The Environmental Impact of AI and Generative AI

The environmental impact of AI, particularly generative AI, has become a hot topic for debate. Two opposing views dominate the discussion: one celebrates AI’s potential in solving environmental challenges, while the other highlights its considerable environmental costs. As companies explore the integration of AI and Generative AI into their operations, it’s crucial to navigate the balance between innovation and sustainability.

AI as an Enabler of Environmental Progress

Proponents of generative AI argue that the technology can significantly advance environmental goals. They point to AI’s ability to analyse complex climate data, offering valuable insights that can aid in efforts like achieving zero waste, reducing carbon emissions, and advancing decarbonisation strategies.

For instance, Google has introduced AI-powered tools designed to help policymakers reduce emissions from tailpipes and better prepare for climate change-induced disasters such as floods and wildfires. Additionally, AI can improve climate models, narrow existing uncertainties, and make more accurate predictions, which is crucial in adapting to climate change.

Leading institutions are also embracing AI for environmental progress. For instance, Columbia University’s Learning the Earth with Artificial Intelligence and Physics (LEAP) initiative is a pioneering effort that aims to create next-generation AI-based climate models. This initiative not only advances AI research but also trains students in sustainable AI technologies, ensuring a future workforce that is equipped to tackle environmental challenges.

Furthermore, AI has the potential to drive efficiency in industries such as energy. It can design lighter materials for wind turbines and aircraft, enabling energy savings. AI can optimize energy use across multiple renewable energy sources and improve efficiency in areas such as smart grids, power plants, supply chains, and manufacturing.

AI as an enabler of ESG

• Google even?showed off new AI-powered tools for policymakers to cut down tailpipe emissions?and prepare communities for climate change-related disasters like floods and wildfires.

• AI can analyse the many complex and evolving variables of the climate system to improve climate models, narrow the uncertainties that still exist, and make better predictions.

• Columbia University's new centre,?Learning the Earth with Artificial Intelligence and Physics (LEAP), will develop next-generation AI-based?climate models, and train students?in the field.

The Environmental Costs of AI

On the flip side, critics emphasize the environmental toll that AI itself incurs. Generative AI systems require vast amounts of energy and water, both of which can exacerbate global resource shortages. In fact, the energy consumption of large AI models is projected to grow so significantly that, within a few years, they could match the energy demands of entire countries.

AI has already contributed to a rise in greenhouse gas emissions. Microsoft, for example, reported a nearly 30% increase in its emissions since 2020, largely driven by its AI investments. A recent analysis found that adding generative AI to Google Search could increase its energy consumption by more than ten times.

As AI technology continues to scale, the demand for resources like water is expected to grow exponentially. Generative AI systems need enormous amounts of fresh water to cool their processors and generate electricity, putting tremendous pressure on water supplies.

Moreover, AI’s energy needs extend beyond the computing power required for running models. Building and operating data centers, the infrastructure necessary for AI, is a major contributor to energy consumption. Even the manufacture of processors that power AI systems carries an environmental cost—one that has yet to be fully quantified.

AI as harmful to ESG

• A study by Dutch researcher Alex de Vries estimated that the "worst-case scenario" suggests that Google's AI systems could eventually consume as much electricity as the country of Ireland each year, assuming a full-scale adoption of AI in their current hardware and software.
• Within years, large AI systems are likely to need as much energy as entire nations.

• AI is accelerating global freshwater scarcity

• Greenhouse gas emissions are soaring thanks to AI

• Microsoft earlier this year said its emissions had risen almost a third since 2020.

• Adding?generative AI to?Google Search increases its?energy use more than tenfold, according to a new analysis.
• In West Des Moines, Iowa, a giant data centre cluster serves OpenAI's most advanced model, GPT-4. A lawsuit by residents revealed that in July 2022, the month before OpenAI finished training the model, the cluster used about 6% of the district's water.

The cost of Generative AI.

• Electricity: The?International Energy Agency estimates that a Google search query requires 0.3 watt-hours of electricity on average. In comparison, a ChatGPT request typically consumes about 2.9 watt-hours. ?
• Water: Shaolei Ren, a researcher at the University of California, estimates ChatGPT gulps up?500 millilitres?of water (close to what's in a 16-ounce water bottle) every time you ask it a series of between 5 to 50 prompts or questions. ?

The Cost of Generative AI

The financial costs of integrating generative AI are another key concern. Researchers have found that the energy consumption and emissions tied to AI systems are significant, but difficult to predict.

Renowned researcher Alex de Vries, who has previously raised awareness of pollution stemming from crypto mining with his website?Digiconomist, says it's still too early to calculate how much planet-heating pollution might be associated with new tools like ChatGPT and similar AI-driven apps. But he says it's worth paying attention now to avoid runaway emissions.

A single AI model can consume thousands of megawatt hours of electricity and produce carbon emissions equivalent to those of hundreds of households annually. For businesses considering the adoption of AI tools, the environmental and financial implications are crucial.

Training AI models, such as large language models, is resource-intensive but the actual cost of a generative AI system will depend on several factors, including:

1. Model Size and Complexity: Generative AI models like Stable Diffusion XL, which produce images, consume considerably more energy than smaller models tailored for specific tasks. Even smaller models for text generation, which are less energy-intensive, still have measurable environmental impacts. For example, using a generative model to classify movie reviews according to whether they are positive or negative consumes around 30 times more energy than using a small language model fine-tuned model created specifically for that task.
2. Location of Data Centers: Energy sourcing and efficiency vary significantly based on location and the energy efficiency of the data centre. For instance, Google’s data center in Finland runs on 97% carbon-free energy, while data centers in Asia may only achieve 4–18% carbon-free energy, directly affecting the environmental costs of AI operations.
3. Task Type: For instance, generating 1,000 images with a powerful AI model such as Stable Diffusion XL can result in carbon emissions equivalent to driving 4.1 miles in an average gasoline-powered car. In contrast, a smaller AI model for text generation may have a carbon footprint comparable to just 0.0006 miles of driving. The number of queries processed and the energy required for specific tasks like video generation—up to 300 times more energy-intensive than text generation—further impact the total energy usage. As companies integrate AI more deeply across applications, the cumulative environmental impact grows.
4. Query Volume and Frequency: The rapid adoption of generative AI has led to models being used millions of times daily, with increasing costs and environmental effects as AI becomes embedded in more business operations. However, If companies employ innovations that will help save money (e.g., efficient AI model architectures, optimization algorithms to accelerate AI training and inference, techniques like weight pruning and quantization) and reduce model sizes then the cost of queries will be lower.
5. The number of AI infused applications a business owns and uses: The number of times GenAI is used is based on how many applications are enabled with Generative AI. The generative AI boom has led big tech companies to integrate powerful AI models into many different products, from email to word processing. These generative AI models are now used millions, if not billions, of times every single day, with costs escalating the more GenAI is inculcated into the fabric of each business.

Even without considering the environmental toll of chip manufacturing and supply chains, AI models' training processes require massive amounts of energy. Beyond training, AI’s carbon footprint extends to regular usage. Each prompt and query, particularly those that generate text, images, or videos, adds to the energy burden, with video generation being especially resource-intensive.

Strategies for Managing AI’s Environmental Impact

To address AI’s environmental costs while still benefiting from its capabilities, organizations need to take concrete steps toward sustainable AI usage. Here are a few strategies for balancing innovation with environmental responsibility:

1. Establishing a Baseline: Companies should begin by measuring their current emissions, leveraging life-cycle assessments to calculate the environmental impact of prospective AI applications. This step can provide a foundation for managing emissions within organizational budgets and informing sustainability efforts.
2. Incorporating Emissions Accountability: As organizations assess the practicality of new AI models, they should factor in the emissions associated with these technologies. Decision-makers should align AI applications with sustainability objectives to ensure environmental responsibility.
3. Collaborating with Suppliers: Seventy per cent of a typical company’s emissions often comes from its suppliers. Therefore, the fastest way to decarbonize IT is to engage suppliers and support their decarbonization journeys. Ask all your IT suppliers to help, and press Microsoft to assess the ESG impact of the program and models it is implementing.
4. Educating and Upskilling Staff: Building awareness among employees around eco-design and efficient AI utilization can help optimize AI applications. Training employees to use appropriately sized models and to avoid unnecessary energy-intensive tasks, such as video generation, is a practical way to minimize emissions.
5. Optimizing Model Efficiency: Companies can significantly reduce emissions by fine-tuning AI models rather than relying solely on multipurpose models. Techniques such as model quantization and prompt engineering can reduce energy usage by up to 70%.
6. Focusing on Domain-Specific Models: Adopting smaller, more targeted AI models for specific applications (i.e., small language models), rather than deploying large language models, can reduce cost, energy and environmental impact.
7. Working with Industry Leaders: Collaborating with companies experienced in sustainable AI practices can provide insights into managing AI’s environmental impact. Partnering with established providers can ensure that sustainability considerations are part of AI deployment.

• Some companies, such as Microsoft, are seeking somewhat novel nuclear energy sources to meet energy demand and move toward their ESG goals. For example, Constellation Energy plans to restart the?Three Mile Island nuclear plant and will sell the power to Microsoft under a 20-year supply deal, demonstrating the immense energy needs of the tech sector as they build out data centres to support artificial intelligence and their desire to meet ESG targets.

Moving Toward Sustainable AI

The question remains open as to whether AI's potential to aid decarbonization and adaptation outweighs the enormous amounts of energy it consumes. Tech companies such as Google, OpenAI, Meta, and Microsoft are all struggling with the environmental costs of AI, including both water and energy consumption because running supercomputers, model buildings, and data centres take up energy.?

So, as AI becomes a central tool for business strategy, companies must weigh its environmental impact against its operational benefits. By embracing sustainability-focused strategies and collaborating with stakeholders, businesses can help minimize the environmental impact of AI while leveraging its power to drive progress in both technology and sustainability. Through careful planning, measurement, and innovation, a balance that aligns with business goals and environmental responsibilities might be achieved.

End.

You might also like to read about:

• 8 GenAI Risks you Should Not Ignore
• Digital Copilots: How AI is Unlocking $50M in Workplace Productivity
• Kieran Gilmurray Chief AI Innovator at the thettg.com
• Need our support and guidance to understand how you might you generative AI and automation in your workplace? Then Find me on social media LinkedIn | Kieran Gilmurray | Twitter | YouTube | Spotify | Apple Podcasts visit website: Https://KieranGilmurray.com or Https://thettg.com.

Photo by Sam Jotham Sutharson on Unsplash