Formative is launching tools powered by ChatGPT’s Artificial Intelligence (AI). These platform integrations hold so much potential, both to expand the boundaries of what Formative can do and to give educators and students valuable time back in the classroom. However, at Formative we also understand that with great power comes great responsibility. Run a Google search about AI or ChatGPT, and you’re likely to find articles about bias.

Since AI tools are “trained” using the World Wide Web, their ability to learn is limited to what exists online. As vast as the internet is, much of what you’ll find reflects a very limited demographic: mainly people in the western world, whereas across the globe, we have a diversity of thoughts, races, genders, belief systems, and socio-economic statuses.

AI tools generally do not represent these broad perspectives. When such tools learn from primarily one demographic, it skews their output to hold similar biases towards people with a lot of privilege. For example, both the DALL-E 2 and Stable Diffusion photo-generating tools primarily output white individuals on the majority of prompts about humans.

That said, the developers building these programs - and many more like them - are striving to diversify their outputs. Acknowledging and addressing the potential for bias - and taking care to thoughtfully interrogate AI results for biased content - helps reduce any such instances that might show up. Formative always strives to keep these matters top-of-mind when developing uses for AI.

To that end, we have prepared five tips to help you minimize bias when using AI-integrated learning tools. After all, the best way to prevent any issues is to be proactive, to think about them ahead of time. Furthermore, this work also increases the quality of the tool itself.

• Be specific! Choose your wording intentionally. When asking an AI for a prompt, add in specific words to broaden the AI’s scope. Instead of asking for an essay about everyone who goes to the park, you could ask for an analytical essay of exactly five paragraphs that focuses on the diverse experiences and talents of those people that visit a park.

• Review! Do not assume that an AI’s answer is true and error-free. AI tools can’t always support a specific context, and they often struggle to understand a sarcastic tone. Always review anything an AI creates for you: read it, question it, and make sure nothing is out of place.

• Critically evaluate! After you’ve reviewed the AI’s work, be critical. If you’re using gendered examples, are they split equally? If you’re seeking content of people walking through a park, did the AI provide a diverse representation of races, ages, gender expressions, body sizes, and abilities?

• Add! If you find that AI-generated work lacks varied representation, especially when a voice or person should be included in a specific topic, add extra material beyond what the AI provided.

• Encourage diversification! The best way to address bias is to expand the demographics of people working in STEM (Science, Technology, Engineering, and Math). This process could simply mean encouraging young minds to explore STEM interests; it could also mean engaging with organizations like We All Code that work to bring STEM to frequently overlooked and historically excluded communities.

AI and ChatGPT are powerful tools that expand the power of individuals and businesses. We are so excited for these technological advances, but we also know that slowing down a bit and reviewing AI outputs will help ensure we are using these tools with the utmost responsibility.