Decarbonization of AI

As per the UN environment programme (UNEP), the world today produces enough food to nourish every child, woman, and man on the planet.? It states that global hunger is not about a lack of food. Food loss is one of the root?causes of hunger?worldwide.? ?In its Food Waste Index Report 2024, UNEP states that ?one-fifth of food?produced for human consumption is lost or wasted globally. It amounts to?a staggering one billion meals a day!? Food loss and waste generates?up to 10 per cent of global greenhouse gas emissions.

It is estimated that around 65% of data generated and stored is only used once – or not at all! This is “dark data”.? Gartner defines?dark data?as the information assets organizations collect, process and store during regular business activities, but generally fail to use for other purposes.? Storing and securing the dark data too requires electricity, space on servers, incurs more expense (and sometimes greater risk) than value.? The World Economic Forum estimates that companies generate 1.3 trillion gigabytes of dark data every day.? Storing that data for a year emits as much CO2 as 3 million flights from London to New York.

As with food and data, when it comes to Artificial Intelligence, we must show our good judgement in how we use AI.? ?

While AI needs data and lots of it, it also requires lots of energy and water.? It is a technology that is driving up energy consumption.? ??A New Yorker report states that the popular chatbot ChatGPT uses a staggering half a million kilowatt-hours daily to handle its 200 million user requests. This translates to?over 17,000 times more electricity than the average US household uses in a day.?? According to the International Energy Agency, a single ChatGPT query requires 2.9 watt-hours of electricity, compared with 0.3 watt-hours for a Google search.

In its “2024 Environmental Sustainability Report”, Microsoft states that “New technologies, including generative AI, hold promise for new innovations that can help address the climate crisis. At the same time, the infrastructure and electricity needed for these technologies create new challenges for meeting sustainability commitments across the tech sector.”

So, how worried should we be about AI’s energy consumption and demands?

The rising electricity demand from AI, in some ways should not be any different from the rising demand on energy from electric vehicles, growth in manufacturing, etc. ?How well we meet that demand is what perhaps really matters.?It is time for decarbonization - the responsible use of knowledge and data to minimize carbon emissions from AI development and adoption.? If rising electricity demand is met through renewable energy and other low-carbon power sources, and we push AI to get more efficient, doing more with less energy, then we can continue to slowly clean up the energy grid, even as AI continues to expand its reach in our lives.?

Given the extent of impact that AI has across the entire spectrum of industries, social activities and the environment, governments are beginning to introduce regulations that lay down a legal framework for the development, placing on the market, putting into service and the use of artificial intelligence systems in their jurisdictions.? One such is the European Union’s Artificial Intelligence Act. Amongst other things, this AI Act?specifies that high-risk AI systems should have logging capabilities to record energy consumption and measure or calculate resource use, and providers should assess environmental impact throughout the AI system's lifecycle.

There are no systematic, detailed, and principled ways yet to determine how much energy an AI model is consuming. It seems the literature that does exist on AI energy consumption are either anecdotal or based on very basic calculations.? This is made amply clear in a University of Michigan news item titled “Power-hungry AI: Researchers evaluate energy consumption across models”, and attributed to Mosharaf Chowdhury, Professor of Computer Science and Engineering at U-M.

How would one manage or improve what cannot be measured?

If the Big tech giants are to be believed – they claim that they are currently investing in developing methodologies to quantify the energy use and carbon impact of AI, while working on ways to make large AI systems more efficient, in both training and application. Significant strides need to be made in the discipline of Sustainable Software Engineering too.?

Until all of these are straightened out, environment and sustainability will take a back seat.??