As the power of sensors and processing rise and their costs drop, those of us who study the “greening” of information and communications technology (ICT) are looking ahead to the challenges posed by artificial intelligence (AI).
Even though we’re only in the early stages of implementing AI, we can foresee the potential it holds for greening many systems, including itself, as well as the challenges it poses for energy consumption and related environmental impacts.
The analogy to greening ICT is apt, though it doesn’t go far enough. Today’s ICT networks gather, transmit and process immense amounts of data and greening it is a priority due to the growth in global, digital activity and the unsustainable amount of energy this requires.
AI’s anticipated role in our world simply adds urgency to ICT-related energy and environmental concerns. One distinct contrast between the two topics is that AI, almost by definition, will require lower latency networks and higher processing power due to the need for real-time decision-making in many cases and the anticipated scale of its various roles.
AI may be defined in various ways, but for our purposes let’s say it refers to systems that are able to learn from their environment and adapt and improve functionalities over time. So AI must apply and evolve various algorithms and techniques to extract data from its environment, analyze it and apply its learnings to provide a better outcome than simpler, static processes or other optimization techniques. One of the promising applications of AI is to use it to green myriad systems, including itself.
Although we cannot yet grasp how pervasive AI will be, it is likely to be useful in automating aspects of our homes, businesses, transportation, manufacturing, innumerable industry verticals, even cities. We also do not yet fully understand how AI will evolve and scale and how that may affect how it behaves. The unknowns are significant and, because they affect the privacy and security of data, as well as the behavior of systems in which autonomous decision-making probably should remain beholden to humans, AI must be developed with our “eyes wide open.”
That said, AI’s potential is so valuable we must explore it and, as we do, we must apply great urgency to greening it in anticipation of its future roles in our world. I tend to be optimistic because we have already made strides in greening ICT and I can envision how our green ICT strategies might be applied to AI.
The Green ICT Challenge
Anyone old enough to witness the past couple of decades has seen the phenomenal growth in personal and enterprise computing, the Internet, mobile communications and all the productivity and information and entertainment applications these advances have wrought. These advances have become integral to our lives, yet they come at an immense cost that is not widely understood.
Relatively recent studies reveal that the ICT industry today generates approximately 2 percent of global CO2 emissions, on par with the global aviation industry, and it is forecast to grow at alarming speed and scale. ICT’s CO2 emissions will contribute about 4 percent of the world’s total in just five years, unless we change course. Use of the Internet, which relies on ICT, is growing by 30 percent to 40 percent per year. That equals perhaps 30 times its current traffic in 10 years and 1,000 times current traffic in just 20 years. Without concerted action, in a decade ICT could consume 60 percent of all global energy resources!
We cannot accurately foresee how AI will exacerbate this challenge, but it will. Let’s look at how we may tackle the challenge.
Greening by and for AI
Two areas we need to address generally are energy use by systems and the demands on those systems. AI can have an impact in both cases by identifying and measuring activity at the network level and at the connected device level by sifting vast amounts of data to extract patterns and trends. AI can then be applied to reconfigure the network in real-time to minimize power consumption.
Another AI application would be in renewable energy. Renewable energy will probably increasingly replace fossil fuels to power ICT and AI, and both the renewables and the AI and ICT will be increasingly distributed rather than centralized. Optimizing both the interlinked ICT networks as well as the renewable energy networks that support them is one possible role for AI.
Traditionally, ICT networks were designed with the assumption that data traffic is somewhat random, but ICT had a constantly reliable energy source. If we design ICT instead so that the energy source is renewable and, therefore, variable — in time and space — we have the flexibility to vastly decrease the ICT carbon footprint if we are able to deal efficiently with the variation in renewable energy. A system that matches both the randomly occurring tasks at hand and the available energy sources will be more optimal. Perhaps the data in need of processing can be sent to a locale where (clean) energy production at that moment is high; conversely, the available energy could be sent to where data needs processing.
The goal is to reduce energy use and the ICT carbon footprint and therefore its environmental impacts and this scenario would be an important AI application. A related metric is cost. Electricity may be cheaper in one area versus another.
You begin to see how managing a multi-dimensional strategy that includes time, space, energy demand and availability and cost would benefit from AI, particularly in a complex web of networks of networks.
Green ICT Progress
It’s difficult to foresee whether and how such an adaptive system will affect end users, though conceivably the quality of service might vary by cost. Still, we should all be aware that our systems, if not our habits, must change or they may collapse under their own weight.
I’m optimistic about meeting the challenges inherent in ICT and AI because we have made strides in greening ICT. A project dubbed GreenTouch, with which I was involved, utilized an international consortium of about 50 member organizations from industry and academia from 2010 to 2015 to improve the energy efficiency of ICT networks by a factor of 1,000 — and succeeded. Such efforts must continue to make ICT and AI sustainable.
In closing I should note related issues needing attention. As with all data-related matters, AI’s future will depend on public policies that enable the private, secure sharing of data from networks of networks. Global standards will need to be developed to support solutions that advance ICT and AI while minimizing their carbon footprints. In an increasingly digital world, this is one of the great challenges of the 21st century.