How Is AI Bad for the Environment?

Artificial Intelligence (AI) has become the foundation of modern technological development. We interact with AI every day, whether it’s through search engines, self-driving vehicles, language translation, facial recognition, or even art creation. However, while AI is awesome, there is an inconvenient truth that is hard to digest: the growing environmental impact of AI.

AI can help solve problems related to climate change by optimizing, predicting, and so on, but ironically, it is also contributing to the degradation of the environment in significant ways. Electricity, e-waste, and other aspects, we will explore how AI can impact the environment when there are no rules.

1. The Energy Hunger of Training Large AI Models

The computation needed to train AI models, especially large-scale ones like GPT-4, BERT, and Google’s PaLM, requires a large amount of computational power. This computational power is found in large data centres, which are massive warehouses filled with servers that run 24-7.

According to a report from the University of Massachusetts Amherst, training a single AI model could emit more than 626,000 pounds of carbon dioxide equivalent—which is almost 5 times the lifetime emissions of an average car (in terms of fuel consumption, etc.).

Why does training AI consume so much energy?

When you train AI, you constantly feed enormous data sets into deep learning algorithms, often improving them for multiple iterations. Because of this it takes a lot to:

Use high-speed GPUs/TPUs
Extended compute resource time (days, weeks)
Interview rains to cool the AI to avoid overheating

These processes require large quantities of energy and, in most instances, are operated on electricity produced from fossil fuels further contributing to greenhouse gas emissions.

2. The Carbon Footprint of AI: A Hidden Polluter

While we typically think of carbon emissions in regard to cars, planes, and factories, AI is emerging as a carbon contributor quietly in the background.

How much carbon does AI produce?

The correct amount varies by model and training scenario, but here is a rough ballpark:

Training GPT-3, for example, reportedly also used hundreds of megawatt-hours of energy.
This can represent tens of metric tonnes or hundreds, of CO₂, especially compared to countries where the grid is powered by fossil fuels, such as coal and gas.

And it is not only the training that uses energy as there is also the (inference) running of AI models that operates every time a user queries it. When multiplied by the billions of users worldwide it adds up quickly.

3. Water Use in Data Centres

Data centres don’t only consume electricity, they also use massive quantities of water to cool their systems. Google reported that its data centres consumed nearly 5 billion gallons of water in 2022 alone.

For regions affected by drought or water shortages, it can become a considerable challenge. The increasing demand for AI-powered services also compounds the challenge of water resources.

4. Electronic Waste and Short Hardware Lifespans

AI development is significantly tied to state-of-the-art hardware. This means:

The GPU, TPU, and other hardware used to create AI applications will be continually updated at an alarming frequency
The cycle of obsolescence means that broken machines will be replaced as newer versions are developed
Increased e-waste, which is often dumped into developing countries that will improperly dispose of that waste

Is AI contributing to electronic waste?

Yes. The demand for increasingly faster and more powerful processing is causing companies to discard hardware increasingly quickly; the electronic waste crisis is a worsening issue. A number of these discarded electronic devices will leach toxic waste into the soil and water.

5. Environmental Inequities: Who Pays for It?

One of the less-discussed issues is the divide in environmental damage costs when it comes to AI.

Most AI organization are in rich countries
Most of the materials (e.g., lithium, cobalt) are mined and disposed of in poor countries
These places also suffer from ecologically damaging practices, unsafe working conditions, and water/soil pollution.

In other words, we are not weighted equally as human beings with respect to the environmental costs of AI.

Can AI be Sustainable?

Yes, there is sustainable AI, or green AI; and it is being recognized. The idea here is to optimize your models for efficiency and ultimately your carbon footprint, not just maximum performance.

How can we reduce AI’s environmental footprint?

Choose energy-efficient models and algorithms
Utilize renewable energy to power data centers
Implement transfer learning to utilized pre-trained models
Enhance hardware reuse and circular economy
Increase the tracking of carbon and transparency throughout AI

Companies like Microsoft and Google, for example, are already investing in carbon-neutral or even carbon-negative plans, within their AI divisions.

Conclusion

AI is an important asset, but with great power comes great responsibility. It is easy to focus on AI’s positive aspects in medicine, transportation, convenience, productivity but what about the environment footprint, are we also ignoring that too? The task will be to develop AI that is not only clever – but ethically and environmentally sound.

As users and developers we need to push for greener AI, but also keep these tech companies accountable while also being aware of the negative digital services that we continue to depend upon. Lets not forget, the planet should not pay the price for our progress!

Read about: What to Say to Someone Who Lost a Loved One?

Xorvex