective prompts to get the results you’re looking for:

For more tips, check out the Prompt Engineering Guide or the Prompt Library from Ethan Mollick .

As AI tools become increasingly sophisticated, writing good prompts is often not enough. Successfully using generative AI tools requires context engineering. Context Engineering means building a system using the right information and tools to get the desire result (Schmid, 2025). This has become increasingly important with the development of Retrieval Augmented Generation (RAG), or the ability for generative AI tools to conduct real-time information retrieval; and with the shift towards Agentic AI, or AI systems with the ability to act autonomously, with limited human supervision, to complete a goal (Stryker, n.d.).

Prompting is still an important aspect of context engineering, but there are multiple other contextual elements to consider, including:

1. Systems prompt or instructions – the rules that have been preprogrammed into a specific tool to limit what it is able to do. For example, these could include guardrails to prevent harmful or unethical responses.
2. Context window of the current conversation – all of the prompts and responses that have already occurred in the current session
3. User profile/memory – the knowledge base built on previous interactions, which might shape communication preferences, focus, or other aspect of the response
4. Access to external resources – allows information to be retrieved in real-time (RAG)
5. Tools and APIs – impacts the ability to complete tasks, like sending email or adding events to calendars

For a deeper understanding of how each of these elements impact the generate output, take a look at my conversation below asking ChatGPT 5 to help generate a lecture plan. I ask ChatGPT to provide a breakdown of the context. You can view the full conversation here.

A conversation with ChatGPT 5 from August 22, 2025.

Critically Appraising AI Outputs

As with any source, it’s important to critically evaluate the output from generative AI tools. As discussed, output can be prone to inaccurate or misleading information or biased representation. Traditional tools for evaluating sources might be helpful, but the nature of generative AI output and the lack of transparency can make it more difficult to assess things like Currency, Authority or even Bias.

Fact-Checking AI Output

The SIFT method was developed by Michael Caulfield for verifying claims made in online sources (Caulfield, 2019). It can be adapted to assessing the validity of AI output.

Modified SIFT method for assessing AI output.

Caulfield, M. (2019, June 19). SIFT (The Four Moves). Hapgood. Retrieved from https://hapgood.us/2019/06/19/sift-the-four-moves/

DAIR.AI. (2023). Prompt Engineering Guide. Prompting Guide. https://www.promptingguide.ai/

Mollick, E., & Mollick, L. (2024). Prompt Library. In More Useful Things: AI Resources. More Useful Things. Retrieved from https://www.moreusefulthings.com/prompts

Office of Equity, Diversity & Inclusion, Western University. (2025, July). Inclusive Language Guide [PDF]. Western University. Retrieved from https://www.edi.uwo.ca/edi-learning/guides/Inclusive-Language-Guide.pdf

Identifying the potential impact of generative AI on course learning outcomes

Some learning outcomes may be easily demonstrated through the use of generative AI tools, which may be undesirable if the student needs to master the learning outcome independently.

Alternatively, incorporating generative AI into an activity or assessment may increase a student’s ability to successfully achieve and demonstrate the outcome, potentially leading to higher level learning.

By thinking about how generative AI may impact learning and demonstrating each learning outcome, we can identify which assessments and learning activities may need to be redesigned so that the outcomes, learning activities, and assessments all align to support each other.

Learning Outcomes

Bloom’s taxonomy is a helpful method for determining whether generative AI can “demonstrate” certain learning outcomes. (more information about learning outcomes and Bloom’s Taxonomy is available here ).

Bloom’s Taxonomy shows 6 levels of higher order thinking skills, often used as a framework for identifying learning outcomes.

Generative AI tools are more likely to successfully demonstrate tasks on the lower levels of Bloom’s taxonomy because:

• Generative AI is particularly good at simple tasks of knowledge reproduction, description, or explanation. When learning tasks at these levels are completed using generative AI, students who have not yet developed the ability to perform these tasks will lack the opportunity to develop them. However, students who have successfully learned these skills may benefit from using generative AI tools to more quickly do this work so that they have more time to concentrate on tasks in the higher levels of Bloom’s Taxonomy.
• Generative AI can be used to support higher order cognitive tasks as well but may struggle to produce accurate or high-quality output for more complex tasks. For these Learning Outcomes, generative AI may be a helpful tool to scaffold certain tasks and increase student learning.

NOTE: Learning Outcomes in the affective (related to attitude or emotions) or psychomotor (related to physical skills) domains of learning are less likely to be successfully demonstrated by generative AI tools.

Analysing learning outcomes

As you review your course Learning Outcomes (LOs), it’s important to reflect on how each LO could be impacted using generative AI.

1. The ability to demonstrate some Learning Outcomes will not be impacted by using generative AI tools.

For example:

• Deliver effective oral presentations in professional settings to a large interdisciplinary audience.
• Proficiently bandage a sport injury to facilitate safe return-to-play.
• Use appropriate speech levels in conversation, including the correct forms of polite (-e/a yo) as well as deferential (-supnita) language.*

*This Learning Outcome was generated with the help of ChatGPT.

2. For some learning outcomes, relying solely on generative AI will not allow students to develop that skill.

For example:

• Paraphrase an excerpt from an academic article to accurately convey the key ideas to a generalist audience.

Generative AI tools would be able to complete this task without the student demonstrating that they have learned the fundamental skill of paraphrasing or audience differentiation.

3. For some learning outcomes, generative AI tools could remove barriers or enable students to focus on higher level cognitive skills in a way that supports learning. This is called Scaffolding.

What is Scaffolding?

The Zone of Proximal Development is the learning that is possible with the assistance of others.

Scaffolding works by initially providing support for learners on tasks that they may not be able to complete without assistance. These supports are then removed as students gain higher proficiency levels.

The concept of scaffolding grew out of Vygotsky’s work on the Zone of Proximal Development, an area of learning where students don’t yet have the skills to independently   demonstrate a learning outcome but can learn to do so by building on their prior skills and knowledge with the guidance and support of others (TMU, 2021).

Traditionally, ‘assistance’ has meant other people, like learners, teaching assistants, or even peers, but generative AI could also help provide this assistance. The next part of this section will outline some ways in which instructors can use generative AI to help scaffold learning.

Oregon State University published an updated version of Bloom’s Taxonomy that distinguishes between Distinctive Human Skills and how generative AI can supplement learning. This can be a helpful tool for understanding the ways in which generative AI can support learning.

Oregon State has reimagined Bloom’s Taxonomy, identifying Distinctive Human Skills and How GenAI can Supplement Learning.

When considering the potential impact of generative AI on learning, it’s important to distinguish between ways in which generative AI tools can replace learning and ways in which generative AI can support learning. Consider the following examples:

Learning with AI

The follow chart provides a summary of some of the potential roles that AI could play to support learning, identifying both the benefits and the constraints or limitations of each.

(Adapted from Mollick & Mollick, 2023) 

Assessment Design

Because generative AI technologies have the potential to impact learning, both negatively and positively, it’s important to consider how assessment designs and practices might need to change.

When reconsidering assessment designs, it’s important to start with the intended learning outcomes. The decision tree below offers help on determining if/how assessments might need to change in light of generative AI.

Levels of AI Integration in Learning and Assessment

AI Planning

Generative AI can help support learners in planning their approach to assessments. It can be useful for:

• managing their time by helping them break down complex tasks into manageable chunks
• identifying a topic or refining ideas to an appropriate scope
• generating research questions
• identifying useful resources to support research

Generative AI can be helpful to support executive functioning, including setting goals and project and time management. For example, students could use generative AI to create a project plan for larger assignment, to break process-based assignments into manageable tasks, or to help with setting short term goals. These tasks could be a barrier for some learners if they are not explicitly part of the assignment objectives or if they have not been taught.

In the example below, CoPilot generated a step-by-step guide to writing a research paper.

A conversation with CoPilot showing how CoPilot can support time management.

Idea generation can present a barrier early on in a process-driven assessment. Students may spend too much time trying to decide on a topic, have too many ideas to explore all of them, or just need a place to start so that they can move on to later stages of the process. Since generative AI tools are trained on large data sets, they can often be used for initial ideation when approaching a task. Using them in this way can provide students with content to respond to and refine rather than generating ideas from scratch. Students can develop ideas generated by LLMs into more robust or context-appropriate topics. In the example below, CoPilot narrowed down the broad topic of “Social Determinants of Health” to more specific topics appropriate for a 10-minute presentation.  

A conversation with CoPilot showing how CoPilot can help provide examples for a topic.

These are just some examples of how generative AI can help learners get started on an assignment, removing unnecessary barriers in the planning process. However, it’s important to keep constructive relevance in mind when deciding if generative AI can be used in this way. If learners use generative AI to complete any of these planning tasks, are they still achieving the intended learning outcomes?

AI Collaboration

Generative AI technologies can be used to support learners at other steps in process-based assignments.

Process-based assignments require a series of tasks usually completed in a specific order. As you look at the following example of a writing process, consider what skills are required at each step in the process. Where could generative AI support learners without detracting from the intended learning outcomes of completing the final paper? Does this change if students have already successfully demonstrated the required skills for this step?

 

Breaking writing into a process can be helpful for identifying if/where generative AI tools can be supportive.

Generative AI is particularly strong at written coherence and can be a useful tool for providing feedback on writing. Rather than making the revisions (replacing learning), generative AI can be used to identify strengths and areas for improvement and suggest changes.

Example Prompt: You are an advisor in the university writing centre. You will give feedback on my paper. You will provide balanced feedback, highlighting both strengths and areas for improvement. Please pay specific attention to clarity of my writing and the completeness of my arguments. You will not provide or offer revisions, but if I suggest a revision, you will offer further feedback to help me refine my writing.

Generative AI can be a useful tool for providing feedback and focus for revisions.

 

Documenting AI Use

If you allow learners to use generative AI throughout an assignment, it is important to give clear instructions on where generative AI can and cannot be used and how students should document and cite their use of generative AI.

Popular citation guidelines have been updated with guidelines for citing generative AI.

APA Citation

Name of Company/creator of generative AI Tool. (Year). Name of the generative AI program (model of program) [Large language model]. URL.

Example:

Microsoft. (2025) CoPilot (GPT-4). [Large language model]. https://copilot.cloud.microsoft

MLA Citation

Author/Creator. “Name of chatbot.” Title of platform where accessed, Full URL, Date Accessed (optional).

Example:

Microsoft. “CoPilot”. CoPilot. https://copilot.cloud.microsoft March 25, 2025.

See the University of Waterloo’s Citation Guide for more details.
You can also state requirements on how you want students to document their use of generative AI in completing the assignment.
Consider the following questions:

• How much detail do you require?
• Do you want students to submit the full dialogue? Do you want them to indicate the types of tasks that they used generative AI for?

The University of Waterloo has developed a tool for detailed documentation of generative AI use. See this resource from the University of Waterloo for more information.

A great way to engage with the complexity of generative AI discourse is to join us in the Generative AI Challenge.

Share your own resources below: