Canada’s GeologicAI has joined a succession of small mining tech scale-ups with artificial intelligence in their DNA that have raised more than US$200 million from investors in the past 18 months, underscoring a growing appetite for sharper data analysis in mining. “Data is the new gold,” BHP’s Laura Tyler repeated at the 2023 World Mining Congress.
The mining major’s chief technical officer wasn’t alone in highlighting data’s rising value, nor the caveats, at the conference in Brisbane, Queensland, which was attended by leaders of many of the world’s big miners.
“Every mining operator generates reams of data,” Tyler said.
“But it is how this information is captured, distilled, analysed, stored and used that makes the difference. You get out what you put in – quality outputs from quality inputs – or the alternative, garbage in, garbage out.”
Get it right and the “the prize is clearly massive”, Tyler said.
Investors in companies such as GeologicAI, Ideon, Exyn Technologies, SafeAI, Plotlogic, Fleet Space Technologies and talpasolutions are seeing an opportunity to ride the so-called “smart mining” wave driven by a convergence of key technologies and the industry’s renewed push for discoveries and efficiencies.
Nasdaq-listed Aspen Technologies, acquiring Australian mining software company Micromine for US$623 million, cited a US$11.5 billion value in 2022 for a market that included data management and analytics, smart asset management and traditional exploration and mine planning software. That has been projected to hit $20 billion by 2025.
The past five years has seen a steady infusion of AI into the digital strategies of companies such as BHP and the growth plans of many mining tech firms.
“I think the rate of change is pretty rapid on this front, and I think you’d find very few vendors in the industry who would not purport to be using AI in their products,” Mark O’Brien, digital technology and innovation general manager at CITIC Pacific Mining, told InvestMETS.com.
“A lot of big vendors have soaked up smaller ones with interesting technology and that’s probably been a win in terms of getting solutions into businesses plus the data to support the requirements of AI.
“It’s always been problematic for smart but small start-ups to get a toe in with a mining company, so being part of a bigger ship that is already supplying to a large customer is a good thing for them.”
O’Brien says the hype and buzz around “AI” means he’s “always a little sceptical about claims”.
“But I can definitely see big improvements happening,” he says.
“AI is very much in use in everything from safety systems to operational systems to maintenance solutions and all the way back to into geological analysis and mine planning.
“Machine learning-type type systems are extremely good at processing fast, bulk data to make predictive decisions, so there’s plenty of scope for this in everything from fatigue management solutions monitoring faces in a cab to a health system on a truck reviewing all the sensor data.
“Over 4-5 years there’s been a rapid climb in sophistication.
“And it’s not just the big stuff where we’re seeing AI used. There’s a lot of little, annoying problems that we’ve solved using these technologies. For example, using image recognition systems to process compliance documentation for onboarding workers, or RPA [robotic process automation] systems helping to process accounts payable data in a faction of the time it previously took several staff members.
“So, it’s just everywhere in some form as part of broader solutions or business processes.
“A very good thing I see happening is that a lot of graduates coming out of the mining schools are coming out with really good data skills and that sets us up for the next generation of miners to have much higher expectations and interests in using data.
“We’ve seen that shift within our own business.”
Executive director of leading mining technology company advisory firm, Atrico, Kheong Chee, says as AI becomes pervasive and commoditised the value of unique and compelling datasets and top development and business talent will surge.
“As AI becomes part of the infrastructure layer and becomes embedded into applications, it will profoundly impact every industry, including mining,” he says.
“We’re already seeing the impact of AI across many different parts of the mining value chain, from regional exploration to discovery, development, and mining operations.
“At the operational end, generative AI with large-language models [LLM] such as ChatGPT becoming mainstream will drive the general acceptance of AI tech and likely help overcome the black-box aversion to existing AI-based mining operational solutions.
“The impact might mostly be felt first in the training and reporting areas.”
Chee says AI-augmented analysis can, in the short term, dramatically improve the speed, accuracy, and reliability of human decision-making.
“This means we’ll see AI being widely used in human-augmented applications where there is a reasonably fast feedback loop and the pay-off for a right AI recommendation is massive compared to the downside of a wrong AI decision,” he says.
“In the longer term, AI has the potential to be applied to the majority of workflows, including automating many of them.”
Atrico, which has advised a host of mining tech firms on the buy or sell side of the industry’s major recent M&A and financing transaction surge, sees AI as a pivotal ingredient in the success of a “new generation of start-ups”.
