Mining’s next big move is thinking small smarter systems set to outpace bigger machines in the race to adapt, thrive and lead under uncertainty

The mining industry’s next leap won’t be powered by bigger trucks or more data, but by intelligent systems built to adapt, anticipate and thrive in uncertainty.

When Eduardo Coloma, chief executive officer of Maptek, took the stage at the AusIMM APCOM 2025 conference in Perth, he promised to take the audience “on a journey” - and delivered exactly that. Over 30 minutes, Eduardo traced a compelling line from the earliest shocks of artificial intelligence breakthroughs to the looming transformation of the mining sector, drawing bold parallels between the scaling limits in both fields.

His core message was as sharp as it was urgent: the next wave of mining innovation will not come from bigger equipment or more data alone. Instead, success will depend on building intelligent, adaptive systems that thrive under uncertainty - systems that can reason, collaborate, and act, much like the human teams they work alongside.

The tipping point

Eduardo opened with a challenge to one of the most enduring beliefs in AI: that scaling up - bigger models, bigger data, bigger compute - inevitably makes machines smarter. This thinking, he said, has driven extraordinary progress since 2012, when the University of Toronto’s AlexNet model stunned the AI world by slashing computer vision error rates. The leap was made possible by two factors: the availability of massive image datasets and the use of powerful GPUs for large-scale training.

“Each success fed the next,” he recalled. “More data, more compute, better results, more investment. But we may be realising we’ve been scaling in the wrong direction - and like any system pushed too far on a single axis, cracks are starting to show.”

To make his point, Eduardo revisited some of AI’s milestone moments: Deep Blue’s brute-force chess victory over Garry Kasparov in 1997; AlphaGo’s 2016 triumph over Go champion Lee Sedol using reinforcement learning; and the emergence of “move 37” - an unexpected, creative play that surprised even its developers. These events, he argued, marked the start of machines displaying emergent behaviour - behaviour not explicitly programmed, but arising from the system’s own learning process.

Now, a new frontier is forming in the shape of agentic AI systems - not just reactive tools, but goal-directed collaborators that can observe, decide, and act with minimal human input.

“Think about the difference between a calculator and a colleague,” Eduardo said. “A calculator gives you answers. A colleague helps you figure out what to think, what to do, and when to course correct. That is the promise of agentic systems.”

Mining’s own scaling story

Mining, Eduardo argued, has its own history of “scaling as a strategy” - from the exponential increase in haul truck capacity over the past century to the speed and size of modern drill rigs. This industrial revolution in motion has delivered huge productivity gains.

“But physical scaling has limits,” he cautioned. “Haul trucks can only get so big before the road collapses under their weight. Drill machines can only penetrate so quickly before rock mechanics fight back. And beyond a certain point, more horsepower doesn’t mean better margins.”

This mirrors AI’s trajectory: performance gains are flattening even as input costs - financial, computational, and environmental - continue to rise. In both fields, the law of diminishing returns is taking hold.

“We are standing at the same kind of tipping point that AI is facing,” Eduardo said. “Where progress demands a different mindset.”

From automation to adaptation

For Eduardo, that mindset shift means moving away from linear, deterministic planning and embracing the probabilistic nature of mining operations. Mines are not controlled environments; they are complex adaptive systems where small changes can cascade unpredictably.

“Geological models are probabilistic by nature,” he explained. “Yet most of our planning systems still operate as if the world is deterministic. We build schedules based on best estimates and averages - but the real world doesn’t work that way.”

Instead, he argued, the sector needs planning systems that are robust, flexible, and scenario-driven - systems that learn from uncertainty, adapt to new data in real time, and recommend actions with built-in risk awareness.

“This is about moving from automation to adaptation,” Eduardo said. “Systems that don’t just execute but evolve.”

The analogy he drew was biological: the human brain doesn’t store knowledge in neat folders or respond to events in isolation. It builds connections, recognises patterns, and uses context to make decisions. Mining’s next generation of intelligent systems must do the same, moving beyond “tidy spreadsheets” to data models that reflect the interconnected complexity of reality.

The synthetic advantage

One of the most powerful tools in making this shift, Eduardo suggested, will be synthetic data - artificially generated datasets that simulate rare, high-risk, or dangerous scenarios. In aviation, pilots train for takeoffs, landings, and emergencies in simulators because the data from these “edge cases” is too scarce or hazardous to gather in real life. Mining could adopt the same approach.

“Why not use synthetic data to simulate geotechnical anomalies, equipment failures, or supply chain shocks?” he asked. “It’s not artificial - it’s augmented reality for intelligence. It allows both humans and AI to prepare for the rare but critical events that can make or break performance.”

Synthetic data, he argued, will be to agentic AI what sensors were to automation - an enabler of entirely new capabilities.

Rethinking the hardware race

Eduardo also tackled the hardware side of the AI boom. From the Cray-2 supercomputer in 1985 to the iPad 2 in 2011, and now to today’s 100-teraflop GPUs, computing power has followed an astonishing growth curve. But, like truck size in mining, this trajectory can’t continue forever.

“Physics doesn’t scale indefinitely,” he said. “We’re approaching the limits of transistor miniaturisation, power density, and affordability. The question is no longer ‘Can we get faster?’ but ‘What happens when the path we rely on stops working?’”

The answer, he suggested, lies in purpose-built, intelligence-first infrastructure. Hardware like NVIDIA’s new “super chips” integrates memory, compute, and connectivity in ways designed not just for speed, but for reasoning - a move from brute force to cognitive alignment.

“This is a new strategy altogether,” he said. “We’re moving from general power to intelligent coordination - and that’s exactly what mining needs.”

Leading under uncertainty

Throughout his keynote, Eduardo returned to the same central premise: in both AI and mining, bigger is no longer better by default. The next leap forward will come from scaling by intelligence - integrating systems, orchestrating data, and enabling cross-functional decision-making that adapts under uncertainty.

“The winners won’t be the ones who move the most,” he said. “They’ll be the ones who adapt best and coordinate the most intelligently.”

This requires a cultural shift as much as a technological one. Engineers, planners, and managers must be trained to think in terms of distributions rather than absolutes, to see uncertainty as a source of opportunity rather than a risk to be eliminated. And leaders must be willing to challenge “old scaling laws” and invest in smarter, not just bigger, capabilities.

Eduardo closed with a striking metaphor from biology. Across the animal kingdom, brain size tends to scale predictably with body size. But humans - and our evolutionary ancestors - broke that rule, developing far larger brains than expected. This “breakout” in scaling created the capacity for language, foresight, and complex problem-solving.

“Maybe the next leap in AI won’t come from more GPUs,” he said, “but from discovering our own version of that breakout - moving from mammoths to agents. Bigger is not better forever. Eventually we hit limits - data, energy, latency, trust. The next step is smarter systems that reason probabilistically, act independently, and collaborate meaningfully.”

Beyond the tipping point

Eduardo’s keynote was not a call to abandon the technologies that have brought mining to this point. Rather, it was a call to recognise when those strategies have run their course - and to have the courage to lead the next chapter.

“This is the beginning of scaling by intelligence,” he concluded. “Our moment to reshape how we build, plan, and lead under uncertainty - with intelligence, and toward impact. As we embrace agentic systems, richly integrated data, and probabilistic thinking, we position ourselves not just to follow, but to lead. Those who lead this shift will change not just the future of mining, but the next era of human–machine collaboration. Let’s go beyond the tipping point together.”

For an industry often defined by its physical scale, it was a timely reminder that the next great competitive advantage may not be measured in tonnes moved or trucks deployed - but in how intelligently we can adapt to the unknown.