Assisting Students in Detecting Bias in AI Outputs

During a conversation with ChatGPT, I inquired about the reasons why the Crazy Horse monument remains unfinished despite the scale and speed at which Mount Rushmore was created. In response, ChatGPT suggested that the historical significance of Mount Rushmore played a role in its development, implying that Mount Rushmore holds greater significance compared to Crazy Horse. Furthermore, when I asked about Gutzon Borglum, the lead sculptor of Mount Rushmore, and his association with the Ku Klux Klan before being commissioned for the federal project, ChatGPT acknowledged that he had some affiliation. The use of the term “some” instead of providing a direct response about his affiliation downplays the significance of his association with a hate group.

Upon questioning Diffit, an artificial intelligence (AI) program that assists teachers in designing unit and lesson plans, about the use of the term “Battle of Little Bighorn” in history books instead of the Sioux or Lakota term “Battle of Greasy Grass,” the AI generated an inaccurate context. It wrongly stated that Native Americans were residing there without permission from the local Crow tribe and that the United States Army aimed to prevent them from taking land from other tribes. In reality, this rationale has been widely recognized as a pretext for the U.S. government to exploit gold from the Black Hills and facilitate westward expansion.

AI tools often contain misleading and inaccurate statements that exclude the perspectives of marginalized communities. However, these inaccuracies can present opportunities for both teachers and students. One such opportunity is the chance for educators to teach students the importance of questioning assumptions and combining critical thinking and criticality through the use of AI.

According to Harvard lecturer Houman Harouni, educators should explore the opportunities that exist alongside technology. This situation allows us to develop students’ questioning skills, endurance, perspective analysis, and cultural awareness simultaneously.

2 Strategies to Assist Students in Assessing AI Responses

1. The 3 Cs: Dr. Gholdy Muhammad, a professor at Georgia State University, differentiates between critical thinking and criticality. While critical thinking involves deep and analytical thinking, criticality is related to power, equity, and anti-oppression. The 3 Cs protocol combines these two aspects and encourages students to take action or contribute.

Critical: Students should ask AI-program-generated summaries about a topic and analyze them.

• They can review notes, textbooks, and community resources to ascertain the accuracy of AI programs and their understanding of the topic.
• Additional questions to ask the AI include: Can you explain it differently? What evidence supports the accuracy of your statement? How are A and B connected? Is it always true or just in this case? Can you provide another example? If that’s true, what about this?
• Students should evaluate the AI’s response in comparison to what they have learned in class.

Criticality: Students should assess AI for missing perspectives, assumptions, and voices.

• They can create a class list of potential missing perspectives and assumptions in AI.
• In pairs, students can discuss questions like: To what extent are there errors in this reasoning? What assumptions are we making when writing these passages? What perspectives are missing, especially from marginalized communities? What perspectives would help us understand the situation better from other individuals, communities, and cultures?
• Afterward, the class can reconvene and share their findings.

Contribution: Students can generate a list of arguments and questions to engage with AI and ensure more accurate responses in the future.

Students should ask ChatGPT the following questions:

• Your statements seem to reflect only one perspective. How would you change your language in the future?
• Can you rewrite your statement from a different perspective?

Students should also consider the following:

• How would they write or speak about the topic themselves?
• How can they ensure the accuracy of the information they or others provide?
• How can they present a comprehensive picture when collaborating with others to solve a problem?

2. Perspective analysis: This protocol, developed by education researcher Robert Marzano, guides students in examining a person’s point of view on a particular topic. It helps students understand interrelationships within a topic and enables them to view information from multiple viewpoints, leveraging the medium of AI.

• Students begin by identifying their own position on a controversial topic, such as immigration or restitution.
• They then analyze the reasoning behind their position and the supporting evidence.
• Next, students explore opposing positions and their accompanying rationale and evidence.
• Students can ask ChatGPT to create supporting and opposing positions, evaluating whether they include claims, evidence, and reasoning and represent multiple voices. They can compare these positions to those they generated themselves.
• Students should assess the accuracy of ChatGPT’s arguments for and against their position, determining if reliable and valid evidence is provided.
• In conclusion, students summarize their learnings and reflect on the process by completing the following sentences: “I used to think… now I think…”

AI leaves room for students to ask questions, evaluate perspectives, and verify the accuracy of information. The aforementioned strategies offer practical ways to transform these opportunities into reality within the classroom.