Environmental Impact Assessment

  • Checklist / Questions to ask your Development team

    • Have you calculated the power consumption for your AI app for both training the API and collecting and store the data (use one V100 GPU can consume between 250 and 300 watts)

    • Have you calculated your carbon footprint (general rule: one kilowatt-hour of energy consumption generates 0.954 pounds of CO2)

    • Is your AI app using evolved transformer (uses 1.6x fewer floating-point operations per second and requires 1.1x -- 1.3x less training time than a transformer);

    • Are you using a cloud based AI app as it is about twice as energy-efficient as a typical enterprise data center.

    • Are you evaluating ways to optimize energy consumption ie how frequently are you retraining, using more efficient hyperparameter search methods, and employing more energy-efficient hardware.

  • External Resources - Tools to use

    • CodeCarbon, estimates the carbon footprint of computing, specifically the power used by privately hosted data centers and the underlying infrastructure from cloud providers.

    • Maching Learning Emissions Calculator

    • Datamaran has developed a materiality analysis tool to help companies to identify material ESG risks and opportunities

    • Templates:

      • Environment Impact Score

  • Case Studies

  • Further Readings