Environmental cost of training AI models in AI for Everyone - Mini Project: Build & Apply

Environmental Cost of Training AI Models

📖 Scenario: You are part of a team raising awareness about the environmental impact of artificial intelligence (AI). Your goal is to create a simple data summary that shows the energy consumption and carbon emissions of training different AI models.

🎯 Goal: Build a clear data summary that lists AI models with their training energy use and carbon emissions, then calculate total energy and emissions to understand the environmental cost.

📋 What You'll Learn

Create a dictionary with AI model names as keys and their training energy consumption (in kWh) as values

Add a variable for the average carbon emission factor (kg CO2 per kWh)

Calculate a new dictionary with AI model names and their estimated carbon emissions

Calculate total energy consumption and total carbon emissions for all models

💡 Why This Matters

🌍 Real World

Understanding the environmental cost of AI training helps researchers and companies make greener choices and raise awareness about energy use.

💼 Career

Data analysts, AI researchers, and sustainability experts use such data to measure and reduce carbon footprints in technology projects.

1

Create AI models energy data

Create a dictionary called ai_models_energy with these exact entries: 'GPT-3': 1287000, 'BERT': 150000, 'ResNet': 50000, 'Transformer XL': 300000

AI for Everyone

# Create the dictionary ai_models_energy with the specified AI models and their energy consumption
# Your code here

Hint

Use curly braces {} to create a dictionary with the model names as keys and energy values as integers.

2

Set carbon emission factor

Create a variable called carbon_per_kwh and set it to 0.475 representing average kg CO2 emitted per kWh of electricity

AI for Everyone

ai_models_energy = {
    'GPT-3': 1287000,
    'BERT': 150000,
    'ResNet': 50000,
    'Transformer XL': 300000
}
# Set the carbon emission factor per kWh
# Your code here

Hint

Assign the value 0.475 to the variable carbon_per_kwh using the equals sign.

3

Calculate carbon emissions per model

Create a new dictionary called ai_models_carbon using dictionary comprehension that multiplies each model's energy in ai_models_energy by carbon_per_kwh

AI for Everyone

ai_models_energy = {
    'GPT-3': 1287000,
    'BERT': 150000,
    'ResNet': 50000,
    'Transformer XL': 300000
}
carbon_per_kwh = 0.475
# Calculate ai_models_carbon using dictionary comprehension
# Your code here

Hint

Use {model: energy * carbon_per_kwh for model, energy in ai_models_energy.items()} to create the new dictionary.

4

Calculate total energy and carbon emissions

Create two variables: total_energy as the sum of all values in ai_models_energy, and total_carbon as the sum of all values in ai_models_carbon

AI for Everyone

ai_models_energy = {
    'GPT-3': 1287000,
    'BERT': 150000,
    'ResNet': 50000,
    'Transformer XL': 300000
}
carbon_per_kwh = 0.475
ai_models_carbon = {model: energy * carbon_per_kwh for model, energy in ai_models_energy.items()}
# Calculate total_energy and total_carbon
# Your code here

Hint

Use the sum() function on the dictionary values to get totals.

Practice

1. What is a major environmental concern when training large AI models?

easy

A. Noise pollution from servers

B. High energy consumption leading to increased carbon emissions

C. Producing excessive paper waste

D. Using too much water during training

Solution

1. Step 1: Understand AI training process

   Training AI models requires powerful computers running for long periods, which uses a lot of electricity.

2. Step 2: Link energy use to environmental impact

   Electricity often comes from sources that emit carbon dioxide, contributing to pollution and climate change.

3. Final Answer:

   High energy consumption leading to increased carbon emissions -> Option B

4. Quick Check:

   Energy use = carbon emissions [OK]

Hint: Think about what powers computers during training [OK]

Common Mistakes:

• Confusing water use with energy use
• Thinking paper waste is a main issue
• Assuming noise pollution is significant

2. Which of the following energy sources reduces the environmental cost of AI training the most?

easy

A. Solar and wind energy

B. Coal power plants

C. Diesel generators

D. Natural gas plants

Solution

1. Step 1: Identify clean energy sources

   Solar and wind energy produce electricity without burning fossil fuels, so they emit little to no carbon.

2. Step 2: Compare with fossil fuels

   Coal, diesel, and natural gas burn fuels that release greenhouse gases, increasing environmental harm.

3. Final Answer:

   Solar and wind energy -> Option A

4. Quick Check:

   Renewables = less pollution [OK]

Hint: Choose renewable energy for lower carbon footprint [OK]

Common Mistakes:

• Assuming natural gas is clean
• Confusing diesel with renewable energy
• Thinking coal is environmentally friendly

3. If training an AI model uses 1000 kWh of electricity from a coal plant emitting 1 kg CO2 per kWh, how much CO2 is produced?

medium

A. 10,000 kg CO2

B. 100 kg CO2

C. 1000 kg CO2

D. 1 kg CO2

Solution

1. Step 1: Understand the emission rate

   Each kWh produces 1 kg of CO2, so emissions scale directly with energy used.

2. Step 2: Calculate total emissions

   1000 kWh * 1 kg CO2/kWh = 1000 kg CO2 emitted.

3. Final Answer:

   1000 kg CO2 -> Option C

4. Quick Check:

   Energy x emission rate = CO2 [OK]

Hint: Multiply kWh by CO2 per kWh [OK]

Common Mistakes:

• Multiplying by 0.1 instead of 1
• Confusing units of CO2
• Adding instead of multiplying

4. A company claims their AI training uses zero carbon emissions but runs on coal power. What is the likely error?

medium

A. Ignoring indirect emissions from electricity source

B. Using solar panels incorrectly

C. Measuring emissions only from hardware manufacturing

D. Confusing carbon emissions with water usage

Solution

1. Step 1: Analyze the claim

   Zero carbon emissions means no CO2 released, but coal power plants emit large amounts of CO2.

2. Step 2: Identify the mistake

   The company likely ignores emissions from the electricity source powering the training, which is indirect but significant.

3. Final Answer:

   Ignoring indirect emissions from electricity source -> Option A

4. Quick Check:

   Indirect emissions matter [OK]

Hint: Check energy source emissions, not just direct use [OK]

Common Mistakes:

• Assuming hardware emissions are the only factor
• Confusing water use with carbon emissions
• Believing solar panels are involved without proof

5. Which strategy best reduces the environmental cost of training AI models while maintaining performance?

hard

A. Ignoring energy source and focusing on speed only

B. Training models longer on coal-powered servers

C. Increasing model size without optimization

D. Using energy-efficient hardware and renewable energy sources

Solution

1. Step 1: Identify ways to reduce energy use

   Energy-efficient hardware uses less power for the same work, reducing total consumption.

2. Step 2: Combine with clean energy

   Using renewable energy sources like solar or wind cuts carbon emissions from electricity.

3. Step 3: Avoid harmful practices

   Training longer on coal or increasing model size without efficiency increases emissions and waste.

4. Final Answer:

   Using energy-efficient hardware and renewable energy sources -> Option D

5. Quick Check:

   Efficiency + renewables = lower impact [OK]

Hint: Combine efficient tech with clean energy [OK]

Common Mistakes:

• Thinking bigger models always improve efficiency
• Ignoring energy source impact
• Believing longer training is better without cost