When Generative AI Summarizes Hurricane Data

This activity was implemented in two different introductory-level, general education courses for non-STEM majors at Penn State University: introduction to oceanography (natural sciences), and when data meets design (interdomain natural sciences and arts). The oceanography courses has no prerequisites, the analog data visualization courses requires college algebra with analytic geometry as a prerequisite. The general education learning objectives include effective communication, and critical and analytical thinking.

Students are first introduced to hurricanes and provided an overview of topics including how hurricanes form, how they are categorized, how they are named, how they are tracked, and the impact on humans and society from the different categories of hurricanes. Students also explore the National Hurricane Center Data Archive and are shown how to access the historic Tropical Cyclone Reports for the North Atlantic and how to read the "spaghetti plots." The content is condensed for the data visualization course.

The first time this activity was offered, it was a take-home activity in place of a class period for the oceanography course. Classes were canceled for Election Day in the fall semester, so this was given to students to complete in place of coming to class. The timing also related to the United Nations celebration of Global Media and Information Literacy Week (Oct. 24-31 annually). Each year, AI relates to the annual theme. The second time the activity was carried out was in-person during a class period of the data visualization course in the spring semester. But even in the spring semester, students were introduced to the United Nations Global Media and Information Literacy Week as a critical connection for why this exercise was important to carry out and is relevant to global conversations and awareness campaigns. Note that students will need a laptop or access to desktop computers to complete this exercise.

The concept goals for this generative AI activity are for students:

• To learn how to produce another form of output in AI (a table) instead of just text
• To explore the inconsistencies of AI output when multiple people use the same prompt
• To learn how to evaluate the output from AI for accuracy against published records (in this case, hurricane archives)
• To be aware of impacts on humans/society if output from a generative AI tool are used without being fact-checked

The goals for students with regards to higher order thinking includes:

• Critical thinking and research skills -- To learn how to evaluate material found online (in this case, produced by generative AI) and how to critically analyze data presented in a data table to determine the accuracy of the data output, to assess what information is missing, and to evaluate the effectiveness to what extent the AI-generated output was in accomplishing its objective from the input
• Reflection before communication -- To think about what it would mean to use the inaccurate/incomplete data produced by a generative AI tool and the impact it could have on people and emergency planning if the accuracy of the output is not verified first

Although not a skill developed by students, one important takeaway for students from this activity was seeing how the construction of a data table for generative AI, even with a specific prompt and openly-available NOAA hurricane data, resulted in errors and inaccuracies in the result. It is increasing the awareness of why students must use caution when expecting information from generative AI tools that is important for students to realize.

This activity starts with an overview and introduction to hurricanes for students (that lecture is not provided on this page). Then, the exercise is presented to students to complete outside of class or during a class period (requires ~40 minutes). Microsoft Copilot is used as the generative AI tool, as that is the tool approved by my institution.

The worksheet provides students all of the prompts necessary to complete the assignment, including the prompt to copy/paste into the generative AI search tool.

Create a table with four columns – the first column is year, the second column is the number of major hurricanes for that year, the third column is the names of the major hurricanes for that year, and the fourth column lists the names of any hurricanes from that year that crossed over any of the Mid-Atlantic states. The top of the table should be 2025, and the table should go back in time to 1995. Align the entries to the left. Add horizontal lines (borders) to separate the header row from the data rows and to separate each data row. The column header should be in bold font.

A table is provided in the worksheet for students to fill out as they compare the AI-generated data table with the data that is on the National Hurricane Center plots. Students are then presented with questions to compare the results. The final question asks students to apply what they have learned with the inconsistencies and incompleteness of the data output from generative AI and how that may impact hurricane reporting. The final question in the worksheet is:

Let's say you are a media producer – either for a newspaper, or local TV, or radio – and it is the beginning of next year's hurricane season in June 2026. You have been assigned to pull together a history of the major hurricanes that have impacted our region of the Mid-Atlantic states, to help people have an awareness and to be prepared for what may possibly happen. Would you use AI to pull together your story? What would the impact be if you only relied on the data from the Microsoft Copilot prompt you entered to learn about the history of natural disasters and hazards?

Generating a hurricane data table with generative AI _ out of class (Microsoft Word 2007 (.docx) 18kB Feb20 26)
Generating a hurricane data table with generative AI _ in class (Microsoft Word 2007 (.docx) 18kB Feb20 26) 

There are several items to make note of for this activity.

• The data table generated by AI will not be the same for every student. Some students may have problems with AI pushing back on generating a table (for example, one student had Copilot asking for the actual data to put in the table). And all of the students have the same prompt. This will be a challenge to troubleshoot any issues and to respond to any questions if students are completing this exercise outside of class, but still requires much moving around the room to help students working on this in class.
• Although this activity was completed for the first time as a take-home exercise, I recommend completing this in-class with students to have a discussion with students while they are completing the activity, especially as they have difficulty getting an accurate response from the generative AI tool. I did have a discussion with my students in the next class period that completed the take-home activity in the fall semester, but it was a more powerful experience for students to be able to look at each other's computer screens and hear their reactions to how inconsistent the data table output was.
• Overall, students were not able to name the Mid-Atlantic states, even students from the Mid-Atlantic. An instructor may want to list the states students use on the worksheet, or change the region to where their institution is if they are located on the east coast of the United States.
• In the end, students showed frustration at how wrong the generative AI tool was but came away with a realization that there can be serious consequences if information from an AI tool is not fact-checked.

This was a low-stakes assignment and was given the same credit as any other in-class activity. Because the generative AI output was different for all students (although none of the data tables were accurate), I gave students full credit for providing complete responses to the application and reflection questions asked at the end of the worksheet. This activity could be extended by asking students to rewrite the prompt fed into the generative AI tool, but I did not want to take more class time on this (my institution wants all students to be taught about AI, and I do only one or two exercises per semester with students to show them the limitations of using this tool).

United Nations Global Media and Information Literacy Week (24-31 October) - https://www.un.org/en/observances/media-information-literacy-week

National Hurricane Center Data Archive - https://www.nhc.noaa.gov/data/