Red vs. Green: Driving efficiency in artificial intelligence

The environmental impact of artificial intelligence (AI) has been a hot topic of late. While AI data can be used effectively to support the development of climate-friendly efficiencies and solutions, it also requires a high amount of energy consumption to accomplish this.

A blog post from OpenAI revealed that the amount of compute required for the largest AI training runs has increased by 300,000 times since 2012. And while that post didn’t calculate the carbon emissions of such training runs, others have done so.

According to a paper presented at the 57th Annual Meeting of the Association for Computational Linguistics (ACL) in July 2019, the average American is responsible for about 36,000 tons of CO2 emissions annually. Training and developing a single machine translation model that uses a technique called neural architecture search was responsible for an estimated 626,000 tons of CO2.

Unfortunately, these so-called “Red AI” projects may be even worse from an environmental perspective than reporting indicates. This is because a project’s total cost in time, energy, and money is often dramatically more than the cost of generating the final reported results. Because of this situation, an active movement toward more environmentally friendly “Green AI” practices is developing and growing.

 

The resource hit of Red AI

Since 2012, AI has reported remarkable progress on a broad range of capabilities, including object recognition, gaming, speech recognition, and machine translation. Much of this progress has been achieved by increasingly large and computationally intensive deep learning models. Trends show an overall increase of 300,000x—with training costs doubling every few months.

An important research paper has estimated the carbon footprint of several natural language processing (NLP) models and argues that this trend is both environmentally unfriendly and prohibitively expensive, raising barriers to participation in NLP research. We refer to this as Red AI.

The Red AI trend is driven by the strong focus of the AI community on obtaining state-of-the-art results, which typically report accuracy (or other similar measures) but omit any mention of cost or efficiency. Despite the clear benefits of improving model accuracy, the focus on this single metric ignores the economic, environmental, and social cost of attaining the reported results.

 

Is Red AI really that bad?

The short answer? No, not entirely. Many of today’s Red AI projects are pushing science forward in natural language processing, computer vision, and other important areas of AI. While their carbon costs may be significant today, the potential for positive societal impact is significant.

Consider the Human Genome Project (HGP), which took US $2.7 billion and 13 years to map the human genome. The HGP’s outcome was originally viewed as a mixed bag due to its cost and the dearth of immediate scientific breakthroughs. Now, however, we can map an individual’s genome in a few hours for around $100 using sequencing technology that relies on the main artifact of the HGP, the reference genome. While the HGP lacked efficiency, it nonetheless helped pave the way for personalized medicine.

Similarly, it’s critical to measure both the input and the output of Red AI projects. Many of the artifacts produced by Red AI experiments (i.e., image representations for object recognition, embedding words in natural language processing) are enabling rapid advances in a wide range of applications.

 

What is Green AI?

Green AI refers to a broader, long-standing interest in environmentally-friendly scientific research. Artificial Intelligence research can be computationally expensive in numerous ways; however, each provides opportunities for efficient improvements. For example, research papers could be required to plot accuracy as a function of computational cost and training set size, giving a baseline for more data-efficient research in the future. Reporting the computational price tag of finding, training, and running models is a key principle of a Green AI practice.

The conversation about Green AI began with a study from the Allen Institute for AI that argued for the prioritization of Green AI efforts that focus on the energy efficiency of AI systems. This study was motivated by the observation that many high-profile advances in AI have staggering carbon footprints.

 

The move towards Green AI

Regardless of its underlying scientific merits, Red AI isn’t sustainable in the long term, due to both environmental concerns and the barriers of entry that it introduces. As we’ve seen, the HGP did succeed in sequencing the human genome, but novel DNA sequencing technologies were required to drastically reduce costs and make genome sequencing broadly accessible. Moving forward, the AI community must work strategically and deliberately to reduce energy consumption when building deep learning models.

 

Green AI and business

Green AI can be a net-positive contributor to environmental sustainability in many industries:

  • Agriculture – AI can transform production by better monitoring and managing environmental conditions and crop yields. The technology can also help reduce the need for both fertilizer and water, all while improving crop yields.
  • Energy – AI can use deep predictive capabilities and intelligent grid systems to manage demand and supply of renewable energy. The cost and unnecessary carbon pollution generation can be slashed.
  • Transportation & logistics – AI can help reduce traffic congestion, improve the transport of cargo, and enable more autonomous driving capability.

 

Moving toward Green AI

Building the future of Green AI requires better collaboration, technology advances, and developing a better understanding of deep learning. Here are five ways we can help drive the development and adoption of Green AI as an industry:

  1. Replication and sharing of intermediate artifacts are crucial to increasing efficiency of AI development. Too often, AI research is published without code, and sometimes researchers find that they can’t reproduce results even with the code. Researchers can also face internal hurdles in making their work open source. These factors are significant drivers of Red AI today, as they force duplicated efforts and prevent efficient sharing. This situation is changing slowly, with conferences like NeurIPS now requiring reproducible code submissions with research papers.

 

  1. Hardware performance improvements offer better results for deep-learning tasks, and also increase performance-per-watt. The AI community’s demand for graphics processing units (GPUs) led to Google’s development of tensor processing units (TPUs) and pushed the entire chip market toward more specialized products. In the next few years, we’ll see NVIDIA, Intel, Samba Nova, Mythic, Graphcore, Cerebras, and other companies bring more focus to dedicated hardware for AI workloads.

 

  1. Deep learning expertise is a gap for researchers. We know that deep learning works, but we as a research community we still don’t fully understand how or why it works. Uncovering the underlying science behind deep learning, and formally characterizing its strengths and limitations, will help guide the development of more accurate and efficient models.

 

  1. Democratization of deep learning is an exciting area of research, and existing models are already accurate enough to be deployed in a wide range of applications. Nearly every industry and scientific domain can benefit from deep learning tools, and when people in many sectors are working on the technology, we’ll be more likely to see dramatic innovations in performance and energy efficiency.

 

  1. Partnership development is essential to the advancement of AI. Most of the world’s largest companies don’t have the talent to build AI efficiently, but their leaders realize that AI and deep learning will be key components of future products and services. Rather than go it alone, companies should look for partnerships with start-ups, incubators, and universities to jumpstart their AI strategies.

 

Moving forward

While it’s easy to look at a self-driving car whizzing down a road in Silicon Valley and think that we’ve reached a technological peak, it’s important to understand we’re still in the very early days of AI. The 2020s may see incredible advances in AI, but in terms of infrastructure and efficient use of energy we’re still in the pioneer age.

 

As AI research progresses, we must ensure that the best platforms, tools, and methodologies for building models are easy to access and reproducible. That will lead to continuous improvements in energy-efficient Green AI.

Similar Articles