# Green AI

**Authors:** Roy Schwartz, Jesse Dodge, Noah A. Smith, Oren Etzioni  
**arXiv:** [1907.10597](https://arxiv.org/abs/1907.10597)  
**Date:** 2019-07-22  
**Summary page:** https://www.summarizepaper.com/en/paper/1907.10597v1/

## Key points

- Authors address the issue of rapidly increasing computational requirements in deep learning research
- Computation requirements have been doubling every few months, leading to a 300,000x increase from 2012 to 2018
- Environmental concerns and financial barriers for individuals in academia are highlighted
- Proposal to make efficiency a key evaluation criterion alongside traditional metrics like accuracy
- Suggestion to incorporate the financial cost of developing, training, and running models as a baseline for exploring more efficient methods
- Goal is to make AI more sustainable and inclusive by enabling anyone with basic resources to engage in high-quality research
- "Green AI" concept gaining traction at institutions like the Allen Institute for AI

## Summary

In their paper "Green AI," authors Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni address the pressing issue of the rapidly increasing computational requirements in deep learning research. They highlight how these computations have been doubling every few months, leading to a staggering 300,000x increase from 2012 to 2018. This exponential growth not only poses environmental concerns due to the large carbon footprint associated with such computations but also creates financial barriers for individuals in academia, particularly those from emerging economies. The authors propose a practical solution: making efficiency a key evaluation criterion for research alongside traditional metrics like accuracy. Additionally, they suggest incorporating the financial cost or "price tag" of developing, training, and running models as a baseline for exploring more efficient methods. The ultimate goal of this approach is to make AI more sustainable and inclusive by enabling anyone with access to basic resources—such as an undergraduate student with a laptop—to engage in high-quality research. The concept of "Green AI" is gaining traction at institutions like the Allen Institute for AI, signaling a shift towards more environmentally conscious and accessible practices within the field of artificial intelligence.