What AI reveals about the state of environmental justice in India

India faces severe environmental challenges – it is currently home to 21 of the world's 30 most polluted cities, with air quality indices regularly reaching "severe" levels in metropolitan areas and unabated water pollution in major rivers. Despite environmental legislation dating back to the 1970s, implementation remains problematic.

The challenge of analysing legal data at scale

Given the failure of the executive and legislative branches of government to address this issue, India’s judiciary – including the Supreme Court and the National Green Tribunal (established in 2010) – has often taken an activist stance and issued landmark rulings to address environmental challenges.

However, a crucial question remains: How effective are these judicial interventions in protecting the environment?

Until recently, researchers faced significant barriers to systematically analysing court interventions. Traditional legal analysis methods struggle with inconsistent data formats across court records and lack standardised tagging for case numbers, dates, and actors.

The sheer volume of records created in the system and the variations in language across the courts also raise challenges. As a result, most empirical studies have focused on small samples or a narrow set of variables for analysis.

Artificial intelligence can evaluate environmental jurisprudence

Our research began with a comprehensive review of India’s environmental laws. We identified three foundational acts (Water Act 1974, Air Act 1981, and Environment (Protection) Act 1986) and found 2,996 judicial rulings that cited these acts on the free legal search engine IndianKanoon.org.

In these rulings, we discovered 23 additional relevant legislative acts cited within these cases. Our final corpus had 12,615 environmental court rulings that cited all these acts. A team of law students manually read and labelled 1,905 cases. Next, we compared the performance of two large language models (LLMs) – OpenAI's GPT-4 and Anthropic's Claude – against human expert analysis.

Only one-third of rulings are pro-environment

Perhaps the most striking revelation in our analysis is that neither humans nor LLM models found widespread judicial activism on environmental issues.

In fact, only approximately 35% of court orders in environmental cases favour environmental protection. When comparing human and AI assessments, we found interesting divergences:

• Human experts classified 25.2% of case rulings as pro-environment ("green")
• GPT-4 initially classified 48.6% as green
• When using identical prompts to humans, GPT-4 classified 35% as green
• Claude consistently classified about 43% as green

Accounting for human cynicism versus AI optimism

We found a striking contrast between human experts and large language models. Drawing on their understanding of implementation challenges and historical enforcement patterns, environmental law experts often assessed seemingly favourable rulings as ineffective in practice.

In contrast, AI models, which primarily analysed formal legal language and outcomes, consistently interpreted the same rulings more optimistically. For example, in a case ruling preventing the use of an illegal polluting machine, human coders classified it as having no environmental impact.

They likely anticipated continued unauthorised use despite the court's intervention. GPT-4, focusing on formal outcomes, coded the same case as environmentally positive. This difference of opinion between humans and AI models is a reminder that while AI offers tremendous potential for scaling up environmental justice monitoring across vast legal datasets, optimal results require combining AI efficiency with human understanding of context.

The accuracy of AI varies across contexts

In the sample coded by both AI models and humans, we found that GPT-4 demonstrated robust accuracy rates compared to human expert analysis. It performed particularly well in cases from more recent years, those with clearly identified parties and judges, and in specialised environmental jurisdictions.

Leveraging AI for environmental research and policymaking

We created a dataset to examine the link between environmental court cases and air pollution levels in Delhi, as well as the rest of India. Combining our legal data set with granular air quality measurements and meteorological controls, we can estimate the initial impact of judicial effectiveness in this context.

By differentiating between Supreme Court directives and local court decisions, we can also quantify the varying impacts of different jurisdictional levels on air quality improvements.

Our dataset has the potential to be leveraged by policymakers for improving environmental governance in India by:

1. Monitoring implementation gaps: Using the data, policymakers can systematically track outcomes, identify “green” court orders, and determine whether these correlate (or not) with environmental improvements.
2. Judicial education: Our analysis reveals how different benches approach environmental evidence – information that could potentially be used to harmonise jurisprudence across India's complex judicial landscape.
3. Accountability: Greater transparency in environmental rulings will enable civil society to hold authorities accountable for implementing court orders.
4. Policy design: Understanding patterns in judicial outcomes can inform more effective environmental regulation design.

AI can transform global environmental governance

Our research demonstrates AI's potential to transform environmental governance by enabling large-scale analysis of judicial outcomes. By revealing that only about one-third of environmental court cases result in pro-environment outcomes, it challenges assumptions about judicial environmentalism.

This methodological approach has applications beyond India. As courts worldwide increasingly digitise their records, AI-assisted analysis could help bridge the gap between environmental jurisprudence and real-world outcomes. For countries struggling with environmental issues or climate change, this approach offers a new lens to examine the judiciary’s role in environmental stewardship.

As AI tools continue to evolve, their application to legal analysis promises to bring greater transparency, accountability, and ultimately, more effective environmental protection. The future of environmental justice may be shaped by this powerful partnership between human expertise and artificial intelligence.