King's College London

Environmental sustainability decisions are often framed as a problem of insufficient information. The vast potential of artificial intelligence appears to be a potential magic bullet to this informational gap. Yet the growing importance of AI in sustainability raises a more fundamental question around the risk of relying on computational systems to reconstruct environmental reality on behalf of organisations.

Unpacking environmental decision-making and outcomes is an ever changing, complex task. Environmental sustainability decisions unfold within systems whose dynamics are difficult to fully observe, predict or evaluate directly at the point of decision. The consequences of those decisions frequently emerge across extended time horizons, while many environmentally significant conditions, including ecological resilience and indirect climate exposure, may remain difficult to observe and evaluate directly at organisational scale. As a result, even the best equipped organisations don’t have a clear source of information to draw on when making environmental sustainability decisions, instead they depend on forms of mediated evaluations of environmental conditions.

Artificial intelligence is often discussed as if it unproblematically improves this process by enabling organisations to model, predict or optimise environmental conditions more effectively. But AI systems don’t only accelerate information processing: they also participate in shaping how environmental conditions become visible in the first place.

Consider a sustainability platform used to assess climate risk across a global supply chain. The organisation may appear to be responding directly to environmental conditions. In practice, however, managers frequently interact with emissions estimates, supplier classifications, projected risk scores and optimisation pathways generated through modelling assumptions, selected variables and computational weighting structures. The environmental system itself is not directly accessible to organisational actors. What becomes actionable is a computational representation of that system. This brings into focus how much we can trust the informational ‘reality’ that AI generates, given the critical importance of organisational decision-making on environmental sustainability. In short, are organisations at risk of making important decisions based on a hallucinated version of environmental conditions.

For decades, organisational scholars have relied on bounded rationality to explain how decision-makers operate under conditions of complexity and uncertainty. Herbert Simon argued that human reasoning unfolds under cognitive and informational constraints that prevent us from fully evaluating environmental conditions and future consequences, forcing organisations to simplify complex environments in order to act.

Bounded rationality remains central to environmental sustainability decision-making because organisations still confront uncertainty, incomplete visibility and competing priorities that prevent full evaluation of environmental consequences at the point of decision. Under AI mediated sustainability systems, however, the primary constraints on managerial decision-making increasingly emerge through computational representations, optimisation structures and model-based evaluations that shape what becomes organisationally visible, measurable and actionable before managerial interpretation even begins.

This shift becomes especially important because AI systems reconstruct environmental conditions through prediction, classification, proxy estimation and optimisation. In doing so, they also shape which environmental relationships receive organisational attention, which sustainability objectives become prioritised and which trade-offs become representable, comparable and actionable within organisational decision processes.

For example, questions concerning long term ecological resilience, irreversible environmental degradation or intergenerational responsibility frequently involve uncertainty and contestation that exceed purely computational calculation.

Artificial intelligence may nevertheless stabilise these tensions into operational decision structures because organisations require systems capable of acting under conditions where direct environmental evaluation remains impossible. This helps explain why AI systems are becoming influential within sustainability governance even when environmental uncertainty itself remains unresolved.

Therefore, the issue concerns how AI mediated sustainability systems reorganise the conditions under which bounded rationality operates within organisations. As organisations depend more heavily on AI systems to navigate environmental uncertainty and complexity, managerial constraints become increasingly concentrated around evaluating computational representations, model credibility, optimisation assumptions and proxy-based estimations through which environmental conditions are rendered organisationally actionable.

This transformation carries implications beyond technological adoption. It raises questions about which environmental conditions become visible for an organisation, which sustainability priorities become strategically actionable and which forms of environmental reasoning gradually disappear because they are difficult to represent computationally? The risk for environmental sustainability governance of over-reliance on artificial intelligence has profound impacts for organisational climate action. To take advantage of the accelerated access to information but guard against falling foul of AI generated hallucinations of environmental conditions, we suggest the following:

Recommendations to improve environmental sustainability decision-making.

1. Environmental sustainability decision processes may benefit from combining AI generated environmental assessments with interpretive and deliberative processes, including domain expertise, contextual interpretation and cross functional deliberation, rather than relying exclusively on automated outputs.
2. Sustainability governance systems may require greater transparency regarding the assumptions, weighting criteria and optimisation priorities embedded within AI systems used to support decision making.
3. Continued scholarly work is needed on how AI systems shape environmental visibility, prioritisation and judgement within organisational decision making, and how these transformations influence the future of sustainability governance itself.