Initial Publication Date: September 22, 2025

Strategies for Effective Use of AI as a Geoscience Educator

Much attention has been focused on the use of generative AI by students. As an educator, one of the best ways to build the foundational knowledge you need to give students informed guidance is to understand how to use it effectively, and when to avoid it, in your own work.

Use AI in Ways that Anticipate Bias and Hallucinations

AI sometimes provides responses that have all the hallmarks and confidence of an expert, but are false, misleading or biased in ways you wouldn't expect from an expert. AI's mastery of language tempts us into evaluating its responses using the same set of intuitions we have developed for contextualizing conversations with humans. But AI tends to fail in ways that don't align with those experiences. So we need to develop new habits when interacting with AI to accommodate its behaviors. We need to assume there will occasionally be hallucinations and subtle errors and adopt strategies to mitigate these problems. Here are some strategies:

  • Have an independent mechanism in place to verify AI responses. If you're working in an area where you have expertise, this might be as simple as reviewing its response. It might mean looking up the papers it cites or the articles it claims informed its response. If the AI is solving a numeric problem (best achieved by having it generate traditional deterministic code ) solve the same problem by several methods and make sure the responses match.
  • Use Generative AI to brainstorm -- as a catalyst for developing your own ideas. Since you aren't using its responses directly, errors won't be as much of an issue. As long as you apply your own good judgment along the way. But beware....
  • You may not want to start your brainstorming with AI. While it can be a great brainstorming tool, starting a creative session by asking an AI may end up unintentionally blinding you to good ideas you might have come up with on your own. Start with your own ideas and then ask the AI to react, expand, and move in different directions.

What strategies have you found useful when engaging with AI?

Geoscience Education Examples

As a support in using other computer tools

Generative AI is helpful when you need to do data manipulation with an existing technical tool (R, Excel, Python, etc...) and you don't recall the correct syntax, or how the tool implements the operation you have in mind. You can describe the issue in plain language and AI suggestions will often lead you in the right direction.

How do I read this CSV file of meteorological data into R, modify the temperature column by adding 2 degrees to simulate climate warming, and then write it back to a new CSV file for use in the GLM lake model?

The AI response will likely get you a long way toward the final commands you need (in less time than trial and error with the documentation). But you'll want to review both the R commands and the final output (maybe with some simpler test data) to head off errors introduced by subtle hallucinations. Note that asking the AI to manipulate the data directly (rather than writing code to do it) is likely to be less robust and will need much more careful review. Also note that many popular chat-centric AI's can now be directed to write and execute traditional code (usually Python or Javascript) internally, so this can be a strong approach when you want AI help with a traditional, deterministic computing task.

To automate routine classroom tasks

There are a variety of routine mechanical tasks that are, in theory, able to be automated, but practically haven't been in the past. AI's ability to move from a general description of a problem to producing a solution makes automation possible.
I'm attaching a copy of a class roster and student responses to a survey about how they like to do group work. Please cluster the students into lab groups of 3-4 students based on their responses to the survey. Prioritize putting students who have stated similar working styles together, and students who have indicated particular affinity groups together. Include a brief label next to each student's name so I can understand your clustering.

This sort of task might have been 20 minutes of manual work in the past. But even with this imprecise description, most AI's will produce a useful first take in a few seconds. Note that we've told the AI to include its reasoning to make it easier for the instructor to review and adjust the lab groups before finalizing. If you had a large class and were concerned the AI might mess up and leave some students out of the final rosters, you could follow up with:

Take your clustering list and the original student roster and write a program that will take these as input and confirm that all students on the roster are represented once and only once in the lab groups.

Using AI to check its own work, especially when you can lean on its ability to write and execute code, is a simple way to help ensure errors don't creep in. You can even use this sort of AI-driven check if you had done the lab group clustering manually on your own. Note that there's a potential privacy issue with putting the student survey data into an AI. You'd want to be sure the AI tool you use has appropriate privacy guarantees and that your use of student responses is in keeping with their expectation of how the data will be used. This is also a situation where running a local AI model, where none of the data leaves your computer, could be the right choice - especially with a simple part of the task like anonymizing the data before feeding it to a more capable AI.

As a starting point for generating student problem sets

 Ask an AI to generate possible homework questions based on some readings you've assigned. For example, if you've given your students a reading about the natural resources issues on the Nez Perce Reservation, you might attach the reading along with this prompt:

Create a quantitative problem set with 3-4 calculation questions based on this reading that integrates the socioeconomic data mentioned (unemployment rates, housing occupancy, population statistics) with relevant environmental science concepts like water quality, energy access, or public health metrics.

The questions it generates are unlikely to be entirely satisfactory and may even introduce incorrect science. But they may serve as a useful first step to creating the final question you actually want to use -- perhaps even engaging in a chat with the AI about what you feel is inadequate about their suggested questions will help clarify your own thinking.

As a proxy student to 'test' possible exam questions

You can leverage AI's affinity for role playing by asking it to point out how students might be confused by a potential exam question. This gives you a chance to do a dry run of your exams with very little effort:

Act as an introductory geology student encountering this GPS velocity vectors exam question for the first time - what parts would confuse you, what calculations would you struggle with, and what common misconceptions might you have about extension, stretch, parallel vectors, or 1D deformation?
 
Again, the responses won't necessarily reflect actual student responses and misconceptions. But you can use your professional judgment to decide whether the issues the AI highlights are relevant for your students and how your existing problem might be improved.

What ways have you used AI to assist you as a geoscience educator?

Going Further - AI Prompts to Explore

These examples are just the tip of the iceberg in terms of how you might use AI as a partner in your work as an educator. Here are a few places you can go for further inspiration:

 In every case, you'll want to carefully reflect on whether the time saved, or the improvement in the resulting work from engaging with an AI is worth it. You'll find some clear wins and a lot of areas where the benefits are less clear. But your experience with these trade-offs will serve as a good foundation when you turn your attention to supporting your students in their encounters with AI.