Generative AI - Faculty Tips
What Is It?
Generative AI, which typically uses the GPT (Generative Pretrained Transformer) or similar LLM (Large Language Model) architecture, is pervasive-available in browser extensions, mobile apps, and being integrated directly into operating systems. Its use is difficult to detect reliably. (For every detection tool created, there is a tool created to beat it.) While the technology has its limitations, it is sufficiently useful and "labor saving" to have earned a likely permanent place in enterprises involving the production of text, code, image and sound. Forecasting its pervasiveness in industry, the arts and academia, we need to teach students to use it with skill and circumspection. There are significant near-term hazards. Overreliance on the technology produces substandard, biased, and inaccurate results. Overuse and misuse in education hinders student learning and cognitive development. As an educator you should consider yourself under strict obligation to become familiar with this tool and facilitate its appropriate use in the learning environment as relevant to your discipline or professional area. What follows are some general pointers to this end. Beneath these you will find links to clearing house sites directing you to a wealth of information about this technology and its implications for education.
- Familiarize yourself with the tool; what practical use can you make of it?
- Enter your assignment prompts into the tool to see how well the results meet your criteria and how the results compare to student work. Experiment - for example, try imposing different constraints or adding criteria (such as, "in three sentences or less...") to see how the tool manages with them. ChatGPT "remembers" earlier sequences in the conversation; ask follow-up questions.
- Use the tool in your area of expertise so you can reliably judge the quality and accuracy. How sophisticated, consistent, reliable are the results?
- Find discussions of the tool in your own discipline or leading professional organization.
- Remember: the tool will provide different responses when repeatedly given the same prompt.
- When integrating use of tools into course assignments, guidelines and expectations should be clear. It is appropriate to expect and demand original work on key components of assignments.
- Use the tools for brainstorming.
- ChatGPT allows users to share links to conversations, enabling you to share results with students and require students to do the same.
- Task students with evaluating AI-generated results for accuracy, bias, sophistication, originality. Have students correct, improve and refine the results.
- The widespread availability of these tools changes the baseline. Adjust expectations upward. Expect students to surpass AI-generated solutions in their own work. (Perhaps a high CopyLeaks score becomes merely a signal of poor or mediocre quality rather than a basis for accusations of plagiarism.)
- ChatGPT presents students with a significant temptation to disengage or cheat. In response, design multiple, smaller assignments worth fewer points. The lower the stakes, the less incentive to cheat
- Similarly, assign fewer points to any elements of assignments where ChatGPT can deliver the goods. Consider additional exams or testing
- The architecture constructs results based on statistical probability rather than logic or an understanding of the world. It does not generate original insights ("stochastic parrot").
- The results are based on training from massive quantities of data, but the data do not reflect the world's diversity ("AI colonialism").
- Despite enormous efforts at screening and curating the training data, they still contain bias, inaccuracies and inflammatory language, and so do the AI-generated results. ("Garbage in, garbage out.")
- At time of writing, ChatGPT (and Bing) was trained on 2021 data, so subsequent trends, events and discussions are missing from the results.
- Warn students of the dangers of "automation bias" - our human propensity to favor suggestions from automated decision-making systems and to ignore contradictory information, even if it is correct.
In our opinion the argument that "it is just a tool" (like a calculator or word processor) is obtuse. Generative AI produces extensive content (essays, lines of code, solutions to math problems, visual art, music), that mimics the products of sophisticated cognition, creative thinking and problem solving, activities thought to be intrinsically valuable for giving meaning to human life. The challenges it poses to society (and education) are unprecedented.
- There are privacy implications - the text entered can be linked to your identity and shared.
- Using ChatGPT is providing free labor to OpenAI in their product development.
- OpenAI has been exposed for its inhumane labor practices targeting the most vulnerable. For example, Kenyan workers were paid two dollars an hour to review distressing graphic content. Some of these workers experienced mental health issues as a result of this work.
- AI technology today represents the concentrated power of a few companies that have access to troves of data and vast computational resources. In the U.S. the use and development of this technology are essentially unregulated and fundamentally undemocratic.
Links to Generative AI