Imagining an Ethical Place for AI in Environmental Governance: Lessons from water management in Guyana

The topic of artificial intelligence (AI) has been ubiquitous following the release of Open AI's ChatGPT language processing model in November 2022. The fast uptake of ChatGPT and other new AI tools since then has prompted governments to weigh their options and seek expert input, even as many questions remain over the tools' benefits and risks.

Early steps by policy-makers to answer the questions posed by AI include the Government of Canada's proposed Bill C-27, the proposed AI Act under consideration in the EU, and the April release by the U.S. Department of Commerce's National Institute of Standards and Technology of a voluntary AI Risk Management Framework.

While laws, regulations, and government-crafted guidance have a role to play in mitigating risks and promoting public benefits, AI is already widely available. As AI uptake increases and new tools emerge, AI users will need to decide how this technology can be responsibly applied in each context, including in the field of environmental governance.

In this piece, I will explore some of the considerations that dominate the current debate around AI, focusing on implications for environmental governance and policy. I will then draw on my knowledge of water management in Guyana to illustrate some of the impacts that previous technologies have had on environmental governance, while outlining some of the questions that remain.

AI: Its promise and pitfalls

The continued development of AI and of the Big Data systems it relies on present some clear benefits to policy-makers and the general public. At IISD, our team working on freshwater science and policy has been exploring the benefits of technologies like AI, Big Data, the Internet of Things, and blockchain in environmental impact monitoring and decision making. For instance, AI, coupled with Big Data, might be used to paint a clearer picture of watershed health and to weigh different solutions, all while engaging the public. In another example, AI is already being used to track progress toward reducing global methane emissions. These benefits are not trivial, and more robust AI systems could make a decisive difference in innumerable environmental issues in coming decades.

At the same time, AI models are often opaque, and it can be difficult for users to evaluate the validity of their findings. For instance, data biases are of special concern for environmental and social justice, with AI potentially compounding other forms of systemic discrimination. Additionally, generative AI like ChatGPT can often muddy the waters by generating its own "source data" (such as fake references or data sources), which can further complicate the task of verifying and validating AI outputs. The opacity of AI can also present challenges for assigning accountability and ensuring that principles surrounding privacy, transparency, and due diligence are met (as was the case in a recent airline lawsuit).  Finally, AI is likely to disrupt environmental industries—such as farming, forestry, and fishing—due to biases, risks of cascading failures, and unequal access. However, these risks are not entirely new or unique to AI, and there are lessons to be drawn from yesterday's innovative technologies in imagining a path forward for tomorrow.

Looking Back: The impact of recent innovations on water management

To illustrate, I will draw on my own knowledge of the issue of water management in Guyana. The Rupununi savannah of Guyana is home to many Indigenous Peoples who live off the land through farming, fishing, hunting, and gathering practices. Recent droughts brought on by climate change are threatening this way of life, and Guyanese policy-makers have been working with international organizations to build new infrastructure, such as wells and dams, and improve year-round water access. Between 2017 and 2019, digital models of water systems, based on satellite imagery, economic datasets and climate science, allowed for infrastructure solutions to be designed remotely and deployed rapidly across dozens of communities.

However, upon speaking with many of those involved, it became clear to me that technology often exacerbated barriers to effective decision making and ultimately to water access. In many cases, the technologies used allowed for data to be collected and decisions to be made without the communities' knowledge or consent. One major infrastructure project faced strong opposition from surrounding communities and was cancelled mere weeks before construction. In another case, a dam flooded a crucial wetland area, impacting local hunting, gathering, and herding practices.

These failures lay, at least partly, in the use of technology as a substitute for other necessary governance practices, such as community consultation and information sovereignty, a problem which Tania Murray Li calls "rendering technical" in her book The Will to Improve. If applied in this way, technologies can impair, rather than improve, environmental governance.

Charting a Path Forward

Yet there is also room for cautious optimism. Many organizations and AI users are combining technologies with a return to basic principles to imagine an ethical place for AI. These include knowledge mapping and human-centred design for collaborative governance, as well as data traceability and blockchain to ensure privacy, transparency, and accountability. 

I asked OpenAI's ChatGPT to write me policy recommendations for climate change adaptation in the Rupununi and in seconds it suggested numerous solutions like sustainable land management and strengthening water resource management. These results were highly relevant to the subject, yet the process used to generate them is completely opaque. There is no clear way for me to decipher what information specific to Guyana and the Rupununi (if any) was used to generate these proposals. There were also omissions of topics, such as fisheries, fire management, and Indigenous subsistence practices, which I know from experience to be of high importance. Despite its polished language, this sort of AI-generated result is better understood as an initial foray into a policy topic rather than a conclusive analysis.

While it remains a useful and impressive technology, generative AI cannot replace years of in-person experience and research, including active engagement with and consultation of the communities impacted by environmental issues. Like other technologies before it, the net impact of AI will depend on our ability to optimize its use, to account for its limitations, and to retain space for the core principles of sound environmental governance.