AI Ethics and Green Energy

The AI tech boom has driven global energy use through the roof – but could it make up for the costs by transforming green energy production from a subsidized sector into a real powerhouse? Like any new technology development, there are a host of ethical questions that have yet to be answered. Navigating these questions is important for setting up infrastructure and standards that guide this new industry-shaker into a powerful tool for improving conditions on Earth instead of another way to exploit vulnerable people. 

Tracking the energy costs of AI is difficult – companies are reluctant to share concrete data on the true cost of training and running AI tools. Researchers and journalists have been cobbling together estimates from whatever data they can get their hands on, including the rising cost of server hardware, the number and size of newly built data centers, and making models of how much energy an AI calculation might cost. 

The Lawrence Berkeley National Laboratory released new projections last December that by 2028, AI could use the same amount of electricity as 22% of households in the United States. Right now, 4.4% of all energy in the United States is going to data centers, which power AI tools – and the demand is increasing so fast that many companies are turning their backs on green energy sources for the types of energy that are more readily available: power plants and burning fossil fuels. In turn, this is likely to drive up electricity costs for consumers – making the costs for the rise of AI come out of the pockets of everyday people.

It's not all bad news: AI has the potential to radically change green energy production to make it more competitive with fossil fuels. If implemented with care, AI can make green energy more efficient, discover new technologies to make improvements in development, and reduce the overall demand for electricity by optimizing its use. Each area of improvement has some potential ethical tripping hazards, which we’ll be talking about next.

Managing variability in clean energy production

Green energy sources like solar panels and wind power have a lot of variation in how much power they produce, depending on the weather and other factors. This makes it difficult to rely on them to cover a big increase in electricity demand – usually they store excess electricity during peak production to cover the lower production hours, but when that stored energy is in demand, it messes up the balance. This makes it difficult to stop relying on fossil fuels, which produce predictable, reliable amounts of power, and usually a lot more of it.

AI can step in to manage variable power systems, improving their efficiency by automatically balancing grid power demand and stability. They can be used to make much more detailed decisions about where and how power flows through a grid, cutting costs by reducing energy waste and the amount of work people have to do to balance power grids with fewer results. AI can process data that covers the entire grid in detail to make its decisions. It’s faster and better informed.

However, there are some ethical complications with integrating AI into power grid management – particularly around data privacy and cybersecurity. For AI to manage power grids, it has to have continuous access to enough detailed information about how much power people and companies are using and when they use it. Some people may not want to share details of their power usage since it could be used to infer private information about how they live their lives. 

Concerns about cybersecurity might also impact how willing people are to share personal data. Monitoring systems for AI power management increase vulnerabilities for both physical system security and cybersecurity. While companies are responsible for safely handling personal data, it’s not always clear what “safe handling” means at the proving ground for a new technology. It’s up for debate whether companies are doing enough to protect people’s privacy even without the added complexities of AI, since data breaches have already become commonplace.

There is also the potential for bias in how AI manages power grids – AI might make decisions based on data that left out groups of people or the needs of specific regions. AI might not have the context to understand why some regions have lower historical power draw and could perpetuate generational patterns of discrimination by routing power based on history instead of equity. It might make decisions prioritizing company profits over protecting local environments, depending on what the company has programmed it to do. 

Making power consumption more efficient

Another way AI could make green energy more viable is by lowering energy demand, which can be done by making power consumption more efficient. This makes it easier to get enough energy from green sources even if they’re less efficient than fossil fuels. The potential here is huge – according to the Yale Clean Energy Forum, optimizing factories with AI could reduce energy consumption, waste, and carbon emissions by 30 to 50%.

AI can reduce power consumption through active resource management and optimizing manufacturing processes, transportation, and building management like automated climate control. It can also actively monitor for failures and make predictions about the best times to conduct maintenance, reducing downtime. 

