Nobody in Exploration Uses AI Right (Yet)

Over the last decade, I’ve watched AI go from a niche academic subject -- to the next hottest thing in mineral exploration -- to the current hottest thing in mineral exploration.

Yet since the launch of ChatGPT in 2022, I fear that AI mineral exploration have gotten too big for its britches, claiming to solve all of mineral exploration’s problems. Today, I want to throw some cold water on the current state of AI in mineral exploration and offer an alternate vision of what AI means for the sector.

AI Hype: Overpromised and underdelivered.

Most people in exploration think of AI as a way to magically find hidden mineral deposits in old datasets. Do they believe it actually works? That's a whole other question. Clearly, the evidence is stacked against this promise. Over a decade into this experiment, AI has not made a single significant discovery.

You know who has made discoveries in that same time? Geologists.

The next Resolution won't be found by Prospectivity Maps.

If we genuinely want to meet the International Energy Agency target of 140M tonnes of copper produced by 2040, we need to find and start producing 40 new Resolution-sized mines in the next decade. But Resolution is one-of-a-kind. It is deep, intact, and high-grade. It will supply 25% of US copper for 40 years once it's up and running.

Could a Prospectivity Map find pre-discovery Resolution? Clearly the answer is no. There would be nothing like it in the training data. We can't expect today's AI to overcome fundamental limitations of physics, geology, and statistics.

We need first principles thinking.

Mineral exploration, throughout history, has always been an endeavor involving trial-and-error. The problem is that the trials are long and the errors are expensive.

Prospectivity Maps will never fix those problems.

The real problems are operational in nature. It takes time to send teams to the field. It takes time to digitize field data collected on pen and paper. You rely on the work of ten different consulting groups. And drilling is still expensive. Whoever solves these fundamental bottlenecks will be rewarded with:

• Early-stage exploration costs in the thousands instead of millions.
• Advancing greenfields projects to drill-ready status in weeks instead of years.
• Drilling accuracy closer to 80% than 10%.

Combine these improvements with AI tools for pipeline and portfolio management and you could feasibly manage 100x bigger portfolios. Lower costs, less risk, and faster discoveries.

Fortunately, we have the science to do this. It's just an engineering problem now. And we're on it.