The Next Decade of Software Is About Climate - What Is the Role of ML?
summary
AI training and deployment have measurable climate costs and concrete levers to cut them: the Software Carbon Intensity spec defines SCI = (operational energy × grid carbon intensity + embodied hardware carbon) / functional unit; a convolutional case shows 11.5% saved by time‑shifting workloads; model scale jumps from 213M → ~200B parameters so pruning, distillation and quantization matter; use pruning, distillation, quantization or federated learning, time/region shifting, specialized TPUs, Azure ML energy metrics, edge inference and right‑sizing to reduce emissions.
AI-written summary · about
chapters (46)
- 0:00Introduction
- 1:32The Age of AI
- 2:24Bill Gates AMA
- 3:19Sara Bergman
- 5:05Agenda
- 6:13Why Software and Climate
- 7:49Machine Learning and Climate
- 9:35Model size is exploding
- 11:09State of AI
- 12:16Carbon Footprint
- 13:23Green Software
- 14:20Green Software Foundation
- 15:45Measuring Software
- 16:41Carbon Efficiency
- 21:21Quantification
- 21:46Quantification Methods
- 22:36Core Characteristics
- 24:25Net Zero
- 26:22SEI Score
- 27:23Machine Learning Climate
- 27:52Patterns
- 28:35Life Cycle
- 28:54Project Planning
- 29:06Carbon Aware Features
- 30:12Service Level Objectives
- 31:04Data Collection
- 31:44Data Hoarders
- 32:43Data Green
- 33:38The Next Phase
- 34:01Graph
- 34:57Training Models
- 36:37Optimizing Size
- 36:52Time Shifting
- 37:42Sustainable Regions
- 38:19Hardware
- 40:18Deployment Maintenance
- 41:11Variable Power
- 42:52Hosting Efficiency
- 44:00Serverless Architecture
- 44:30AI Models on the Edge
- 45:32Where to Start
- 46:48Summary
- 47:17Free course
- 47:46Questions
- 51:09What about software in general
- 53:11What is the Green Software Foundation doing
related talks
45mGreen AI: Making Machine Learning Environmentally Sustainable
01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory- 01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory
29mHow AI and Deep Learning could effort the Climate Change
38mGreen software: How can tech contribute to worldwide sustainability effort?
34mPath to production: An environmentally sustainable approach
44mThe Invisible Monster: The Carbon Footprint of AI
32mOptimizing AI/ML workloads for sustainability