Tag: Carbon

Categories
AI Governance Uncategorised

Scope 3 Calculator for Gen-AI usage

Following on from ESG Calculations for AI Usage, I decided to create the calculator in Excel. It’s been an interesting experience.

Part of the challenge is the lack of data, notably around efficiency gains, average carbon impact for journeys, and carbon per query when taking into account training emissions.

What’s the purpose?

There are two goals to the spreadsheet

  1. To provide input for Scope 3 emissions resulting from gen-AI
  2. To provide input for discussions on the relative environmental impact of using gen-AI for generating code

What is the calculator?

The Gen AI Carbon excel file has multiple tabs, focussed on two main sheets: Calculator and Coding

The Calculator is the overview bringing together all the calculations from the other sheets and providing input for scope 3 returns.

Coding is the sheet to handle specific use of gen-ai for code generation. This one is integrated to Ways of Working sheet, which allows for modelling remote, office and hybrid scenarios. These scenarios weren’t added to the main calculation since it’s expected that organisations are already calculating for those emissions.

Issues

However, there are some issues described in the intro sheet. Feel free to use it at your own risk, but DO NOT EXPECT to simply input your own figures. There’s a bit of fixing the calculations to make it work properly.

Why aren’t I providing this completely working?

I started this to answer a question regarding at what point does using Gen-AI for coding become environmentally irresponsible. I settled on carbon rather than water or other factors as in indicator. I figure out enough for me to understand that assuming productivity gains can be realised, then there’s a net benefit to using Gen-AI for coding. The minor errors left in the calculations were not of sufficient size to deflect that result. So I’m comfortable. However, I can see it also being of use to others.

Gen AI Carbon 0_01Download

Categories
AI

ESG Calculations for AI Usage

Following on from a number of conversations in the last couple of weeks on the subject of carbon reporting for the use of AI, specifically Foundation Models (including LLMs), I started some initial research.

I couldn’t find a decent calculator. I could find calculators for current usage, i.e. for prompts and replies. But they weren’t clear if they included embodied costs, such as training new models.

I’ve created this document to provide some foundation for those tasked with compiling Scope 3 figures.

Notes

  • It’s draft
  • There may be errors
  • Check the figures
  • Double-check the defaults
  • It doesn’t yet include using LLMs for code development/generation. That will come, but was an additional nuance.
  • It doesn’t include use of AI outside of Foundation Models, e.g. machine learning analysis of large volumes of data
  • There are other figures that may be of interest for ESG that are not carbon-related, e.g. water consumption, affect on local environment, etc

Aims

  • To provide a view of what may need to be considered when calculating Scope 3 figures for using Foundation Models
  • To act as a common reference across organisations
  • To prompt a conversation regarding what may need to be considered

Version

As the document evolves, I’ll post the current version on this page.

Foundation Model Scope 3 Emissions Calculation Advisory 0_01Download