AI has, almost overnight, become an integral part of all our devices. No longer limited to chatbots, AI features are today conspicuously available on several apps used every day for basic functions such as text messaging, emailing, internet searches, etc. The appearance of such features on our interface was automatic and often cannot be switched off, their use thereby becoming unavoidable. The goal of these features is to simplify everyday social interactions and searches, the apps predicting your needs and providing corresponding results. The potential of AI to positively contribute towards the simplification of complex tasks in order to improve the quality of human life is undeniable. This potential also extends to its ability to assist in tackling the effects of climate change by supporting the design of smart grids, creation of low-emission infrastructure, and in the modeling of climate change data. However, AI use is a double-edged sword. AI systems utilize require significant power, both, in terms of computing and electricity. By integrating its use into everyday functions to simplify already straightforward tasks, we may be causing more damage than good. 

In order to be effective, Artificial Intelligence devices are always learning, and therefore, they are always running. For this, the establishment of ever-growing and energy-intensive neural networks is a necessity. These costs add up at every stage of the process. Technology plays a key role in development and quality of life. However, stepping on the gas pedal in a car with no breaks can only lead to us careening down a highway with no breaks, waiting for an impending crash. Any benefits of AI must be balanced against likely consequences, optimized by robust plans to counteract and minimize pollution and emissions.

The cost of development

In 2019, researchers at the University of Massachusetts Amherst analysed the life cycle of several common natural language processing training models in order to estimate the energy cost of training them. They found that the process can emit more than 626,155 pounds of carbon dioxide equivalent, which is nearly five times the lifetime emission of an average American car. 

Source: MIT Technology Review

It is important to note that this data only covers the carbon cost of training large machine learning models. While simpler models produce relatively insignificant quantities of carbon dioxide, it is possible for larger applications to result in substantial emissions. For instance, it took emissions equivalent to 500 tons of carbon dioxide to train OpenAI’s large language model, GPT-3. As datasets and models become increasingly complex, the energy needed to train and run them will increase too.

The AI industry is divided on whether AI methods that leverage computation are better and more efficient than those that leverage human knowledge. While it is difficult to determine which is true, current trends indicate a reliance on increasing amounts of computes and data, requiring greater infrastructure and power. This increase is directly proportional to the industry’s rising footprint. In addition to its emissions costs, the extraction of metals necessary to build the components for AI technology can be connected to human rights violations and environmental damage. 

Putting AI to good use

The carbon impact of the infrastructure associated with Big Tech’s deployment of AI adds to climate concerns. A significant portion of the impact will be connected to the applications these systems are built for. For instance, fossil fuel giants such as ExxonMobil have been using AI to increase and optimize its extraction operations. Adding more fossil fuels, and hence, more greenhouse gases to the atmosphere, is an obvious outcome of this application of AI. AI use in advertising has also been linked to negative impacts, with targeted ads resulting in an increase in overall behaviours of consumption and waste. 

The other side of this coin, of course, is the use of AI to deal with environmental concerns. Machine learning models are used to predict the trajectories of weather events, identify the damage caused by natural disasters, monitor emissions, etc. The potential for this technology to slow down the effects of climate change and save lives is undeniable. 

The cost of maintenance

Given AI’s potential to do significant good, if its negative climate impacts can be minimized, it could become a powerful tool in addressing climate change and other environmental issues. One of the challenges associated with gauging AI’s impact on the environment is the lack of quantification of its energy consumption and carbon emissions. A study in 2019 found that in training a neural network model, the emissions depended on the location of the training server and the energy grid it uses. As a result, experts increasingly recommend treating emissions as only one part of the impact of AI on the environment. Tech companies largely remain hesitant to share emissions data, resulting in a general lack of publicly available information. 

The information gap should not prevent governments from regulating tech companies and their infrastructure requirements, especially in countries with lower compliance thresholds. Environmental standards must be developed and companies must be incentivized to connect to renewable energy grids to reduce emissions associated with the AI use. Furthermore, they must be made accountable for any human rights violations or environmental damage caused by their reliance on AI. 

The cost of disposal

AI’s emissions are only one aspect of it’s footprint; its disposal poses other risks to the environment. The environmental challenges associated with e-waste are not new. By the year 2021, manufactured electronic waste amounted to 57.4 million tonnes. This amount is projected to increase at a rate of roughly 2 Mt per year. The rapid evolution of AI hardware is only likely to reduce device life cycles and result in a greater surge of e-waste generation.

The risks this poses are manifold. The lead and mercury contained in electronic components has the potential to pollute the soil surrounding the areas in which they are disposed. In countries like India, where temperatures soar in the summer, heating of e-waste is likely to result in the discharge of toxic compounds such as lead, cadmium and beryllium into the air. Monsoons, thereafter, can aid in carrying these toxic compounds to other regions and in the seepage of these compounds into ground water, affecting the ecosystem. According to the WHO, exposure to these compounds can have severely negative impacts on health.

What can individuals do?

While an individual’s ability to change these outcomes might be limited, awareness of the effect AI use has on the environment can inform your decision to upgrade your devices, to recycle your e-waste, what you demand of your representatives in the government, and what brands and technology you engage with. These can have small but significant effects on perceived consumer trends and may have the potential to influence the behaviour of companies that produce this technology. Awareness and collective action are the only way to achieve a sustainable and equitable future. 

To this end, there is little individuals can do about AI features included in the applications they use regularly. However, small individual and collective behavioural changes can go a long way. Some steps you can take include:

1. Switching to energy efficient devices. 
2. Minimizing reliance on AI software for basic tasks such as drafting unimportant emails and documents or for entertainment. Each time you send ChatGPT a query, it produces around 4.32 grams of carbon dioxide.
3. Opting out of generative AI features in search engines through their browser settings. 
4. Turning off AI through the tools option on online word processors.
5. Many other apps are yet to make AI enhancements optional, but it will be useful to check the settings to see if such features have been introduced.
6. Turning off autoplay on video sites.
7. Switching to search engines like Ecosia that help plant trees.
8. Using devices with intention so that they last longer – the mining of minerals necessary to produce every new device has far-reaching environmental and social consequences.
9. Switching to audio only when making calls. 96 per cent of the footprint of a video call can be reduced by switching to voice-only. 
10. Using apps such as Commons, Klima, CoolClimate etc. to map the carbon footprint of your individual or family habits, and use this information to optimize lifestyle choices, and to offset your emissions. 
11. Supporting organisations that advocate for fair working conditions and compliance with environmental and other regulations along the tech supply chain.
12. Recycling your electronic devices or donating operational ones that you no longer use, instead of throwing them away. 

In the larger scheme of things, reducing time spent using electronics and the internet will be better for the environment, and for mental health. Demanding frameworks for transparency and ethical utilization from tech companies and policy makers will also go a long way towards regulating AI’s impact on the planet and society. 

Article by: Gauri Anand