GEN AI usage Impact on Environment

AI Generated Images - The Hidden Environment Cost.

By Adwait Aphale

AI technology is transforming industries and revolutionizing the way we work, create, and interact. Personally, I’m an avid user of AI for a variety of tasks. But recently, I started thinking about something I hadn’t considered much before—the environmental cost of using AI, particularly when it comes to generating images.

With the launch of AI models that can create stunning visuals, including Ghibli-style images, there has been an explosion in their use online. While this technology is groundbreaking, it also comes with significant environmental impacts that are often overlooked.

The Hidden Energy and Cooling Demands of AI

The powerful GPUs that drive AI image generation require vast amounts of energy to function. As AI use increases, so does the demand for computational power, and this can lead to environmental strain on data centers.

AI technology is transforming industries and revolutionizing the way we work, create, and interact. Personally, I’m an avid user of AI for a variety of tasks. But recently, I started thinking about something I hadn’t considered much before—the environmental cost of using AI, particularly when it comes to generating images.

With the launch of AI models that can create stunning visuals, including Ghibli-style images, there has been an explosion in their use online. While this technology is groundbreaking, it also comes with significant environmental impacts that are often overlooked.

The Hidden Energy and Cooling Demands of AI

The powerful GPUs that drive AI image generation require vast amounts of energy to function. As AI use increases, so does the demand for computational power, and this can lead to environmental strain on data centers.

Energy Consumption: How Much Power Does AI Use?

To illustrate the magnitude of energy consumption, let’s look at specific figures:

• GPT-3 Training: According to a study from University of Massachusetts Amherst, training a large-scale AI model like GPT-3 (which has 175 billion parameters) requires approximately 256,000 kWh of electricity. To put that in perspective, 256,000 kWh is enough to power 24 U.S. homes for a year.
• Energy for AI Image Generation: For generating images, a model like DALL·E 2 or similar image generation systems might consume around 200-300 watts of power per inference (the time it takes to generate an image). Over many users generating thousands or millions of images, the energy consumption scales significantly. The total electricity used across large datasets and simultaneous requests can be equivalent to the energy consumption of a small city.

Water Usage in AI Data Centers

Water usage in data centers, while often overlooked, is another significant concern. Cooling systems, particularly those using evaporative cooling (which is common in AI data centers), can use substantial amounts of water.

• Water Consumption: A study by Stanford University in 2020 found that data centers can consume between 10,000 to 50,000 liters of water per day depending on their size and the type of cooling used. For an AI model generating images or running queries like ChatGPT, this could translate to 5 to 50 liters of water per image generated, depending on the scale of the data center and the cooling methods.

While not all of this water is directly drinkable, it still represents a substantial environmental cost, especially in areas where water resources are scarce.

Carbon Footprint of AI

The carbon emissions associated with running AI models are also considerable, particularly if the energy used comes from non-renewable sources.

• Carbon Emissions: Research from OpenAI and Microsoft shows that training and running AI models can result in significant carbon emissions. For example, training GPT-3 (the large language model behind ChatGPT's) has been estimated to produce up to 284 metric tons of CO2 emissions. That’s the equivalent of the annual carbon footprint of approximately 35 average cars.

The carbon emissions depend heavily on the energy sources powering the data centers. Data centers powered by renewable energy sources, like wind or solar, have a much smaller carbon footprint compared to those relying on fossil fuels, like coal.

What Can We Do About It?

While it’s impossible to entirely eliminate the environmental costs of AI, there are ways to mitigate its impact:

• Opt for more efficient models: AI models are evolving to become more energy-efficient. Newer architectures aim to reduce the computational resources needed to train and run models, reducing energy consumption.
• Sustainable energy: Supporting data centers that use renewable energy can significantly lower the environmental footprint of AI. Many large companies, like Google and Microsoft, have pledged to power their data centers with renewable energy, but there's still work to be done in this area.
• Use AI responsibly: Before using AI, ask yourself: Is this task necessary? Could I achieve the same result with fewer resources? Being mindful of the frequency and necessity of AI usage can help reduce the overall environmental impact.

Conclusion: Responsibility in AI Usage

AI is an incredibly powerful tool, but it’s important that we, as users and developers, recognize its environmental cost. With growing usage comes increased energy consumption, water use, and carbon emissions, which all have real-world consequences. Let’s be proactive in supporting more sustainable practices in AI development, from using energy-efficient models to supporting renewable energy-powered data centers.

By being mindful of these factors and using AI responsibly, we can help balance the incredible benefits of AI with a commitment to sustainability.