Targets are central to environmental policy. Governments regularly establish objectives for emission cuts, biodiversity recovery, water quality and pollution abatement, nature restoration, circular economy targets and climate adaptation. Such goals are important because they offer direction and foster accountability.
But targets do not implement themselves.
In many ways, the most challenging part of environmental governance is not the setting of an ambition, but how it is turned into action. Ambitions need to be funded, monitored, enforced, regularly updated and coordinated between multiple agencies. Data needs to be gathered and interpreted. Local authorities require resources. Businesses require clear regulations. Citizens require transparency. Regulators need to be able to react when issues arise.
It is in these areas that AI is starting to attract interest.
AI cannot solve environmental governance on its own. It is not a replacement for political decision-making, public institutions or local expertise. But it can support governments and regulators to better use data, spot risks in a timely manner, evaluate policy performance and go from broad commitments to more focused implementation.
In short, it is not a question of whether AI can write good targets, but whether it can assist in delivering them.
From targets to delivery
In many areas, the problem is the space between setting targets and their execution.
Governments might pledge to support the restoration of ecosystems, yet not know where to spend money first. Regulators may have stringent pollution legislation, but not enough capacity to visit all facilities. Water authorities could be aware of river pressures, yet not know precisely where the pressures arise in real time. A planning system could grant permits for new clean energy facilities, but only have partial information on the cumulative impacts of these developments in terms of habitats and land use.
This implementation conundrum is a familiar one and was discussed in Environmental Governance After the Green Deal. Environmental policies in most areas increasingly include a long list of aspirations at the top of their hierarchy. What gets done in practice depends on institutions, budgets, data, monitoring and enforcement.
In a number of places, AI can help to address the issues at stake by analysing large volumes of data, spotting patterns, finding anomalies and informing more targeted decisions. It can help politicians, regulators and others know where efforts are lagging and where risks are accumulating.
But AI will only make a difference if institutions have the capacity, transparency and accountability to use it effectively.
Enhancing the monitoring of environmental change
Good environmental governance must be based on knowing what is happening.
AI can assist in this by monitoring and analysing satellite data, sensor readings, inspection reports, climate data, water quality data, land-use changes and biodiversity data. Such techniques have been used to spot instances of deforestation, pollution occurrences, illegal land conversions, shifts in rivers, crop stresses, urban heat events and habitat fragmentation.
If applied to biodiversity, governments could track changes to habitats and species more rapidly. Under the water policy umbrella, AI could reveal patterns of pollution across river catchments. Climate adaptation efforts could benefit from forecasts of higher-risk areas, including those exposed to flooding, droughts and heat stress.
AI has a role to play in building the evidence base on environmental change, which is a key need for the protection of freshwaters and the use of public funds for nature restoration.
But monitoring is not, in itself, action. Detecting deteriorating conditions is one thing, but someone then needs to look for the source, make sure standards are followed, invest in getting the water back up to scratch, or shift land use on the ground.
AI makes issues more visible; it does not generate political will to address them.
Smarter environmental enforcement
One of the most promising applications of AI in environmental governance could be in compliance and enforcement.
Regulators typically have insufficient capacity to visit every site or inspect every development. AI-enabled tools could help pinpoint which cases carry the highest risk by combining insights on compliance history, emissions levels, complaints, permit status, remotely sensed data, and local site characteristics.
The idea would be to enable more risk-based enforcement.
If a facility has a history of poor compliance, sits next to a sensitive water body and has unusual emissions patterns, it might be flagged for closer inspection. If satellite imagery indicates unauthorised clearance of land, authorities could react more speedily. If water-quality sensors detect a sudden deterioration downstream from industrial zones, regulators could begin a probe before further damage is done.
AI could make enforcement more targeted. But it cannot function as a black box. Should the technology affect which establishments are targeted for inspection, what penalties are imposed, or whether permits are granted, the rationale must be clear.
Firms and communities should be able to follow the reasoning behind the judgements. Ultimately, human authorities should retain final decision-making authority.
Supporting spatial planning
Much environmental governance fails because it does not grasp the wider spatial picture when decisions are taken on a project-by-project basis.
Planners could deploy AI to weigh up competing priorities on a given area of land or river catchment, combining knowledge about habitats and biodiversity, renewable energy infrastructure, agriculture, flooding, transport networks, conservation-designated sites, species connectivity, urban growth and public exposure to pollution.
This could be particularly useful for clean energy spatial planning. Renewable energy generation is crucial, but it does have to be sited carefully. Using the technology, authorities could find spots for renewables that are least likely to harm nature, have good grid connections and cause least friction with food production, water protection or communities.
It fits with the ideas in:
How Better Spatial Planning Can Protect Biodiversity and Support Clean Energy
What Makes a Renewable Energy Project Environmentally Responsible?.
Planners could employ it to spot connections that might otherwise be missed. It should not, however, supplant public participation in planning decisions. The process is about values, trade-offs and experience, which no machine can grasp. It can only ever aid in presenting the options, but not determine which is right.
Helping evaluate policy outcomes
Environmental legislation can be passed without much knowledge of what is likely to work most effectively.
In principle, AI could make it easier to evaluate policy by sifting through large volumes of policy, environmental and economic data. This could help governments understand which programmes have reduced air or water pollution, boosted the success of restoration schemes, encouraged circular economy objectives or relieved pressure on biodiversity.
This matters because the environmental management system ought to learn and adapt. If a subsidy is not helping biodiversity improve, the scheme should be redesigned. If a limit on pollution is not reducing emissions, it may be necessary to tighten up enforcement.
