.png)
Problematic :
This research paper focuses on the environmental impact of using artificial intelligence (AI) systems for writing and illustration tasks, compared to the impact of performing these same tasks by humans.
Key Points :
- Comparison of AI vs. Humans' Emissions: When AI performs writing or illustration tasks, it generates significantly fewer CO2e emissions than humans performing similar tasks.
- Quantification of Emissions: The training of models like GPT-3 and BLOOM, and the emissions per query are analyzed. For example, training GPT-3 produces approximately 552 tons of CO2e.
- Additional Considerations: The study acknowledges that social impacts such as job displacement, legality, and rebound effects are not accounted for in this analysis.
- Non-Complete Substitutability of AI: The study emphasizes that AI does not replace all human tasks and that collaboration between AI and humans can be more advantageous.
Methodology :
- Quantitative Analysis: The researchers conducted a quantitative analysis based on previously published data regarding the environmental impacts of AI systems and human activities.
- AI Emissions Calculation: They considered the training costs of models (amortized over many queries) and the emissions per query for AI.
- Human Emissions Estimation: For human activity, they used average annual emissions per person and adapted them to the specific tasks of writing and illustration.
- Direct Comparison: They compared these data to obtain a direct ratio of AI vs. human emissions for the same tasks.
Key Figures :
- GPT-3 Training: Produces approximately 552 metric tons of CO2e.
- ChatGPT: Around 2.2 grams of CO2e per query (including both training and operation).
- BLOOM: Around 1.6 grams of CO2e per query.
- Human Writing: Between 180 and 1427 grams of CO2e per page.
- Human Illustration: Between 690 and 5500 grams of CO2e per image.
- Lower Environmental Impact for AI: Currently, using AI for certain significant tasks is much less polluting than performing the same tasks by humans.
- Future Considerations: It is essential to monitor the evolving environmental impact of AI and its applications, especially in relation to technological and societal developments.
- AI-Human Collaboration: The authors advocate for a collaborative approach between AI and humans, maximizing the strengths of each entity for greater efficiency and sustainability.
To conclude, this paper highlights the importance of considering AI not only as a powerful tool but also as a potentially more environmentally friendly option for certain activities, while cautioning against potential social and legal side effects.
-min.png)
