Green AI

When you think about Artificial Intelligence, the first thing that comes to mind probably isn't carbon footprint. But perhaps it should be. The computations required for deep learning research have been doubling every few months, resulting in an estimated 300,000x increase from 2012 to 2018, and these computations have a surprisingly large carbon footprint. If we don’t want AI research to become as electricity hungry as bitcoin mining, we need to do things differently. The chart below shows the total amount of compute used to train well-known deep learning models.

In addition to increased energy inefficiency, the barriers to participation for those who can't afford the rising costs of developing, training, and running models are significant. As covered by the NYTimes, if trends continue, artificial intelligence research will be a field of haves and have-nots.

At the Allen Institute for Artificial Intelligence, we recently published a position paper, Green AI, advocating a practical solution- make efficiency an evaluation criterion for research alongside accuracy and related measures. Despite the clear benefits of improving model performance in AI, the focus on this single metric ignores the economic, environmental, or social cost of reaching the reported performance. The figure below shows the proportion of papers that target accuracy, efficiency, both or other from a sample of 60 papers from top AI conferences.

In addition to considering additional criteria, we propose reporting the financial cost or “price tag” of developing, training, and running models to provide baselines for the investigation of increasingly efficient methods.

The term Green AI refers to AI research that yields novel results without increasing computational cost and ideally reducing it. Red AI refers to AI research that seeks to obtain state-of-the-art results in accuracy (or related measures) through the use of massive computational power—essentially “buying” stronger results. While Red AI has been yielding valuable contributions to the field of AI, it’s been overly dominant. Our goal in championing Green AI is to make AI both greener and more inclusive—enabling any inspired undergraduate with a laptop to write high-quality research papers.