How to reduce AI’s harmful effects on the environment

LINK: https://bit.ly/46NEWE8

Equities.com spoke with Bill Wong, AI research fellow and lead AI researcher at Info-Tech Research Group. He is an expert on artificial intelligence strategies with a long career consulting, developing and writing about anything related to AI. He has held executive positions at a number of technology companies across the world, and is often asked to speak about AI at conferences and trade shows.

We wanted to know what Wong thought about the effect of AI on the environment, whether it is helpful or hurtful, how much power AI uses, if the government should get involved, and what might happen if technology companies and consumers ignore the effects of climate change.

Equities.com: Is AI a friend or enemy of the environment?

Bill Wong: There are both positive and negative impacts. On the negative side, it uses an extraordinary amount of compute resources, and that hurts the environment. It takes a lot of power, a lot of water and produces carbon emissions to develop these AI models. Power consumption is especially pronounced with the big technology vendors building large language or multimodal models.

EQ: What’s the positive side?

BW: On the plus side, AI has been trained to find best practices for lowering carbon emissions. You can run simulations and have interactive visualizations of the impact of AI on climate change or various weather anomalies. Many AI research laboratories are funded by the Department of Energy in the United States and are currently performing research on mitigating the impacts on climate change.

The EPA uses AI to help enforce environmental regulations. Of course, NASA uses AI to evaluate hurricane intensity and help to mitigate risk. There’s also a program from the Department of Energy called “Wildfire” that uses satellite imagery to assess the risk of wildfires.

EQ: How much computing power do the tech companies use?

BW: The big tech vendors don’t disclose how much resources they use. There have been estimates made, but all we have are ballpark figures. I think some people would be surprised at how much of our resources are needed to do even a simple search.

For example, one of Google’s initiatives was to build a data center in Chile. Everyone in the country was excited about it. However, when they did an analysis on the climate impact, the government was shocked at how much water this data center would require. A Google spokesperson said they submitted a change to the original design to have the center air-cooled instead.

EQ: So the tech companies who use AI are not transparent?

BW: They will argue that they are transparent. They will share some information but not all. What is often lacking is the hardware configuration used to produce the latest capabilities their AI models can deliver.

EQ: Why don’t more people discuss AI power consumption in the States?

BW: In the G7 countries, where water is plentiful, the impact of AI is kind of hidden, unless you are the data center manager who can see the monthly bill for power and water. In countries that are not as well off, the government can see the impact directly.

EQ: What other resources does AI use?

BW: It’s not only the running of the machines but it’s also chip manufacturing that plays a role. The chip manufacturing process requires a ton of resources that leaves a huge carbon footprint, and the data center makes its own carbon footprint.

EQ: Is there a solution to the amount of power AI uses?

BW: At this time, there still seems to be a notion that “bigger is better for AI model performance,” but it is not sustainable, in my opinion. I am joined by several researchers who believe there will be a number of breakthroughs where we may use AI models that don’t consume as much resources as they do today. We can’t continue on this current trajectory indefinitely.

EQ: What can the companies do?

BW: Today, companies planning to use AI can architect their solution using an agent-based approach. Instead of sending all the prompts or queries to ChatGPT or GPT-4, which is a common scenario, you could implement a solution that uses AI models that are more fit-for-purpose. Instead of ChatGPT receiving every request, smaller models can be used to address prompts and queries that can be processed more effectively and efficiently.

EQ: Are companies starting to think about the environmental impact of AI?

BW: We engage with a variety of executives and boards regarding the development of their AI strategy. Often, their primary focus is leveraging AI to improve operational excellence, customer engagement, sales or profit contribution, and their risk mitigation programs. We are also seeing new value-based drivers for their AI investments emerge, such as improving their environment, social and governance programs.

Organizations want to be more energy efficient — it reduces costs and waste. ESG programs can be a catalyst for transforming an organization’s AI strategy for improving business outcomes in a responsible and sustainable manner.

EQ: Should the government get involved?

BW: If the environmental requirements of these AI models continue and innovations do not get introduced to curb the energy consumption requirements, then the government may have to step in. I know the government doesn’t want this responsibility because they don’t want to stifle innovation. I understand that. But that has to be balanced with the health of the planet. Nobody wants to see a scenario where the government has to take drastic measures.

EQ: What happens if people ignore the climate change research?

BW: The current rash of wildfires, flooding and other extreme environmental events will
continue, but at an accelerated rate. If we continue to consume all this energy and resources, and
don’t pay attention to the possible impact on the environment, things will get worse.