![A vibrant and futuristic illustration of a person's profile with glowing augmented reality glasses, surrounded by intricate digital circuitry and neon lights in red, blue, and orange tones, creating a high-tech, cyberpunk atmosphere.](/offices/digital-learning/blog/ut-ai-conference-takeaways/aiconference.png)
UT System Conference on AI Takeaways for Instructors
Written by Stephen Myers, Instructional Designer
Innovating Teaching and Learning in the Era of Generative AI
UT System Conference | March 6-8, 2024
This spring, I had the pleasure of attending a UT System conference focused on AI
and its impact: Innovating Teaching and Learning in the Era of Generative AI. UT San
Antonio graciously hosted our many discussions, seminars, and keynotes. It was an
excellent time for learning and collaboration. As a result of that time, I have a
few things for faculty here at UT Tyler to consider as AI continues to be an issue
on our campus:
Try it Yourself
Directly trying out a Large Language Model (LLM), such as ChatGPT or Microsoft Copilot, is often the best way to learn and understand it. Without personal experience using the tool, it will be difficult to advise and instruct students on its use and combat its harmful effects in our classrooms. When you interact with an LLM, you can see how it responds to different types of prompts and where it might struggle. Trying giving it a prompt from a writing assignment from your class. Like any tool, it has strengths and weaknesses. And with knowledge of those nuances, you can craft better assignments to circumvent its misuse and find ways to improve your instruction and workflows. Our colleagues at UT El Paso conducted a series of workshops to introduce their faculty to AI tools in which faculty responded in many ways: UTEP helps faculty use ChatGPT, AI in class to curtail academic dishonesty - El Paso Matters
Understand Hallucinations
Our colleagues from the Center for Excellent in Teaching & Learning at UT Dallas addressed
generative AI hallucinations, defined as generated responses that are incorrect, nonsensical,
or unfactual. Many LLMs are not able to say when they do not know something and will
instead create language which it thinks will satisfy the prompt. These hallucinations
are created because the models generate text based on learned patterns and probabilities
rather than understanding or verifying the factual accuracy of the information. Some
models, like Microsoft Copilot, have been found to generate fewer hallucinations because
they are connected to the internet and are able to cite its sources. While other models,
like ChatGPT, will create fictitious sources and journals to mimic the look of a reputable
source. While improvements are being made to prevent generative AI hallucinations,
it is currently not possible to completely eliminate the possibility of errors or
inaccuracies. Here is recent article from MIT on this issue: When AI Gets It Wrong: Addressing AI Hallucinations and Bias - MIT Sloan Teaching
& Learning Technologies
AI Detectors Beware
While we are all hoping for a plagiarism detection tool that is able to identify AI generated content with a percentage, there is not a reliable tool on the market currently that can make this distinction with confidence and consistency. OpenAI even had such a tool which they have since removed due to its unreliability. Other tools like GPT Free and Turnitin’s AI Detector have been labeled “neither accurate nor reliable” by a myriad of academics. Given that generative AI tools are constantly updating, and with new tools release every week, there may never be a way to technically determine when a generative AI tool has been used. Many institutions across UT System shared this current struggle. While, nationally, this has been recognized as an issue: Professors proceed with caution using AI-detection tools (insidehighered.com)
Learning Management System (LMS) AI Design Help is Coming
While we are a Canvas school, it is often useful to see what other LMSs are doing.
One such useful insight was brought up by UT El Paso and their experience with Blackboard’s
Artificial Intelligence (AI) Design Assistant. This tool can generate learning modules,
rubrics, question banks, and assessments to help build your course all within Blackboard.
While currently this is not integratable into Canvas, it is exciting to see these
tools begin to come available for course creation, and we anticipate other LMSs to
release similar products in time. AI Design Assistant (blackboard.com)
Student Panel Discussion
A popular highlight of the conference was the inclusion of a student panel. These
students came from a handful of UT System schools and were mostly made up of tutors,
graduate students and other high performing students. Their insight and use cases
were enlightening as they work with AI, many of them every day, and anticipate using
AI in their careers. UT San Antonio has provided a recording of that final session
featuring the student’s thoughts about generative AI, its potential impact on student
lives, and advice for faculty: Student Experiences with Generative AI (youtube.com)
All in all, the conference was a great time to come together as a system and collaborate
about a pressing issue in our industry and classrooms. I am grateful to the folks
at UT San Antonio’s Academic Innovation and hope we can continue to host targeted
events like this in the future! For more information on the conference and to watch
the keynote speakers at this conference, UTSA has provided recordings in a playlist
available to all: Innovating Teaching & Learning In the Era of Generative AI - YouTube
OReferences:
- Coffey, L. (2024, February 9). Professors proceed with caution using AI-Detection
Tools. Inside Higher Ed | Higher Education News, Events and Jobs. https://www.insidehighered.com/news/tech-innovation/artificial-intelligence/2024/02/09/professors-proceed-caution-using-ai
- MIT Sloan Teaching & Learning Technologies. (2024, May 7). When ai gets it wrong: Addressing ai hallucinations and bias. https://mitsloanedtech.mit.edu/ai/basics/addressing-ai-hallucinations-and-bias/