The paper "Green AI" by Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni from the Allen Institute for AI, Carnegie Mellon University, and the University of Washington, advocates for a practical solution to the growing computational and financial costs of deep learning research. The authors argue that the increasing computational demands of deep learning, which have doubled every few months since 2012, have led to a significant carbon footprint and financial barriers, particularly for researchers in emerging economies. They propose making efficiency a primary evaluation criterion alongside accuracy, and suggest reporting the financial cost of developing, training, and running models to promote more efficient methods. The paper highlights the need for Green AI, which is both environmentally friendly and inclusive, enabling inspired undergraduates to conduct high-quality research. The authors also discuss the concept of Red AI, which refers to AI research that focuses on achieving state-of-the-art results through massive computational power, often at the expense of efficiency and cost. They propose several measures of efficiency, including the total number of floating-point operations (FPO), and advocate for their widespread adoption in AI research. The paper concludes by emphasizing the importance of Green AI in reducing computational costs, improving inclusivity, and potentially leading to more cognitively plausible models.The paper "Green AI" by Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni from the Allen Institute for AI, Carnegie Mellon University, and the University of Washington, advocates for a practical solution to the growing computational and financial costs of deep learning research. The authors argue that the increasing computational demands of deep learning, which have doubled every few months since 2012, have led to a significant carbon footprint and financial barriers, particularly for researchers in emerging economies. They propose making efficiency a primary evaluation criterion alongside accuracy, and suggest reporting the financial cost of developing, training, and running models to promote more efficient methods. The paper highlights the need for Green AI, which is both environmentally friendly and inclusive, enabling inspired undergraduates to conduct high-quality research. The authors also discuss the concept of Red AI, which refers to AI research that focuses on achieving state-of-the-art results through massive computational power, often at the expense of efficiency and cost. They propose several measures of efficiency, including the total number of floating-point operations (FPO), and advocate for their widespread adoption in AI research. The paper concludes by emphasizing the importance of Green AI in reducing computational costs, improving inclusivity, and potentially leading to more cognitively plausible models.