Age of AI

Since November 2022, students (as well as faculty) have had access to Large Language Model (LLM) Generative Artificial Intelligence (AI).  Students can use AI to complete some assignments but

• that may not be the instructor's intent, and can be an academic integrity violation

• the AIs do somethings very well, and other things not so well, so students may not benefit from the AI's assistance.  

It can be challenging for instructors to keep up with developments in AI.  COLI maintains a resource to assist faculty in developing, implementing, and assessing student work amidst AIs, but here are a few quick tips:

Briefly: Large Language Model AIs

What are currently called Generative AIs are language engines.  Consulting millions of lines of human-written text, they generate responses to prompts by determining, probabilistically, what is the next best word, phrase, sentence, and paragraph.  They simulate, rather than perform thinking.  They are remarkably capable of producing credible, if formulaic prose that describes events, people, places, and theoretical concepts.  LLM AIs draw upon the the texts they were trained with, and in some cases have access to the current internet.  

However, AIs have limits; problems arise when they are prompted to simulate critical thinking, or provide things that are not in their training corpus.  As they simulate human prose, LLM AIs are liable to provide false but plausible-sounding information, what experts call hallucination but in COLI we call fabrication.   More mundanely, AIs may make mistakes remarkably similar to new students in a given topic: poorly understanding and applying theoretical concepts.  Obviously, AIs are only capable of generating content on the internet and as such, may not help in oral assessment or real-time discussion activities.

Experiment with AIs

Every professor in every discipline needs to experiment with AIs.  There is no assurance whether or not an AI can do coursework for a student, unless the professor determines this by asking the AIs to do assignments similar to those a professor assigns to students.  Faculty also come to better understand LLM AI's implications for thier discipline or profession.  

After experimenting with AIs, be clear to students what you learned, and even provide examples.  This can help them see where (over)reliance on AI may be harmful in your course or discipline.  It also makes clear that you are familiar with what AIs will likely present.  (When AIs are prompted in similar ways, their responses even to different uses can be remarkably similar and even use the exact same language.)  

Clear Course Policies

Make clear to students, within your syllabus or assignment instructions, what the rules are concerning AI.  Can they use AI?  If not, make it explicit that you want to see the student's work and not that of an AI.  If they can use AI, then for what?  Brainstorming or outlining?  Writing text that the student then edits and augments?  Describe specifically where students can use AI to complete work in your course.  Reference your experiments, because there you determine what AI is capable of doing with respect to your scholarly or professional practice.  This informs how you expect students to use AI.

Beyond AIs

Even brief experimentation prompts the question:  what must my students be able to do, stemming from learning goals and objectives, that AIs cannot do?  How must I assess those skills, abilities or procedures?   

Obviously, if a student must present to classmates via Zoom or a video recording, AIs cannot help with the presentation itself.  AIs may provide content, or even compose a script, but if the assignment is for students to provide a candid set of talking points, rather than a formal presentation, use of AI may be awkward and obvious.

In the course of your experimentation, you will likely determine that AIs fall short of simulating certain types of critical thinking.  For example, they struggle to match theoretical concepts to real-work examples.  While an AI might pose a fairly good hypothetical example of, say, institutional racism, they are hard-pressed to come up with theoretical examples unless their training corpus includes explicit examples of that concept.