A dash of optimism and a dollop of caution: Building AI for climate action - Hindustan Times
close_game
close_game

A dash of optimism and a dollop of caution: Building AI for climate action

ByHindustan Times
Dec 19, 2023 07:29 PM IST

This article is authored by Urvashi Aneja, founder and executive director and Dona Mathew, research associate, Digital Futures Lab, Goa.

The use of Artificial Intelligence (AI) can support climate action, but also entails significant environmental costs. We need a more judicious approach to developing AI for climate action - one that is based on a participatory and inclusive political vision for climate action, not market dynamics alone.

AI(AFP)
AI(AFP)

The earth is on the verge of crossing five critical tipping points that are likely to trigger a climate catastrophe in the next 10 years. Urgent steps are needed to slow down and arrest this trajectory.

HT launches Crick-it, a one stop destination to catch Cricket, anytime, anywhere. Explore now!

Technology like AI could play a critical role in climate action strategies by improving our understanding of complex systems and advancing scientific research. For instance, ClimateAI used AI-based simulations to predict that extreme heat and drought in Maharashtra would reduce tomato outputs by 30%. Producers used this information to shift to more climate-resilient seed varieties and change their cropping patterns, averting a shortage.

AI solutions for climate were also a key highlight at the recently concluded COP28. The UN Climate Change Technology Executive Committee, together with a non-profit open-source AI community, launched the AI Innovation Grand Challenge to identify and support the development of AI-powered solutions for climate action in developing countries.

But at what point do the environmental costs of technologies like AI and machine learning outweigh the climate good they supposedly promise?

According to the International Energy Agency, data centres already contribute to 1-1.5% of global greenhouse gas emissions, a number that is expected to drastically increase, especially with the current fascination with building larger and larger models. For example, ChatGPT consumes up to 500 millilitres of water for every conversation of 5-50 prompts or questions. With larger models and more parameters, training time increases, leading to increased water and electricity consumption. Not only is training ML models resource intensive, but their usage also has a significant environmental cost - for example, generating a single AI image consumes as much energy as charging a smartphone. Water required to cool data centres is also likely to increase in tropical, warmer geographies, such as in Asia.

Metals like nickels are also needed to build data centres and other physical infrastructure for technology. Nickel mining is a carbon-intensive process. For example in Indonesia, every ton of the metal-equivalent produced emits an average of 58.6 tonnes of carbon dioxide-equivalent compared to the global average of 48 tonnes.

Research indicates that by 2027 if current growth rates continue, the energy consumption of AI will be 85-134 terrawatt-hours (TWh) of electricity each year, roughly the amount of power used by the Netherlands annually.

The extractive processes entailed in the development of AI also impacts local communities and their land rights. For example, metal mining has caused large scale displacement and destroyed forests. In Chile, lithium mining forced out indigenous communities from their ancestral lands. In Indonesia, 76,301 hectares of land - roughly the size of Bengaluru, were deforested for nickel mining.

As we deliberate how AI technologies can support climate action, we must keep these environmental costs front and centre. Unless we do so, the use of AI is likely to exacerbate climate injustice, with environmental costs disproportionately borne by already vulnerable and marginalised communities.

Balancing between tech working for climate good while also mitigating its harmful effects on the environment will require a radical shift in our current approach to technology. It calls for a more cautious and judicious approach to AI development and use, one that is purpose-driven, participatory and not dictated by market forces alone.

First, current innovation trajectories tend to be top-down and driven by commercial incentives. Instead, we need to build purposeful AI solutions that are driven by specific community needs and priorities. This can help develop solutions that are in the genuine interests of impacted communities.

For example, IIT Delhi’s CoRE Stack initiative recognises that communities are best placed to address challenges around groundwater and forest management. Through a consultative process to identify these challenges, the initiative provides support to adopt technological components in their traditional practices and helps in unlocking funding channels to enable these processes.

Second, the current fascination with larger and all-purpose AI models may not be the best pathway. Instead, we need to focus on smaller, more specific models, that are built on intentionally and locally curated data sets and enable auditability and accountability. Climate solutions need to be highly contextualised, factoring in local climatic factors, economic conditions and demographics. For example, a crop advisory developed for rice varieties domesticated for cold climates will require very different parameters for rice grown in tropical climates.

Third, environmental impact assessments should be mainstreamed for AI. There are currently no standardised frameworks for evaluating the environmental costs of AI production and deployment. Developing measurement benchmarks can facilitate a more judicious use of AI, providing an additional metric for evaluating an AI-based intervention versus other low-tech or no-tech solutions.

Fourth, assessing AI tools is crucial to gauge their efficacy in climate action. Despite numerous AI pilots and research in the pipeline, there is a lack of comprehensive reporting on their impacts. In science in general, research shows that experimentations with AI have led to a reproducibility crisis. A lot of time and resources are invested in research producing claims that cannot be replicated or are useless for practical problems. Evaluating the tools and establishing a repository of best practices and lessons learned, ensuring the protection of values such as equity and privacy, and disseminating findings to researchers, policymakers, and the public is essential to understand the value-addition of AI use.

The suitability of AI for climate solutions requires careful examination. With evidence of their environmental costs, these interventions should be very deliberate and not driven solely for the sake of leveraging AI or for commercial interests.

Finally, it is important to remember that AI, or any other technology, is eventually only a tool - for it to serve climate action, its development and use has to be tied to a broader political vision for climate action. Without this, there is a risk that the use of AI in fact distracts from a coordinated political strategy.

This article is authored by Urvashi Aneja, founder and executive director and Dona Mathew, research associate, Digital Futures Lab, Goa.

SHARE THIS ARTICLE ON
Share this article
SHARE
Story Saved
Live Score
OPEN APP
Saved Articles
Following
My Reads
Sign out
New Delhi 0C
Thursday, April 18, 2024
Start 14 Days Free Trial Subscribe Now
Follow Us On