������ý����

Generative AI is “any type of artificial intelligence (AI) that can be used to create new text, images, video, audio, code, or synthetic data” (Rouse, 2023). Although generative AI has existed since the 1960s, most of us only began to experience its impacts on teaching and learning firsthand in the past couple of years (White, 2023).��

Note: This page focuses specifically on generative AI (gen AI). Interested in other types of AI and their role in teaching and learning? Check out this insightful article:

In generative AI, a statistical model is trained–that is, given the opportunity to search for and “learn” patterns–in a large data set. When prompted, it then generates (hence��generative��AI or gen AI) novel outputs based on what is probable given the patterns detected in the training data.��

For example, imagine a generative AI statistical model is trained on all of the text messages you have ever sent. Would it find that the word “thank” is more likely to be followed by “you” (as in “thank you”) or “me” (as in “thank me”)? Let’s assume “thank” is more likely to be followed by “you.” Now, if you prompt it to compose a text message to let your friend know that you enjoyed coffee with them that morning, it will be more likely to include the phrase “thank you” than “thank me.”��

Want to dig deeper into how generative AI works? Explore these expert resources:

• – Learn the fundamentals of how generative AI works from leading researchers.
• – Watch this accessible explainer video designed for educators and learners.
• – Dive into more technical, visual explanations of how specific generative AI models operate.

Generative AI has several significant limitations. Being aware of these limitations and ensuring that you educate your students on them is an essential component of AI Literacy. These limitations include but are not limited to:

High Risk/High Reward

Although some AI tools show promise for supporting student learning, misapplication of AI tools can have a negative impact (;��). Initial studies found that when students use gen AI for customized tutoring and resources tailored for their learning style, outcomes were comparable to traditional education. However, when students offloaded too many tasks to the gen AI, they had worse outcomes–even worse than not studying.��

Hallucinations

GenAI can often “hallucinate,” inventing spurious facts and details. This can be especially dangerous to users trying to learn from the gen AI tool, given that they might not have the subject mastery to identify when facts are fabricated (;��).

Misinformation

Gen AI tools can be used to create false information in any media format. For example, there is software, such as Hume or ElevenLabs, that can reliably clone voices and generate audio. There is also software, such as Veo and Synesthesia, that can generate photorealistic videos with only a text prompt (note that, although it’s currently limited to short clips or stationary angles, that could change soon). In addition, nearly every LLM (large language model) now has an image generation feature that can generate photorealistic images. Although these AI-powered media generators have safeguards, these safeguards are easy to circumvent. Thus, users acting in bad faith could potentially use gen AI to create media to fake events that never actually occurred.��

Manipulation

LLM chatbots are built to effect a communication style and tone that is engaging and potentially flattering of the user (). Moreover, they are programmed such that they can state false information with a tone of confidence. This can make them particularly effective at swaying people’s views ().

Bias

AI has various issues of bias. Its training data is disproportionately based on WEIRD (Western, Educated, Industrialized, Rich, and Democratic) sources, which creates corresponding cultural biases in its outputs (). In addition, gen AI has been shown to reproduce non-inclusive language stereotypes along a variety of lines (Sun et al., 2023; Johnson et al., 2022; Abid et al., 2021, Whittaker et al., 2019). Finally, studies have shown that this bias is present covertly in many cases (); therefore, any decision being made by gen AI should be scrutinized accordingly.��

Data Privacy

Sam Altman, CEO of OpenAI, has made it clear that user logs are saved and that there is no confidentiality (). All users should be cautious of what information they share with these chatbots.��

Copyright Concerns

LLMs are trained on libraries of date that leave debts to contributors unacknowledged, which can constitute a form of plagiarism and theft of intellectual property (). While some cases have started setting precedents (), there are still many cases pending () in this rapidly evolving space.��

Special thanks to Chris Ostro, Assistant Teaching Professor and Learning and AI Strategist with CE's Learning Design Group, for contributing this section and for providing input on this webpage.

As educators, we can provide students who wish to explore gen AI with guidance on how to use AI responsibly and ethically, as well as opportunities to practice essential AI literacy skills. This includes skills like fact-checking, verifying AI-generated outputs, and writing effective prompts.������

Other resources:��

• Cornell University Center for Teaching Innovation. (n.d.).��.

• Notre Dame Learning. (n.d.).��.