The Environmental Cost of Generative AI

Generative AI is taking over the world, but an unseen storm may be brewing.  While this powerful technology has huge potential for positive change, its environmental impact cannot be disregarded.

Today’s article explores the less-discussed environmental impact of deploying generative AI. Here are some aspects to consider;

Energy Consumed from Training

Training these AI models requires massive amounts of computational resources. A 2024 study published in Nature projected that training a single large language model can produce up to 300 tons of carbon dioxide, which is equivalent to the lifetime emissions of several dozen cars. This energy need is driven by data centers, which are massive warehouses loaded with computer servers that are continually functioning and these servers require the same amount of energy to cool them. (https://www.nature.com/articles/d41586-024-00478-x)

Energy Consumed by Using

The problem doesn't end with training. According to a recent study by Nvidia, a leading manufacturer of AI processing chips, using generative AI models can account for 80-90% of total energy consumption because, while each individual query has relatively little environmental impact, the number of queries made around the world each day can reach hundreds of millions, if not billions.  (https://charitydigital.org.uk/topics/the-environmental-cost-of-generative-ai-11196#:~:text=Far%20more%20significantly%2C%20actually,millions%2C%20or%20perhaps%20billions)

Energy Consumed from Storage

The environmental impact goes beyond the initial training phase.  The massive volumes of data generated by these models must also be stored, which increases energy usage in data centers. This continual energy consumption increases the overall environmental impact.

So What?

While generative AI has the potential to change a variety of industries, its specific impact on energy consumption will most likely be determined by how it is deployed and used. Some measures to consider include;

Investing in Green Energy: Powering data centers with renewable energy sources such as solar or wind can considerably lower the carbon footprint of generative AI.

Optimizing model efficiency: Researchers are continually working on more efficient AI models that demand less computational power. This can considerably reduce the energy requirements of generative AI.

Considering the Actual Cost: When deploying generative AI technologies, we must account for both the environmental impact and the economic rewards. A comprehensive cost-benefit and Life Cycle Analysis can help us make informed decisions.

You can reduce your environmental impact of AI by;

• Staying informed about your AI carbon foot print; Do you know where the generative AI systems that you use operate from? What source of energy does these data centers use?

• Do you need to invest in developing that new generative AI platform, or can you leverage the technology of existing Open sources for developing your tools?

Do you really need generative AI?

Sustainably yours
Chiny