“Similar to mobile and the cloud, we’re seeing generative AI and more broadly, AI, as the next foundational platform for an explosive number of new applications,” managing director Ivan Gustavino says.
“Most of Atrico’s clients already either use AI/ML as a key enabling technology in their solution or are using it as a productivity-enhancing tool. With developer tools such as GitHub Copilot and productivity-enhancing AI agents and tools such as Auto-GPT, we think AI can have an 60-90% reduction on the costs of funding a new generation of start-ups plus accelerate their speed to market.
“In turn, we expect to see reduction of staff at large mining tech companies as they derive productivity uplift from AI and adjust to the higher cost of capital.
“Mining technology companies that are AI from the start will have a strong competitive advantage over other companies. In addition to using a proven playbook for getting product-market fit and scaling, they have a clear strategy for building and improving a valuable dataset and model for AI.”
Walking the talk
Intellisense, a Cambridge, England, based software company that promotes itself as being “at the intersection of AI and critical minerals demand the world needs to get to net zero”, has seen its headcount triple and revenues more than double over the past three years.
Founder and CEO Sam Bose says bringing benefits of “scientific AI” to an industry such as mining continues to be challenging, but agrees that “for those mine sites that get it right the returns are fast and large”.
“For example, AI-driven insight into stockpile contents and ore properties allow for better blending decisions, downstream processes such as grinding can increase throughput with AI-led recommendations on setpoints, and improved operating parameters in the flotation circuit allow for incremental increases in metal recovery,” Bose says.
“These are just some of the gains being experienced by our customers.
“Net zero remains the overriding theme that is driving the cycle.
“We haven’t seen any slowdown across the sector, though the typical inertia that is inherent in mining operations towards adopting new technology always remains a challenge.
“We still see market confusion being created by point, ad-hoc solutions being created that can only work in one site and it is difficult to get scale or network effect benefits from the data. These kinds of internal initiatives often supported by generic consultancies have high failure rates and that tends to feed into the overall inertia preventing miners from getting huge productivity and efficiency advantage provided by these technologies, especially when they are being asked to produce more metals at the lowest environment and input cost footprint.”
Bose says Intellisense’s development of “small” proprietary machine learning models built for specific industrial processes, which allow users to “look inside and influence the model performance” to build trust between end users and AI outputs, has been key to its success.
“The deployment architecture needs to factor in a network edge component,” he says.
“The fusion of first principle or phenomenological models with machine learning approaches is a completely new way of looking at the real world”
“Specifically, with mining being in remote places, there is a need for model retraining to be done on a local hardware at the edge, away from cloud computing infinite resources. The computing limitations of these environments drive the requirement of running small models, reducing operating cost.
“The fusion of first principle or phenomenological models with machine learning approaches is a completely new way of looking at the real world,” Bose says.
“This enables laws of physics are embedded into the data-driven world resulting in more trusted outputs.
“AI solutions have successfully been deployed in other industries such as healthcare to accelerate patient diagnosis and even in the creation of new treatments and drugs. AI has played a big role in manufacturing for quite some while with preventative maintenance solutions that maximise production line operating uptimes.
“However, the operating conditions in the mining industry are completely dynamic with sparse data sets making the successful deployment of AI considerably more challenging.
“This probably accounts for why AI adoption and success has been slower in this industry.
“For example, one of the biggest hurdles is the need to cater for the large degree of variance in ore properties including impurities. The inputs to a manufacturing production line don’t vary greatly, if at all, but the opposite is true for mining.”
Bose says harsh mine operating conditions that can hinder or stop sensors providing crucial raw data is an ongoing challenge.
“Another dimension that often gets forgotten is change management and safety,” he says.
“For users to adopt recommendations generated by AI solutions in environments where the cost consequences and risk to human safety for getting it wrong are so large, the AI needs to be explainable to build user trust.”
Roy Pater, Asia Pacific executive with South Africa’s Ramjack Technology Solutions, told a recent mining data science forum the availability of skilled, experienced data scientists and other tech professionals was a constraint on new mining technology deployment projects, but “system adoption is where things fall over the most”.
“Having the right team on board and getting that change management right is key,” he said.
“If you’re a small operation then obviously it’s better to go and outsource that. If you’re a larger operation it makes sense to build a team of data scientists and business improvement specialists. But most important is having what I call a transformation officer; someone who is responsible for changing mindsets and ensuring that there is system adoption.”
