By Ryan Hurnevich
During our study of pre-Columbian North and South America, U.S. History Honors students were tasked with an assignment that posed the following question: How can we utilize ChatGPT to enhance our understanding of history while also realizing the limits of AI and testing its accuracy?
The specific question is part of a much larger question in the field of education, and the world at large, as all grapple with the proliferation of artificial intelligence and its potential. In the social studies department, teachers are well-versed in how to prompt AI software to enhance their productivity, research, and consider best practices for lesson design. In short, the full-grown humans feel prepared enough to engage with the robots, but what about the growing humans whose care we are entrusted with? How best should our students engage, if they engage at all?
For our project about the peopling of North and South America, students first read a traditional academic text, “First Peoples: A Documentary Survey of American Indian History” by Colin G. Calloway, which explores various theories of how and when the Americas became populated with human beings. Next, students were provided a text generated by ChatGPT by prompting OpenAI’s artificial intelligence chatbot with the following: “What are the most up-to-date theories on how and when the Americas were populated?” The task then became a sourcing and fact-checking exercise. Students needed to confirm what the AI-generated text wrote by comparing it to what the traditional academic text had to say about the subject. If the AI-generated text provided information not found in the traditional academic text, students needed to research, source, and confirm whether or not the AI-generated claims were accurate.
Our findings were eye-opening. We discovered the value of traditional academic texts written by a human expert, as much of what ChatGPT produced was already discussed in the traditional academic text. Additionally, we discovered that AI has the potential to provide more up-to-date information. A few of the leading theories for the peopling of the Americas have recently been updated, informed by new understandings of genetics that weren’t available when the academic text was written.
Perhaps most importantly, though, we gained valuable insights into the potential biases of ChatGPT. In the traditional academic text that we read, Calloway included various indigenous perspectives on the peopling of North and South America. He explained the creationism stories of multiple tribes, such as how the Iroquois believe their ancestors fell from the sky, and the New England Algonquin believe their ancestors were transformed from ash trees into people. Other indigenous perspectives assert that their people did not cross the Bering Strait or travel by sea across the Pacific, as outlined in the Coastal “Kelp-Highway” Migration Model. Instead, they assert that they have always been here. While we can’t fully know the motives of why ChatGPT would not include the indigenous perspective on the peopling of North and South America, we can use the exclusion of this perspective as evidence that such indigenous history is lacking in the internet sources from which ChatGPT derives its content. Helping students contextualize that information remains safely in the hands of the humanities.
Beyond updating our knowledge of this rich history and how it helps appraise our modern understanding of what it means to be “American,” we also stepped into the new frontier of AI as a research tool. The consensus was clear: this emerging technology is only as strong as its users. If users are thoughtful, clear in prompting, skeptical, and equipped to verify the information produced, artificial intelligence can keep students of history on the cutting edge of knowledge and research. Without those uniquely human guardrails, however, AI has the potential to lead history students astray.
