This is the second post in a series exploring the nine areas of AI ethics outlined in this original post. Each post goes into detail on the ethical concern and provides practical ways to discuss these issues in a variety of subject areas. For the first post on bias and discrimination, click here.
The original post Teaching AI Ethics provided an overview of nine areas of ethical concern with AI. Although I primarily work with generative AI like text and image generation, many of these can be applied to other types of AI from facial recognition to predictive algorithms.
The environmental concerns of artificial intelligence are less reported on than algorithmic bias, but just as important. In this post I’ll explore the impact of AI technologies on the environment and what AI developers are doing – or not doing – to mitigate those risks.
Here’s the original PDF infographic which covers all nine areas:
Kate Crawford – researcher, author, and leading AI scholar – refers to Artificial Intelligence as an “extractive” technology. In her book Atlas of AI she compares the AI industry to mining, drawing comparisons between oil and precious metal extraction.
The use of rare earth minerals and metals in the manufacturing of electronic components is a crucial aspect of the development of AI. These materials are used in the production of components such as batteries, memory chips, and processors. Lithium, for example, is a key component in the production of batteries used in devices such as smartphones, laptops, and electric cars. Similarly, cobalt is a vital component of rechargeable batteries used in many portable electronics and electric vehicles, while copper is essential for wiring and other electrical components.
However, the extraction and refining of these materials are resource-intensive processes that have significant environmental impacts. The mining of rare earth minerals and metals can result in soil erosion, deforestation, and water pollution. It can also lead to the displacement of local communities and the destruction of their habitats. The production of electronic components also generates a significant amount of greenhouse gas emissions, contributing to climate change.
The demand for these materials is expected to increase dramatically as AI technologies continue to develop and become more widespread. This increase in demand will only exacerbate the environmental impact of their extraction and use. It is therefore essential to find sustainable solutions that reduce the environmental impact of these processes.
In Atlas, Crawford also discusses the human impact of this “extractive” technology – something which I will discuss more in a later post in this series.
The hidden costs of the cloud
Cloud computing relies on massive data centres and infrastructure that consume a significant amount of energy and produce waste. These data centres require constant cooling, lighting, and other support systems to ensure the optimal performance of servers and other hardware. The construction and operation of data centres also require huge amounts of energy, water, and other resources, leading to carbon emissions and other forms of environmental damage.
One of the most significant environmental impacts of cloud computing is its contribution to climate change. The energy consumption of data centres is massive, and as more and more computing moves to the cloud, this demand will only increase. According to one estimate, the carbon footprint of the IT industry is 1.8% to 3.9% of global greenhouse-gas emissions.
Many companies have pledged to make their data centres carbon-neutral or powered by renewable energy sources, but critics argue that these efforts are not enough. Offsetting carbon emissions or engaging in carbon trading does not address the underlying problem of energy consumption and waste production.
Case Study: The Carbon Cost of Training Large Language Models
Large language models (LLMs) are powerful artificial intelligence systems that can generate natural language texts for various applications, such as content generation, summarisation, and code generation. This type of AI has been thrust into the limelight by OpenAI’s ChatGPT, and they have many applications. Unfortunately, training these models requires a huge amount of computational resources and energy, which has a significant environmental impact.
According to a study by researchers at the University of Massachusetts Amherst, training a single LLM can emit as much carbon as five cars in their lifetimes. The study estimated the energy consumption and carbon footprint of four popular LLMs: Transformer, ELMo, BERT, and GPT-2. The results showed that the most energy-intensive model was Transformer, which consumed 656,347 kWh of electricity and emitted 626,155kg of CO2 equivalent. This is equivalent to “nearly five times the lifetime emissions of the average American car”.
Organisations like Microsoft, OpenAI, and Google are investigating ways to reduce this impact, including:
- Choosing more efficient models or algorithms that require less energy or data to train
- Using pre-trained models or transfer learning techniques that leverage existing knowledge
- Reducing the frequency or duration of training sessions
- Using renewable energy sources or carbon offsets to power the training process
- Implementing best practices for data collection and processing
- Adopting ethical principles and guidelines for developing and deploying LLMs
Teaching AI Ethics
Each of these posts will expand on the original and offer a few suggestions of how and where AI ethics could be incorporated into your curriculum. Every suggestion comes with a resource or further reading, which may be an article, blog post, video, or academic article.
- Legal Studies: What laws and regulations exist to govern the use of AI in environmental contexts? How can they be improved to better protect the environment?
- Environmental Science: How can AI be used to address environmental challenges such as climate change, pollution, and deforestation? What ethical considerations should be taken into account when designing AI systems for environmental applications?
- Mathematics: How can algorithms and machine learning be used to model and predict the impact of human activities on the environment? What are the potential ethical concerns associated with these models?
- Social Studies: How does environmental degradation disproportionately affect different communities, particularly marginalised and vulnerable groups? How can AI be used to address these inequities, and what are the ethical implications of doing so?
- Engineering: How can engineering principles be used to design environmentally sustainable AI systems? What role can engineers play in ensuring that AI is used in an ethical and environmentally responsible manner?
- Business and Economics: How can AI be used to promote sustainable business practices and reduce the environmental impact of industries such as agriculture, transportation, and manufacturing? What are the economic incentives for companies to adopt eco-friendly AI?
- Health and PE: How can AI be used to address public health issues related to the environment, such as air and water pollution? What are the potential ethical concerns associated with using AI in public health contexts?
- English and Literature: How can we use literary criticism like ecocriticism to analyse representations of AI in literature and popular culture? What are the potential ethical concerns associated with these representations?
The next post in this series will explore academic integrity and the concept of “truth” in AI. In an age when machines can generate content faster than people, there is a huge risk of misinformation and malicious use. Join the mailing list for updates:
Got a comment, question, or feedback? Get in touch: