Learning from the Past: What Environmental Catastrophes Teach Us About AI Ethics

From the Great Smog of London to the Bhopal disaster, the 20th century is filled with examples of technological progress that caused serious harm when introduced without ethical safeguards or oversight. Each innovation promised something positive – greater efficiency, more convenience, or faster growth – but in many cases, the risks were overlooked until it was too late. These failures offer important lessons for how we think about the development and deployment of artificial intelligence today.

A Familiar Pattern

Many of the most significant environmental disasters followed a similar trajectory. First, a breakthrough in technology offered clear economic or industrial benefits. Then, the associated risks – whether chemical, atmospheric, or ecological – were dismissed or ignored. Regulation lagged behind. And when disaster finally struck, the response was reactive rather than preventative.

Take chlorofluorocarbons, or CFCs. These compounds made refrigeration safer and more affordable, but their impact on the ozone layer wasn’t understood until global damage had already begun. Or consider leaded petrol, which improved engine performance while poisoning air, soil, and the bloodstream of millions. Even seemingly localised disasters, like Minamata disease in Japan, had lasting global implications for how we understand industrial pollution and public health.

These cases aren’t just moments in history. They are warnings about what happens when innovation is pursued without asking hard questions about who it serves, what it costs, and what systems it might disrupt.

The Lessons for AI

Artificial intelligence presents a different kind of breakthrough, but many of the warning signs are familiar. AI systems are already being adopted at scale across health care, finance, policing, education, and communication, often without clear accountability or public oversight.

We’re also beginning to see the cracks. Algorithms have reproduced and amplified racial and gender biases. Large language models consume huge amounts of electricity and water, adding pressure to already strained ecosystems. Deepfake technology is eroding trust in what we see and hear. And predictive systems are being used in high-stakes decisions with little transparency.

All of this suggests that AI, like the technologies before it, could deliver more harm than good if we fail to apply what we’ve learned from history.

Five Principles from the Past

1. Don’t wait for catastrophe to act. The earlier we introduce safeguards, the less likely we are to face large-scale harm. With AI, that means building regulation in now – not after the damage is done.
2. Invisible risks are still real. In the past, we ignored pollution because we couldn’t see it. Today, AI harms are often just as hidden – whether they involve discriminatory algorithms, misinformation, or environmental impacts. We need tools that help us see and measure these effects clearly.
3. Vulnerable communities suffer first. Many of the worst environmental harms fell hardest on low-income, Indigenous, or marginalised communities. The same is already true for AI. Ethics must consider who is most likely to bear the brunt of unintended consequences.
4. Innovation is not neutral. The idea that technological progress is always good has been challenged time and again by environmental history. AI development must be guided not only by what is possible, but by what is responsible.
5. Global risks require global cooperation. Environmental crises like ozone depletion and climate change eventually led to international agreements. AI is also a cross-border issue, and it requires shared ethical frameworks and standards, not just isolated national policies.

Looking Ahead

Artificial intelligence is not inherently dangerous or destructive. But it is powerful. And that power needs to be matched by wisdom, governance, and humility. If we ignore the lessons of past environmental disasters, we risk repeating them – only this time, at digital speed and planetary scale.

There is still time to guide AI in ways that align with social and ecological well-being. But that will require more than clever code or commercial success. It will take courage, transparency, and public accountability. We need broad debate, clear regulation, and genuine restraint from those developing the most powerful systems.

We have to move beyond asking what we can do with AI. We need to ask: What should we do? Who does it serve? And perhaps most important of all, who will pay the price if we get it wrong?