Junior uses AI to hunt for arctic gold

In a move many considered the stuff of science fiction or fantasy just 20 years ago, junior miner Auryn Resources Inc. has used artificial intelligence to help identify gold deposits at the Committee Bay Gold Project in Nunavut.

Vancouver, B. C.-based Auryn gained control of the remote arctic exploration project in 2015 through its acquisition of North Country Gold Corp. At the time, Committee Bay was known to host significant gold mineralization, including its Three Bluffs deposit. But the folks at Auryn were convinced that the 380,000-hectare (939,000 acres) property situated along the Committee Bay Greenstone Belt in the central Kivalliq region hosted a lot more gold.

The greenstone belt lies some 180 kilometers (110 miles) northeast of Agnico Eagle Mines Ltd.'s Meadowbank Mine and not far from that company's Amaruq and Meliadine advanced development projects, both of which are set to begin gold production later this year.

During the three years since it acquired 100 percent interest in Committee Bay, Auryn has aggressively explored the property, especially along the 300-kilometer (180 miles) strike length where Three Bluffs was discovered. Numerous gold occurrences have been identified along this corridor. In 2017, the junior reported a NI43-101-compliant resource at Three Bluffs with an indicated resource of 524,000 ounces grading 7.85 grams-per-metric-ton gold and an inferred resource of 720,000 oz averaging 7.64 g/t gold.

A year later, Auryn reported discovery of three other gold-bearing systems at Committee Bay in the Aiviq, Kalulik and Shamrock targets.

A second opinion

In January, Auryn reporting taking the unusual step of calling in artificial intelligence experts specifically to employ a technique called "machine learning" to target high-grade gold mineralization at Committee Bay.

Machine learning is the use of algorithms that improve over time through exposure to more data.

Computational Geosciences Inc. of Vancouver joined the Auryn team to employ its proprietary "VNet segmentation deep-learning algorithm."

VNet is a type of convolutional neural that can handle an arbitrary number of geoscience data inputs in either two dimensions or three dimensions. It is also sensitive to sparse or dense data areas, can detect multiple feature resolutions (i.e. regional trends vs. local anomalies) and is scalable across large areas.

This style of deep learning for mineral exploration is an emerging technology that requires expertise in geoscience data processing, data interpretation, and artificial intelligence, according to the AI company.

Auryn said the machine learning platform was designed to generate additional targets across the "Three Bluffs playing field" at Committee Bay, a 1,600-square-kilometer (617.6 square miles) area which includes the Three Bluffs deposit and the 20-kilometer (12.4 miles) shear zone hosting the Aiviq and Kalulik targets where significant gold mineralization has been encountered. The platform will process the vast amount of data (collected through extensive surface geochemical sampling, geological mapping, geophysical surveys, and drilling) from work Auryn and prior explorers have completed on the project.

Emerging trend in mining

Still, Auryn isn't the first mineral explorer to use artificial intelligence to identify mineral deposits, as a wave of machine learning and other AI applications has swept over the mining industry worldwide in recent years.

Though AI is still considered a cutting-edge technology in mining, a growing number of companies, both miners and tech geeks, have already jumped into this industry niche with enthusiasm.

GoldSpot Discoveries Inc , a Montreal-based company formed in 2016 to hunt for gold and other minerals with new technologies, has successfully applied AI applications in mineral exploration and succeeded in anticipating 86 percent of current gold deposits in the Abitibi gold belt by using only 4 percent of the geological, mineral and topological data of the surface area.

Goldspot became the first AI company to go public in late February.

Goldcorp recently collaborated with IBM to inject its smart technology into mining exploration. The IBM Watson services are being used to study drilling reports and geological survey data and to specify which areas to explore on Goldcorp's properties.

Falco Resources Ltd., another Canadian junior, uses and has had recent success, with pattern recognition algorithms known as CARDS – computer aided resource detection system – to predict and find gold.

Falco appointed AI company Albert Mining Inc. in December 2017 to analyze historical data in the Rouyn-Noranda Mining camp in Quebec using its pattern recognition algorithms. The AI software learned the signatures of positive and negative gold and base metals targets and identified new targets with high discovery potential. Fifty anomalies, including 15 gold, 13 copper, 11 zinc and 11 silver were identified and grouped into 11 exploration areas.

In September, Falco reported that it discovered new gold showings in the Rouyn-Noranda Mining camp using Albert Mining's AI technology.

Ideal AI testing ground

Analysts say efforts to be more precise when finding areas to mine by using machine learning can help the mining industry accurately target commercial mineral deposits and ultimately be more profitable.

According to Auryn, the biggest advantage of a data-driven solution is the ability to extract subtle correlations across multiple datasets over a large spatial area, all while reducing human bias.

"By generating targets with deep learning, and vetting them with an experienced geoscience team, the expertise of human beings is still utilized, but is complemented by the power of the machine," according to Auryn's Chief Operating Officer and Chief Geologist Michael Henrichsen.

Henrichsen told Mining News that the junior "is always looking at innovative new methods to enhance our exploration efforts, but the concept of machine learning was introduced to us organically through longstanding relationships we have with experts in the industry.

"They saw Committee Bay as the perfect project to test their new technology because of the wealth of high-quality data we have from years of exploration on Comm Bay's 300-km greenstone belt," Henrichsen explained. "We saw it as a great opportunity to leverage that data and work with some of the best people in the artificial intelligence space."

AI delivers results

On Feb. 19, Auryn reported results from the targeting exercise using VNET at Committee Bay. Among the highlights: 12 new gold targets, including two targets that overlap Auryn's geologist-derived targets adjacent to the Aiviq and Kalulik discoveries, which are effectively drill ready. The machine learning technique also identified two important targets that create east and west extensions of the Three Bluffs deposit; multiple targets hidden beneath shallow lakes and glacial-fluvial cover; and a third structure (in addition to the Three Bluffs structure and the Aiviq and Kalulik structure) with 15 kilometers (nine miles) of strike length that offers new opportunities.

Auryn said the two new targets east and west of Three Bluffs will be advanced to drill stage, hopefully to significantly expand that deposit.

"The machine learning process is valuable because it removes bias and its in-depth analysis of our extensive, high-quality data sets outreaches the capabilities of the human brain," Henrichsen told Mining News.

"The resulting targets have brought our exploration plans into focus and have given us confidence in our emerging discoveries at Aiviq and Kalulik. In addition, the machine learning identified new targets under shallow lakes and glacial-fluvial cover, where surface geochemical sampling has not been possible," he added.