From gun searches to kimberlites, AI makes inroads in Canadian mining

An artificial intelligence (AI) executive who co-founded a company searching for concealed weapons is now fronting another firm using similar technology to hunt for hidden mineral deposits.  

The company, Montreal-based Windfall Geotek, has worked with explorers such as Canada Nickel (TSXV: CNC; US-OTC: CNIKF), which just landed a $24 million investment pledge from Anglo American (LSE: AAL), and Puma Exploration’s (TSXV: PUMA) Williams Brook gold property in New Brunswick, among other projects.  

Dinesh Kandanchatha, chairman and interim CEO, helped start the company Extract One, which scans for hidden weapons at concert venues and other sites using AI technology seeking out anomalies. Kandanchatha had also worked on scanning patients’ eyes for signs of diabetes using the same concept. 

“The foundation of the algorithm was anomaly detection which made a lot of sense in the mining context,” Kandanchatha said in a phone interview. “We have two data models, one for base metals and one for gold and silver.”  

AI’s surging importance was shown this month when Microsoft announced it would spend billions to revamp its Bing search engine and Edge browser with AI in a bid to take on Google. And news of the AI chatbot application ChatGPT, which mimics human responses, continues to sweep across marketing and academia.  

While chatbots are widely different from the mining industry’s AI efforts, the programming shares the ability to learn with experience. The industry is seeing how AI can not only help in crunching data to find deposits, but cut costs by increasing processing efficiency, upgrading automation and improving safety.  

The global mining industry’s investment in AI may be increasing by 10% to 23% a year from US$769 million in 2021, according to London-based GlobalData. Consultant McKinsey & Co. has estimated mining sector investment in AI may realize savings to miners of US$390 million a year by 2035.  

“Deep machine learning (DML) has potential to help with exploration without any doubt,” Patrick Mercier-Langevin, a research scientist in Quebec City with Natural Resources Canada,” said by e-mail. “It is difficult to assess if DML, although seemingly successful [in some cases], represents an improvement on targeting and outperforms conventional geologists/geophysicists interpretations.”  

Kandanchatha noted how humans can figure out data in a small set of variables or dimensions, often helped by their ability to discern patterns. But mining’s geophysical, geological and geochemical data, satellite imagery and soil types are overwhelming. So, AI makes connections among the data to identify sets of signatures and contexts for tens of thousands of models and compares them to different datasets from around the world.  

“When people go out and drill, we can see that the pattern they saw was validated and it goes back into the algorithm,” he said. “Or maybe in this other case you were wrong because sometimes you drill and there’s nothing there, so the AI is constantly learning based on the validation.”   

Next-level searching 

It was more than three years ago when AI technology developed by GoldSpot Discoveries, now part of Brisbane-based ALS, was behind the drill hole that rocketed New Found Gold (TSXV: NFG; NYSE-AM: NFGC) to the forefront of the central Newfoundland gold surge. Hole NFGC-19-01 returned 19 metres grading 92.9 grams gold per tonne and 6 metres of 285.2 grams gold from 96 metres downhole at the Queensway project just west of Gander, NL.  

“We got a lot of good validation out of that work,” GoldSpot and New Found Gold co-founder Denis Laviolette, who continues to run Toronto-based parent firm EarthLabs, said by phone.  

Now, Vancouver-based North Arrow Minerals (TSXV: NAR) is betting AI will help it find new diamonds in places across Canada’s north where humans wouldn’t think to look. It wants to emulate Arctic Canadian Diamond, the owner and operator of Canada’s oldest operating diamond mine, Ekati, which used DML programs to find new kimberlites in the country’s most-explored gem area in the Northwest Territories.  

North Arrow has used the programming written by Vancouver-based technology company Mineral Services Canada to find preliminary targets at its Pikoo project in central Saskatchewan. It plans to make a second pass with the AI this year.  

“We’re really confident that there are new discoveries to be made,” Ken Armstrong, chief executive officer of North Arrow, said in an interview. “This is another tool we can use to help try and identify and build out those targets for drill testing later on.”  

Vancouver-based Mineral Services compiles magnetic, electromagnetic and gravity data from property flyovers and feeds them into DML algorithms to predict the location of kimberlite pipes, Armstrong said. He said he’s especially encouraged by how the AI found the 1.7-ha Bear kimberlite for Arctic Canadian, a relatively large pipe in an area that’s been scoured by prospectors for decades.  

“This Bear discovery is really intriguing because it’s so big in such a heavily explored brownfield environment,” Armstrong, a 25-year veteran of diamond exploration, said by phone from Vancouver. “It shows the platform clearly has promise.” 

Still, there’s no word at this point if Bear holds economically viable gems. Arctic Canadian didn’t reply to an email seeking comment.  

Gilles Bellefleur, a research scientist in Ottawa with Natural Resources Canada, said he was impressed with how the Mineral Services team prioritized the predicted targets with human expertise to avoid drilling false-positive anomalies, a potentially costly error.  

“Deep learning techniques are very efficient at recognizing complex patterns in data from known examples, e.g., kimberlites, and at finding similar patterns on a large volume of data elsewhere,” Bellefleur said in an emailed reply to questions. “Of course, predictions are never 100% perfect. Their method may have missed some kimberlites or misidentified other geological bodies as kimberlites.”    

Armstrong said he couldn’t be specific about the cost of using Mineral Services. 

“But fair to say it is not high because it does not require additional field work,” he said. “It uses existing exploration datasets which were much more expensive to acquire in the first place.” 

Geotek charges a fee for processing an explorer’s data and may seek equity stakes or potential royalties for other projects depending on the size and time required. Most contracts are three to four years, Kandanchatha said.  

“Our goal is to provide a platform where you’re able to upload your data and leverage the AI to generate targets,” he said. “We’ve spent the last two years taking what took us 18 weeks and three people using a supervised algorithm to now do in two hours, unsupervised.”