“Miners who talk the talk on innovation but don’t walk the walk on changing their culture and incentives will probably burn a lot of time on ideas that their antibodies will reject”
Rob Foster, CEO and founder of Australian mining digital twin start-up Geminum, says the more novel the AI technology, the more necessary it is to validate the use case and get front-line supervisor and worker buy-in, or “we’ll create another generation of shelfware”.
“Ultimately, AI is just software, albeit data-centric software,” he told InvestMETS.com.
“A shocking amount of software is built and not used, and a lot of software is bought and hardly used.
“We’re doing a lot in AI, and in fact one of our core value props is accelerating AI adoption, or perhaps more correctly prediction and prescription adoption, because the AI term is now being used to cover an awful lot of ground.
“I personally think the LLM hype is going too far for repeatable use cases, and the ecosystem around LLMs that will make them valuable – and sufficiently low risk – is only just forming, and there will be a lot of missteps and dead-ends along the way.
“It’s also very focused on consumer and social tech, not mission critical tech. To me this proves it’s just too early for miners to embrace it.
“But it’s the perfect time to experiment … Let the start-ups light up and in many cases burn out and then let’s see what’s robust enough to survive front line use.
“Juxtaposed against this of course is the example of Tesla: they build and own mission critical advanced systems and it gives them tremendous competitive advantage. But they’re built from the ground up to be innovative, not risk-averse.
“Miners who talk the talk on innovation but don’t walk the walk on changing their culture and incentives will probably burn a lot of time on ideas that their antibodies will reject.”
Foster says enhanced data science workflows and improved machine learning operations that enable engineers to produce cheaper, narrow predictive models – models built and trained to do one thing well – and faster learning model iterations can help miners and vendors rapidly improve single-model use cases in areas such as predictive maintenance and data pre-processing.
“These use cases are already very data model centric and the ever-increasing capabilities in the market will bring multi-model and more advanced learning which should improve the outcomes,” he says.
“We should expect to see every OEM bundling a predictive maintenance solution with their products if they want to protect their after-market servicing.
“I also think the chat capabilities of the pre-trained LLMs will drive improved search experiences, improved aggregation of data, and improved reporting, all of which are really more efficient use of language.
“We’ll see a lot more narrow data models that are ML-based and provide the first narrow predictions in legacy applications.
“These will show up as notifications, or suggestions, and the tech companies will use these to start pushing into predictive workflows that bend current business processes, and hence will have very sporadic adoption.
“This is the real envelope that will be pushed: probabilistic workflows in a rule-based culture.”
Miners are going to test a range of boundaries with the help of AI this decade, according to industry leaders, in areas as diverse as mineral discovery, raw material sorting, environmental monitoring, and security.
Perhaps nowhere will AI’s intersection with mining be more emblematic, though, than via mobile autonomy.
Speaking recently in Denver, Colorado, SafeAI founder and CEO Bibhrajit Halder said advanced generative AI models would effectively train smart heavy machinery – the world’s largest robots – to operate autonomously, “improving precision, productivity, and safety”.
“[Around the world today] there are about 1200 mine trucks running fully autonomously – no human operator, no human control – and they have moved about 7-8 billion tonnes of materials [without] a single fatal accident,” Halder said.
“I think that is an amazing statistic for this industry.”
Halder said the next phase of mining autonomy 2.0 was on the horizon.
“In the next 3-5 years … I think this industry will really connect data in a ubiquitous way,” he said.
“In [mining], where data is abundant and intricate, generative AI can revolutionise numerous tasks. Rather than thinking of the use of ChatGPT, think of generative AI as a powerful platform to develop the next generation of applications.
“Just as electricity enables the operation of refrigerators, which in turn enables the production and distribution of beverages like Coca-Cola, generative AI [can be] the driving force behind transformative applications.”
Tom Palmer, CEO of gold major Newmont, told the World Mining Congress the “exponential acceleration of technological change, particularly in artificial intelligence and long language models” necessitated a “people-centric, values-based” response from corporations and governments to harness the rate and direction of change as the benefits were explored.
“Like any technology, AI, including LLMs like ChatGPT, are advancing and self-learning so rapidly that even their own creators are not sure how they are able to do what they do,” he said.
“AI’s accelerating advancement is creating unknown unknowns.
“We must prepare our businesses and our workforces to responsibly navigate these technological opportunities and threats by anchoring ourselves in our core values so that we can all make moral and people-centric decisions in fast-moving and complex situations either driven or exacerbated by technology.”
