AI | March 7, 2024

Image: Freepik

AIs Energy Consumption is on an Unsustainable Trajectory and Demands Urgent Efficiency Measures

As artificial intelligence (AI) and machine learning (ML) technologies evolve, their energy consumption has skyrocketed, posing significant sustainability challenges. This trend, driven by the development of larger models and the quest for higher accuracy, raises concerns about the long-term viability of AI advancements.  In the article from Peterson Institute for International Economics Industry leaders are sounding the alarm, urging a shift towards more energy-efficient practices to ensure the future of AI aligns with global energy capabilities and environmental goals.

• Machine learning's energy consumption is on an unsustainable trajectory, threatening to outpace global energy production. The demand for larger models and extensive training sets has led to an exponential increase in power usage, primarily in data centers for both training and inference phases.  Figures from AMD's CTO Mark Papermaster highlight the stark reality of ML systems' energy consumption compared to the world's energy output. The tech industry, historically driven by efficiency innovations like Moore's Law, now faces a period of "anti-efficiency," focusing on performance at the cost of increased energy use.

See:  Sustainability: A Must for Fintech Growth

• The pursuit of higher accuracy in AI applications, such as voice and speech recognition, has led companies to prioritize results over energy consumption. This focus on profitability, however, overlooks the potential long-term impacts on energy resources and environmental sustainability.
• AI's reliance on data centers significantly contributes to its carbon footprint. These centers not only consume vast amounts of electricity but also require continuous cooling through air conditioning, which further increases energy use. As AI becomes more widespread, the carbon emissions from data centers are expected to rise, exacerbating the environmental impact. There is a growing commercial pressure from consumers to reduce the carbon footprint of AI technologies. Companies striving for carbon-neutral solutions may find a competitive edge, as consumers increasingly favour environmentally sustainable practices.
• The process of training LLMs, such as GPT-3, is extremely energy-intensive. A recent study by Cornell University cited in the article found that training such models can consume electricity equivalent to 500 metric tons of carbon, comparable to a coal-fueled power plant running for nearly half a day. Given that these models require frequent retraining to stay updated, the cumulative energy consumption and carbon emissions are substantial.  While training AI models is known to be energy-intensive, the inference process (responding to queries) may consume even more energy.  This is alarming since there are not only more users interacting with LLMs but they are also increasing their reliance and usage.

See:  Canadian Banks Face Scrutiny Over Sustainability Claims

• Beyond cloud data centers, the proliferation of smart edge devices contributes significantly to the overall energy consumption of AI technologies. These devices, integral to the Internet of Things (IoT), are expected to use more power than the world generates, highlighting the need for energy-efficient solutions across all facets of AI deployment.
• There is a lack of transparency from AI companies regarding the environmental costs of developing and operating their systems. This opacity makes it difficult to assess the full extent of AI's carbon footprint and to implement effective regulations to mitigate its environmental impact.
• A coalition of environmental groups warns that the growing use of AI will lead to increased energy use from data centers. The coalition recently published a report AI Threats to Climate Change where this rise in AI-driven energy demand is expected to result in an 80% increase in planet-heating emissions, even with improvements in energy efficiency at these centers.

Ways to Reduce AIs Carbon Footprint (according to Google Researchers)

Recent Google research on reducing AI's carbon footprint suggests four key practices aimed at minimizing the environmental impact of AI systems:

• Reduce the number of parameters (read accuracy), these models require less computational power for both training and inference, leading to lower energy consumption and, consequently, a reduced carbon footprint.
• Use specialized processors designed specifically for machine learning tasks are more efficient than general-purpose processors. These specialized processors can handle AI workloads more effectively, reducing the amount of energy required for training and running AI models.

See:  How TinyML Is Unleashing AI Power in Everyday Devices

• Use cloud-based data centers which are generally more energy-efficient than local data centers. They benefit from economies of scale and can implement advanced cooling and energy management technologies more effectively. Additionally, cloud providers often invest in renewable energy sources, further reducing the carbon footprint of AI operations hosted in the cloud.
• Optimize cloud infrastructure to use data center locations based on the availability of cleaner energy sources. By choosing locations where renewable energy is readily available and affordable, AI companies can significantly reduce the carbon emissions associated with the power consumption of their data centers.

The Outlook for Sustainable AI Development

AI's vast energy use, driven by the development of larger models and the quest for higher accuracy, is on a collision course with the planet's environmental and energy sustainability goals. The tech industry, once celebrated for efficiency-driven innovations, now faces the challenge of reversing the "anti-efficiency" trend that prioritizes performance over environmental impact.

The environmental cost of AI's reliance on data centers, the intensive energy requirements for training large language models, and the increasing energy consumption for inference processes highlight the multifaceted nature of AI's carbon footprint. Moreover, the proliferation of smart edge devices threatens to exacerbate this issue, underscoring the need for comprehensive energy-efficient solutions across all facets of AI deployment.

See:  Bitcoin’s Energy Blueprint for the AI Revolution

Google's research points towards actionable strategies for reducing AI's environmental impact, including the adoption of sparse models, specialized processors, cloud-based data centers, and optimizing the location of these data centers to leverage cleaner energy sources. These recommendations offer a roadmap for the AI industry to mitigate its carbon emissions and align with global sustainability efforts.

The National Crowdfunding & Fintech Association (NCFA Canada) is a financial innovation ecosystem that provides education, market intelligence, industry stewardship, networking and funding opportunities and services to thousands of community members and works closely with industry, government, partners and affiliates to create a vibrant and innovative fintech and funding industry in Canada. Decentralized and distributed, NCFA is engaged with global stakeholders and helps incubate projects and investment in fintech, alternative finance, crowdfunding, peer-to-peer finance, payments, digital assets and tokens, artificial intelligence, blockchain, cryptocurrency, regtech, and insurtech sectors. Join Canada's Fintech & Funding Community today FREE! Or become a contributing member and get perks. For more information, please visit: www.ncfacanada.org

Buying a Canadian MSB and Reviewing FINTRAC RPAA and Financial Due Diligence

Why MSCI Is Removing Strategy From Its Indexes

How Is Crypto Market Cap Calculated? What a Change in the Top Coins Really Tells You

What Happens When AI Agents Start Trading Together?

Canada’s VC Totals Are Reliant On Mega Deals

Why Canada Is Making It Cheaper to Leave Your Bank

CSA Tokenization Report Shows Where Canada May Scale First

Creative Ways to Boost Team Morale

Webull Adds Free Data as Canadian Brokers Compete Beyond Fees

BBVA Deploys Canadian AI for Derivatives Pricing