Artificial intelligence could reduce drilling by a factor of five

Artificial intelligence applied to mineral exploration could reduce drilling by a factor of five, generating substantial savings in time and capital while enabling more informed go-or-no-go decisions.

This was the key message delivered by Jef Caers, Professor of Earth and Planetary Sciences at Stanford University, one of the world’s leading experts in applied geosciences and decision-making under uncertainty, during a webinar held as part of the World Mining Congress 2026 series.

According to Caers, this AI-based approach can reduce drilling requirements by fundamentally changing the logic of traditional exploration. Instead of drilling on a fixed grid to estimate grades, the system plans drilling campaigns to falsify human-generated geological hypotheses and strategically reduce uncertainty until a company can make an informed “go-ahead vs walk away” decision.

From self-driving cars to the intelligent prospector

To explain the concept, Caers referred to autonomous vehicles currently operating in San Francisco: “These things work extremely well. They’re very sophisticated. And so we call this type of AI an intelligent agent.”

This “intelligent agent” is not simply a predictive tool. “An intelligent agent is an AI for sequential planning under uncertainty. This is an AI that makes decisions while at the same time optimizing data collection.”

Exploration: Drilling to falsify hypotheses

Caers applied this framework directly to mineral exploration, stating that “all critical mineral supply chain challenges can be seen as sequential planning under uncertainty problems, starting with exploration.”

In conventional practice, companies typically build a single deterministic subsurface model and drill that basis. “An intelligent agent will plan drilling to falsify human-generated hypothesis, then only drill to define grades and tons,” Caers states.

“You can imagine if your hypothesis about the subsurface is completely wrong, then your drilling will be extremely inefficient,” Caers explained.

Rather than filling in a grid regardless of emerging information, the system dynamically adjusts drilling locations to reduce geological uncertainty as efficiently as possible.