For Faculty: Generative AI (ChatGPT, etc.): Bias, inequity, security, privacy, and copyright concerns

Bias is very much a concern when we are using gAI, like ChatGPT. From the letters GPT, the "P" stands for "pre-trained."*  The GPT-3 model was  trained on around 45 TB of free web text data from multiple sources which include books, articles, webpages. and more. In simple terms, ChatGPT has absorbed a lot of our knowledge and ideas, and our biases, into its algorithm. For example, a student who launches a prompt about the signs and symptoms of a pending heart attack may not realize the inherent bias in the output results; most of our heart disease research has been conducted on males, who exhibit distinctly different signs and symptoms compared to females. This bias will not be apparent, and may not be easily overcome unless the user already knows about this bias in research and can craft a prompt to compensate for that.

Still, other entrenched biases may present themselves in unexpected ways, and unless a user is specifically engineering their prompts and their conversation with the chatbot in a deliberate attempt to uncover biases, bias can be subtle and play into our own predisposition for confirmation bias. 

Here is just one example, that is making its way around social media currently, giving ChatGPT a chance to solve a classic circa 1970 gender/sex-bias riddle, and to develop some interesting rationales to explain its inconsistencies and double-standards (remember: it learned it from us).

See the link to the chat transcript below, in an image format, and and accessible Word document format.

---------------------

*see the Start Here tab for definitions

In part due to the built in biases of our world, and therefore our technology and the algorithms that make them hum, technology sets us up for inequity whether it's through the access and cost of technology, or the products of technology.

Here are a few thoughts from around the Web on AI as an equity issue.

When interacting with generative AI (gAI) models, you should be cautious about supplying sensitive information, including personal, confidential or propriety information or data. AI prompts and conversations belong to the AI tool and are used in their research and development.

For this reason, please:

• Do not include any PII (personal identifiable information) in your generative AI (gAI) prompts.
• Do not include student work directly into a gAI prompt
• Do not include sensitive information from our institution in a gAI prompt (HR information, student information, IT information, etc.)