However, the widespread adoption of AI is driving the energy used to run AI tasks through the roof. The energy cost of an AI task varies depending on the model used and the type of task, and while it’s constantly evolving, what we do know is the sheer number of tasks is skyrocketing. Companies powering AI systems have been turning to fossil fuels to meet the increased energy demand, although Microsoft and Meta have both made promises to develop new nuclear power plants as an alternative.

It's hard to know the true energy AI demands – companies aren’t releasing data on their energy use for AI, so people are creating rough estimates based on sales of tech, new data centers, and other clues. Beyond that, current estimates might not reflect how we use AI in the future since the industry is still in its infancy. When combined with the lack of transparency from AI companies, it’s difficult to build a good estimate. That being said, researchers at the Lawrence BerkeleyBerkey National Laboratory have put together a report accounting for as much of this uncertainty as they can, estimating that by 2028, electricity going to US data centers could triple from what it was in 2024. In four years, it could rise to as much as 326 terawatt-hours per year, or enough energy to power 22% of households in the US.

Companies developing AI at the forefront of a tech revolution aren’t going to want to share data for a lot of reasons – competitors undercutting them, people reverse-engineering their designs, and also the backlash from making the environmental costs public. Making too much data public could make it harder for a business to make AI more energy efficient and unlock new capabilities that could reshape how we tackle climate change. Hiding too much data runs the risk of not holding companies accountable for their climate impact.

The right balance between measuring AI energy costs and a company’s data privacy is up for debate. Wherever we decide the ethical line is, it will have to be enforced somehow. Government regulation could help address the lack of corporate transparency once lawmakers across the globe catch up to AI’s development. The EU requires large data centers to release info about their power consumption as of 2023, which goes a long way towards covering the gap in data. 

Equity in the quest for greener energy

AI tech development is not affecting everyone equally. The AI boom has increased demand for much more than electricity – all the resources needed to build processing chips and data centers are also in higher demand. Mining rare earth elements needed to make computer chips is damaging to local environments. There’s also more electronics waste, which is either recycled or ends up in landfills, which can leak pollution into the local water table and have other harmful impacts on a community’s environment.

Running data centers could also disproportionately impact local communities through water use. AI needs a lot of water for cooling – it’s estimated that ChatGPT needs 519 mL of water to write a 100 word email (one of the simpler AI tasks, more complex ones will need more) – that’s about one 16 oz water bottle per email. Google reported increasing their water use by 20%. Data centers could have a serious community impact if located in places like California where water scarcity is contributing to natural disasters, like the LA fires early in 2025. 

There’s also the financial burden of the rising electricity demand. Data centers are often built in clusters to share infrastructure, so they have a concentrated local impact. Virginia has more data centers than any other US state and more are being built there as we speak. Each consumes the same power as tens of thousands of residential homes. 

This raises concerns over where the power will come from and how it will be transported to these data center hotspots. It’s driving Virginia’s electricity costs much higher – both from the increased demand, and also from the costs to build the new electrical infrastructure data centers need – that utilities have been sneakily putting on the bill of Virginia’s residents instead of the corporations using the data centers.

Summary and further thoughts

AI could rewrite how we approach green energy, making it efficient enough to overcome some of the challenges it faces with irregular power output and more. There are still so many questions we have to think about so we can use AI as a powerful tool for making our future brighter without (inadvertently or not) causing harm to our communities. Should we prioritize privacy rights over a more efficient energy grid? What about privacy and holding companies accountable for their environmental impact? What if some governments don’t prioritize setting up regulations for AI power consumption and green energy? Do other countries have an ethical responsibility to change how they interact with countries that don’t regulate AI? Should we have a global set of ethical AI standards? How can we invest in developing AI without putting the economic burden on people who don’t get most of the benefit from it?

What questions or thoughts do you have about AI ethics and green energy? Start conversations with the people in your community – it’s a great way to understand how such a far-reaching topic affects each of us, and together we can come up with new ways to move forward.