Teaching the Environmental Impact of AI Through PBL

In a previous article, I covered how secondary students can enhance their learning and content creation skills in projects using generative artificial intelligence (AI) platforms and AI-empowered tools. Now that they understand how these tools work, let’s take their learning further by having them examine the real-world impact of AI use on the environment.

A math- and science-themed project-based learning (PBL) unit that examines scaling industrial carbon footprints and environmental responsibility will present students with opportunities to make their AI use more efficient and less harmful to the environment.

Explain the Environmental Costs of AI Use

Many students don’t fully comprehend the massive environmental cost of using AI. Every query on ChatGPT or a text-to-image tool requires electricity, water, and the physical infrastructure of a data center to process the request. GPT-4 queries can consume 0.42 watt-hours of energy—and when scaled with the hundreds of millions of daily queries, the use of AI can be compared to the electricity usage of more than 35,000 homes in the United States.

Water in data centers is evaporated for cooling, and although it returns to the atmosphere, it still depletes local freshwater supplies. The water required to train AI is significant, and researchers predict demand will continue to increase. Training GPT-3 alone consumed an estimated 700,000 liters of water in roughly two weeks. Researchers predict AI usage could grow to 4–6 billion cubic meters of water by 2027. Additionally, carbon emissions from training large language models are comparable to and in some instances exceed emissions from high-pollution industrial processes in other industries like aviation, cement production, and steel manufacturing.

Engage Students in Inquiry, Research, and Data Gathering

Teachers can guide students through a PBL activity focused on investigating AI’s environmental impact. This type of inquiry can help students understand the hidden cost of everyday AI use while piquing their interest in energy and sustainability. Assign students to small teams and direct them to investigate AI’s effect on the environmental aspects of energy use, water consumption, and carbon emissions. 

Note: Teachers who don’t specialize in science or math can collaborate with their STEM colleagues to support the data analysis explained below.

1. Explore Energy Use Related to AI Queries: Students can learn to estimate the amount of energy needed for common AI queries and model training using figures reported by researchers—such as around 0.42 watt-hours for a GPT-4 query, as I mentioned already. Calculations can also be made for earlier models, like 0.34 watt-hours for GPT 3.5 queries, as reported by OpenAI. These numbers can be multiplied to estimate usage at scale and compared to everyday real-world energy that kids are already familiar with.

Calculation Example: 2,000 GPT-4 queries x 0.42 watt-hours = 840 watt hours, which is the same amount of energy needed to power an LED light bulb for more than 80 hours.

2. Investigate Water Consumption From AI Training: Have students explore how water is cooled in data centers and learn about the amount of water required for training AI models. Students can relate reported estimates (e.g., 700,000 liters for a single GPT-3 training cycle) to things they’re familiar with, like household water usage.

Calculation Example: 700,000 liters ÷ 1,135 liters (the average daily household usage reported by the Environmental Protection Agency) = over 616 days or 1.7 years of water consumption for one family.

3. Calculate AI Carbon Emissions: Help students understand AI’s impact on global emissions by having them calculate carbon dioxide (CO₂) output from AI energy use. Then have them convert kilowatt-hours (kWh) into estimated CO₂ emissions using a standard conversion factor (e.g., 1kWh ≈ 0.7 kg CO₂, based on the U.S. national marginal emissions rate). They can then compare AI’s carbon footprint to other sectors and activities (e.g., transportation, aviation, agriculture, manufacturing). 

Calculation Example: Let’s use the same calculation mentioned in the energy use section:  

2,000 GPT-4 queries x 0.42 watt-hours = 840 watt hours (or 0.84 kWh).

Now use the U.S. marginal emissions rate: 0.84 kWh x 0.7 CO₂/kWh = 0.588 kg CO₂—roughly the same amount of carbon emitted by driving a typical gas car 1.5 miles. 

The EPA Greenhouse Gas Equivalencies calculator is a powerful tool for helping learners explore other comparisons they can easily visualize and comprehend. When students use this tool, have them use the “energy data” option and enter their result(s) in kWh. 

Discover Career Connections

Once students get the hang of making these calculations and comparisons, they’ll be ready to expand their learning into projects that make connections to authentic career pathways. They may consider professional roles in energy auditing, sustainable computing, and nuclear power operations.

Students can explore other important roles in the green tech workforce—such as sustainable systems engineers, environmental data scientists, and AI infrastructure engineers—that seek to make AI more efficient and less harmful to the environment. 

Additionally, professionals such as data center operations specialists, water resource engineers, environmental compliance officers, and cooling systems technicians help manage and reduce the water footprint of high-tech infrastructure.

Emissions auditors, climate data analysts, clean energy policy specialists, and sustainability consultants assist businesses in measuring, reducing, and reporting their carbon footprints. These professionals are instrumental in making AI and tech environmentally responsible.

Here are some project ideas to get your students engaged in deeper exploration of possible careers:

• Create an AI Sustainability Awareness Campaign,
• Design a “Greener AI” Blueprint, and
• Prepare an AI Career Profile and Job Shadow Simulation.