Environmental impact of AI in Prompt Engineering / GenAI - Model Metrics & Evaluation

Metrics & Evaluation - Environmental impact of AI

Which metric matters for this concept and WHY

For understanding the environmental impact of AI, the key metrics are energy consumption and carbon footprint. These show how much electricity AI models use and how much greenhouse gas is released. Tracking these helps us know if AI is using resources wisely and if it harms the planet.

Confusion matrix or equivalent visualization (ASCII)

Energy Use (kWh) | Carbon Emissions (kg CO2)
------------------------------------------
Training Large Model   |  1000 kWh  |  500 kg
Training Small Model   |   100 kWh  |   50 kg
Inference per Request  |  0.01 kWh  | 0.005 kg
------------------------------------------
Total AI Usage         |  1100.01 kWh  |  550.005 kg
    

This table shows energy and emissions for different AI tasks. It helps compare how big or small models impact the environment.

Precision vs Recall (or equivalent tradeoff) with concrete examples

Here, the tradeoff is between model performance and environmental cost. For example, a very large AI model may give better answers (higher accuracy) but use much more energy. A smaller model uses less energy but may be less accurate.

Choosing the right balance means picking a model that is good enough but does not waste energy. This is like choosing a car that is fast but also fuel efficient.

What "good" vs "bad" metric values look like for this use case

Good: AI models that use less than 100 kWh for training and keep carbon emissions low, while still performing well. Efficient code and hardware help.

Bad: Models that use thousands of kWh and emit hundreds of kg CO2 for small improvements in accuracy. This wastes energy and harms the environment.

Metrics pitfalls (accuracy paradox, data leakage, overfitting indicators)

One pitfall is ignoring environmental metrics and focusing only on accuracy. A model can be very accurate but cause huge energy waste.

Another is not measuring energy use consistently, leading to wrong conclusions about impact.

Also, overfitting large models wastes energy training on data that does not improve real-world use.

Your model has 98% accuracy but 12% recall on fraud. Is it good?

This question is about fraud detection, not environmental impact, but it shows tradeoffs. High accuracy with low recall means many fraud cases are missed.

For environmental impact, a similar question is: "Is a model that uses 1000 kWh worth a 1% accuracy gain over a model using 100 kWh?" Usually, no, because the environmental cost is too high for small benefit.

Key Result

Energy consumption and carbon footprint are key metrics to evaluate AI's environmental impact and balance performance with sustainability.