The Ugly of AI

In my previous discussions on the complex nature of artificial intelligence (AI), I have explored its benefits and the darker aspects that accompany its rise. As we delve into the “ugly” side of AI, it is crucial to examine the environmental impact of these technologies, drawing a parallel between the needs of the human body and the requirements of AI neural networks.

Just as the human body relies on essential elements—air, food, and water—to survive and thrive, AI systems depend on energy, data, and cooling mechanisms to function effectively. The human body requires oxygen to fuel its cells, nutrients to support growth and repair, and water to maintain hydration and regulate temperature. Similarly, AI neural networks need electricity to power their operations, vast amounts of data to learn and improve, and cooling systems to manage the heat generated by their computational processes.

The environmental implications of this comparison are alarming. The energy consumption of AI systems is substantial. Training large language models (LLMs) can be likened to filling a swimming pool with water—once the pool is full, it takes a significant amount of energy to maintain its temperature and keep it from overflowing. A study by the University of Massachusetts Amherst found that training a single AI model can emit as much carbon as five cars over their lifetimes, with some models consuming over 600,000 kilowatt-hours (kWh) of electricity (Strubell et al., 2019). This demand for energy often relies on non-renewable sources, contributing to greenhouse gas emissions and exacerbating climate change.

Imagine a world where the energy consumed by AI technologies continues to rise unchecked. The consequences could be dire: increased carbon emissions, further depletion of natural resources, and a planet struggling to cope with the effects of climate change. The data that fuels AI systems is not just a passive resource; it requires significant infrastructure to collect, store, and process. Data centres, which house the servers that run AI applications, are estimated to consume about 1-2% of the global electricity supply, akin to the energy consumption of entire countries (International Energy Agency, 2021).

Even seemingly simple AI applications, such as text-to-image generation, have hidden environmental costs. Generating a single image using a text-to-image model can consume approximately 0.5 kWh of electricity, comparable to the energy used by a refrigerator running for an hour (Henderson et al., 2020). When we consider more complex tasks, such as text-to-video generation, the energy consumption can increase dramatically, raising serious questions about the sustainability of these technologies. For example, generating a short video can require several times the energy needed for a single image, similar to the difference between boiling a kettle for a cup of tea versus heating a large pot of water for pasta.

Despite the growing awareness of climate change and environmental degradation, the conversation around the ecological impact of AI remains largely absent. As we continue to develop and integrate AI into our daily lives, it is imperative that we acknowledge and address these environmental concerns.

We must advocate for transparency in the energy consumption of AI systems and push for the adoption of renewable energy sources to power data centres. Companies like Google and Microsoft are already investing in renewable energy to power their data centres but is this enough? Additionally, promoting research into more energy-efficient algorithms and technologies can help mitigate the environmental impact of AI. Initiatives such as the Green AI movements encourage researchers to consider the energy costs of their models and strive for more sustainable practices.

While AI holds immense potential for transforming our lives, we must remain vigilant about its environmental consequences. By recognising the parallels between human needs and the requirements of AI neural networks, we can better understand the urgent need for sustainable practices in the development and deployment of these technologies. It is our responsibility to ensure that the evolution of AI does not come at the expense of our planet. Together, we can advocate for a future where AI enhances our lives without compromising the health of our planet.

—

References:

– Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and Policy Considerations for Deep Learning in NLP. *Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics*, 3645-3650.

– Henderson, P., et al. (2020). Towards the Systematic Evaluation of AI Models. *Proceedings of the 37th International Conference on Machine Learning*.

– International Energy Agency. (2021). Data Centres and Data Transmission Networks.