Ai environmental approvals risk 'robodebt' disaster, experts warn
Australia’s push to leverage artificial intelligence for faster environmental approvals is facing fierce scrutiny, with leading conservationists warning it could replicate the devastating failures of the infamous Robodebt scheme. The Minerals Council of Australia’s proposal, seeking $13 million in government funding for a trial program, has ignited a firestorm of concern that automated assessments could push vulnerable species closer to extinction.
The ghost of automated errors
The comparison to Robodebt—the automated debt recovery system that wrongly accused hundreds of thousands of welfare recipients of owing money—is not hyperbole, experts argue. The Biodiversity Council, a coalition of 11 university researchers, asserts that while ai can assist with routine tasks, automating complex environmental assessments presents a significant risk. “The current environmental legislation is riddled with ambiguity, allowing for broad ministerial discretion,” explains Lis Ashby, the Council’s policy and innovation lead. “This inherent vagueness, already a bottleneck in the assessment process, would be amplified by an ai tool lacking the nuanced judgment of a human assessor.”
The core issue lies in the data—or lack thereof. Brendan Sydes, biodiversity policy advisor at the Australian Conservation Foundation, voices skepticism, stating, “Technology can certainly improve efficiency, but ai is a poor substitute for rigorous, data-driven decision-making.” He advocates for a focus on filling critical data gaps regarding threatened species and habitats, rather than relying on potentially flawed algorithms.
Professor David Lindenmayer, a forest ecologist at the Australian National University, highlights a stark reality: “A staggering one-third of Australia’s threatened species remain unmonitored, and data on many others is patchy at best. Human assessors routinely consult with experts to bridge these gaps. ai, however, is only as reliable as the data it consumes—and robust data is conspicuously absent for a vast number of Australian species.” The risk of decisions based on incomplete or outdated information is, therefore, alarmingly high.
Beyond the data: a flawed training ground
Even if data availability were improved, Professor Hugh Possingham, a leading conservation biologist at the University of Queensland, raises a critical point about the training data itself. “ai tools require extensive datasets for training. The past two decades of EPBC Act approvals are demonstrably unsuitable—the Act has consistently failed to protect the environment.” He advocates for bolstering the workforce of assessment officers as a more effective solution.
Tania Constable, CEO of the Minerals Council, dismisses the Robodebt comparisons as “disappointing,” insisting that the proposed AI tools would merely support human decision-making, not replace it. A government spokesperson echoed this sentiment, stating that while AI’s potential is being explored, “decisions about project approvals will always be made by assessment officers.” But the lingering question remains: can a system built on a foundation of flawed data and ambiguous legislation truly deliver meaningful environmental protection?
The Albanese government’s recent reforms to environmental laws, enacted after a damning 2020 review, underscore the urgency of the situation. The prospect of accelerating these reforms with AI, only to repeat the mistakes of Robodebt, is a gamble Australia’s threatened species can ill afford.