For example, if restoration funding is yielding tangible results in one landscape but stalling in another, it is important for policy-makers to know why. AI could potentially help examine patterns between different regions and sectors more rapidly than is feasible through conventional assessment methods.
That could help facilitate improved policy-making decisions on the economics of biodiversity and public spending on restoration. It could also help governments understand where environmental spending generates the most long-term return.
Yet even assessment needs human insight. Even if an AI model can discern a correlation, policy-makers will still need to understand the cause, the context, and any spillover effects.
Practicalities of circular economy governance
Circular economy policy needs to track material flows through a supply chain.
AI could help analyse waste flows, product lifecycles, recycling quality and efficiency, material demand, repair infrastructure, and secondary material markets. It could support better materials separation and sorting at waste facilities, find opportunities for reuse and remanufacturing, and help business owners understand where materials and value are being lost.
This could make circular economy governance more practicable.
AI could help with this information. However, circular economy AI should not only look at how waste flows can be optimised after a product has reached the end of its functional life. It should also help identify how waste can be reduced during the design phase.
Otherwise, it is possible that we could improve the efficiency of a wasteful system, rather than making the system circular.
Assisting farmers without substituting for their knowledge
AI is also having implications for agriculture and land management.
Digital tools can help monitor soil health, forecast disease and pest risks, optimise irrigation requirements, detect crop stress, improve productivity, and reduce inputs. This should support the development of lower-risk farming systems and contribute to reducing the environmental pressure on water, biodiversity and soils.
In agriculture, AI also raises governance questions: who owns the data? Could farmers afford access to such systems and solutions? Are the recommendations produced by AI models transparent? Does the technology encourage farmers to reduce their use of pesticides and fertilisers, or does it only optimise their productivity?
Is it supporting farmers, or making them overly dependent on the development and services provided by third-party suppliers of potentially expensive, proprietary solutions?
AI should help farmers to make better decisions. It should not constrain the scope for farmers to make their own decisions, or force them into farming systems that are detrimental to environmental outcomes.
Risks of poor-quality data and weak institutional accountability
The quality of AI outputs will be dependent on the quality of the underlying data, assumptions and institutions.
Data on environmental issues is often incomplete. There are regions where monitoring data on environmental quality is not available. There are communities and groups who may not feature in environmental data. There are sources of pollution that go unmonitored and undocumented. There are parts of the biodiversity data that are not available, or are patchy.
If AI is trained with such weak or biased data, the outputs it produces could prove misleading or inaccurate.
This has real implications for governance. The model may not accurately estimate risks where the data quality is poor. It could focus on problems which are easier to measure and address, rather than those where action is required the most.
It could recommend solutions that are technically efficient but may not necessarily consider social implications. It could make decision-making harder for stakeholders to challenge due to the complex, technical nature of the system being evaluated.
Environmental governance, therefore, should view the use of AI as a tool, with checks in place to prevent abuse.
Data sources and methods, alongside uncertainty and limitations, should be open. Those in public office who use these tools should be able to articulate how decisions are made with their aid. Any community subject to these decisions should have a mechanism to challenge them.
AI is not environment-neutral itself
AI should not be considered environmentally neutral.
Data centres, hardware and AI cooling require energy, water, land and materials. Hardware depends on mineral extraction. Digital infrastructure creates electronic waste. If AI applications become popular without appropriate environmental guardrails, they add stress to the same systems environmental policy seeks to maintain.
This is relevant for governance.
Policymakers should ask whether the environmental gains of an AI application outweigh the costs. For example, an application that can prevent a large-scale environmental polluter from operating or improve the efficiency of an energy system may be worth its own impact, while a new tool that adds complexity without producing better outcomes may not.
Environmental AI should be assessed on the full value chain.
Energy, water, hardware and procurement requirements, the energy used for data storage, emissions and waste generated, and the potential rebound effects should all be taken into account. In other words, tools that are used for environmental governance should themselves meet environmental standards.
AI can help implementation but should not replace it
The potential for AI to support environmental governance should be welcomed, but with clear limits.
AI may help to identify risks, monitor changes, evaluate policies, inspect and plan, and understand complex systems. It can make governments move faster with data and enable decisions that use more evidence.
However, the fundamentals of governance cannot be substituted for by artificial intelligence.
There must still be:
- Robust laws;
- Well-funded regulators;
- Transparent institutions;
- Public participation;
- Adequate budgets;
- Independent science;
- Enforcement capacity;
- Political accountability.
Artificial intelligence may be able to identify which rivers have poor water quality, but it cannot make the decision that water quality matters. AI tools may identify which locations need more enforcement activity, but they cannot replace the power of the regulator to do so.
A land-use model generated through AI cannot decide which development plans a community should have or refuse.
AI tools can improve implementation of environmental policies, but they must not become a substitute for democratic choices.
From digital to green
Data-enabled governance will be the next frontier for environmental policymaking.
Governments will have access to greater amounts of satellite imagery, sensor networks, modelling and predictive systems, along with automated reporting and analysis tools. AI will have a role in helping turn all these new data into improved implementation.
The core challenge remains political and institutional. Targets will only have traction insofar as there is a working delivery system behind them. AI can strengthen these systems, but only if used responsibly, transparently and in pursuit of a clear public purpose.
The best use of AI in environmental governance may be to help the system identify environmental challenges earlier, act more fairly, spend money better and ensure the polluter or the public decision-maker ultimately bears responsibility for any negative impacts.
In this case, AI can help environmental policy move from promises to practical action.
Responsibility for that action must still lie with people and public institutions.
