Will AI Save the World - or Destroy It?

 

I was asked that question in 2018 at a large sustainability conference in Birmingham, UK. Nearly a decade later, my answer has not changed as much as you might expect.

In 2018, I spoke at a conference about the circular economy, artificial intelligence, and the Fourth Industrial Revolution.

AI was rather different then. ChatGPT did not exist. Most people were not asking a machine to write their emails, diagnose the strange noise coming from their washing machine, produce a business plan, and explain quantum mechanics before breakfast.

But one of the questions we discussed was surprisingly prescient: will AI save the world — or destroy it?

My answer was firmly at the optimistic end of the spectrum. Not because I thought technology would magically fix climate change, inequality, or resource scarcity. But because I believed AI could democratise knowledge at extraordinary scale.

Nearly ten years later, I still think that. But the answer has become considerably more complicated.

We are already seeing what happens when capabilities that were once scarce become almost universally accessible.

Tasks that previously required specialist knowledge, significant time, or expensive external support can increasingly be accelerated by AI. That is enormously powerful. It is also disruptive.

Some jobs will change substantially. Some business models built around information asymmetry will struggle. Some skills that commanded a premium because they were scarce will become less scarce.

And there is an uncomfortable irony here: the technology we hope can help solve some of humanity’s biggest environmental problems also consumes significant resources itself.

Data centres consumed about 1.5% of global electricity in 2024, according to the IEA. Its base case sees that roughly doubling to around 3% by 2030, with AI an important driver of the increase.

So no, technology is not impact-free. But neither is the story remotely as simple as AI equals enormous energy consumption equals bad for sustainability.

The clean-energy transition provides a useful reality check.

In 2025, the world installed a record 800 GW of renewable capacity. Solar represented around three-quarters of it. Renewables supplied about 34% of global electricity, compared with 23% only a decade earlier.  (Source: IEA, Global Energy Review 2026)

That did not happen because everybody suddenly became more environmentally virtuous. Technology improved. Costs changed. Capital moved. Deployment scaled. That is important.

We sometimes talk about sustainability as though progress depends primarily on persuading billions of people and businesses to make slightly better individual choices. Of course behaviour matters. But history suggests that making the better choice cheaper, easier, faster, and more useful is considerably more powerful.

That is what technology can do. And AI potentially takes that principle into an entirely new category: knowledge and decision-making.

Data everywhere, converging on a human decision......this is something we see constantly at Rio.

 

Large organisations can have extraordinary quantities of information: energy, carbon, buildings, investments, waste, water, suppliers, materials, procurement, travel, climate exposure, compliance obligations, and much more.

 

The problem is increasingly:

• What does it mean?
• How does our performance compare?
• Where is the risk?
• What should we do first?
• What happens if we make a different decision?
• And perhaps most importantly: can we trust the answer?

That is where I think the next phase becomes genuinely interesting.

 

AI can make sophisticated analysis accessible to vastly more people. But combine AI with trusted enterprise data, scientific methodologies, rules, benchmarks, and human expertise, and you move beyond simply democratising information.

 

You begin to democratise decision intelligence.

This is the bit that sometimes gets lost in the excitement.  Generative AI is extraordinarily good at producing plausible answers. Businesses do not need plausible sustainability answers.

 

If you are deciding where to deploy capital across 5,000 sites, calculating financed emissions across an investment portfolio, assessing climate exposure, or making a regulatory disclosure, “the chatbot sounded convincing” is not an adequate control framework.

 

We need to know where data came from.

This is one reason we are particularly interested at Rio in approaches that combine the flexibility of AI with structured data, domain knowledge, and deterministic rules — including the potential of neurosymbolic AI.

 

The objective is not to build an AI that knows everything. It is to build intelligence that can help humans make better decisions and explain why.

 

That is probably expecting rather too much from a collection of GPUs.

 

Could it cause serious problems? Absolutely. The economic disruption, resource requirements, misinformation risks, and concentration of technological power deserve serious attention.

But I remain an optimist.

 

I believed in 2018 that democratising knowledge could be an extraordinary force for progress. I believe that even more strongly today, although we can now see much more clearly that democratisation comes with a price and requires thoughtful governance.

 

I also still believe something else. People who want to build, achieve, solve problems, and create value will continue to find ways to do so as technology changes the structures around them. The tools change. Human ambition does not disappear with them.

 

Our challenge is therefore not to decide whether AI is inherently good or bad. It is to decide what we want to do with it.

 

For sustainability, the opportunity is enormous: understand complex systems, make expertise accessible, compare performance, identify interventions, model consequences, and direct human attention and capital towards the places where they can have the greatest impact.

 

Perhaps the question I was asked in 2018 was the wrong one.

 

AI probably will not save the world. But, properly harnessed, it might give us some extraordinarily useful tools to do the job ourselves.