The Exploration Problem
Exploration in the mineral resource industry has become increasingly difficult due to the easiest and most accessible deposits being found and extracted over thousands of years. Traditionally the first metal deposits to be mined propagated to surface, making their discovery relatively easy due to erosion leaving small fragments of the metals (ex. gold nuggets) spread across a wide area. Geologists or prospectors merely had to stumble across these fragments, recognize their significance, and begin a geological campaign to both trace them to their source and to assess the extent of the metal deposit. Should the deposit be significant, mines were developed to extract the resource.
Today, however, the industry’s ability to discover profitable deposits has become more of an art form where geologists use their understanding of an area’s formation to create a hypothesis for the potential locations of metal deposits. The surface clues upon which these hypotheses are based often no longer include trace metals, which may still be buried. Occasionally a geologist’s hypotheses can result in exploration targets, which (if convincing enough for investors) are then explored using costly drilling programs. Overall, the success rate of exploration programs today is very low – equating to roughly 0.5% in greenfield areas, and about 5% in brown field areas.(GoldSpot Discoveries, 2018) This equates to hundreds of millions of dollars of wasted exploration funds every year.
Machine Learning at GoldSpot Discovery Inc.
GoldSpot Discovery Inc. (GoldSpot) is a Canadian based mining technology company that specializes in applying technology to mineral resource discovery and modelling. This work is completed using complex data sets consisting of drill core information, geophysical and geochemical data, geological fault logging, topographic and satellite imagery, and more. GoldSpot has traditionally been contracted to use this aggregated information to build a cohesive picture of a mineral anomaly upon which it’s extent and properties are extrapolated.
Recently, GoldSpot has expanded in to machine learning techniques to increase the efficiency of target identification for gold deposits. Partnering with a number of small Canadian gold mining companies GoldSpot has built a machine learning program that uses data available to the mining companies to identify gold exploration targets.
GoldSpot has had promising preliminary results using the existing data on deposits to identify existing gold deposits in the Abitibi geological region of Quebec. The algorithm was able to correctly identify 86% of the existing known deposits, and even identified previously unexplored targets. (GoldSpot Discoveries, 2018).
While GoldSpot has started with gold deposits, the technology is certainly applicable to many different types of metal deposits as most have anomalistic patterns that can be used to identify them. However, the nature of these signatures can change dramatically by region, depth, and metal composition.
Short-term Challenge – Data Availability
Data on existing deposits is typically privileged information available to the mining company that owns the lease and mineral rights that the deposit resides upon. Controlling and understanding that information is a strategic advantage to the mining company, which means these companies are hesitant to share the information for fear of their competitors benefitting and outperforming there exploration programs.
To date, GoldSpot has only been able to partner with a number of small mining companies with interests in expanding their own operating assets in a brownfield setting. This data has also been localized to specific mining regions of Canada – which inherently has its own unique signatures due to its geological history. This introduces bias to the model and could make a global rollout difficult.
GoldSpot’s goal should be to greatly increase the information that have access to. This may be possible through a partnership with a major miner of the likes of Barrick Gold Corp., Rio Tinto, or BHP Billiton to name a few. By partnering with one of these organizations GoldSpot could exponentially increase their access to data.
However, if GoldSpot is not in control of the information they require, can they compete in this field or will they be replaced by large mining companies developing their own in-house machine learning programs?
Long-term Challenge – Greenfield Exploration
Given the localized nature of mineral deposits it is very possible that GoldSpot’s machine learning is only suitable for brown field opportunity screening in regions that already have significant mining activity, and thereby deposit data. However, to be a truly game changing development the algorithm would need to be able to function in previously unexplored regions that has unique aspects to its signatures.
Given the localized nature of mineral deposit and unique geological histories, can a machine learning algorithm ever be reliable in green field applications?
If so, we may be on the cusp of an innovation that won’t only improve capital efficiency in the mining industry, but also the environmental footprint of exploration activities and the ability of developing nations to discover and develop their own mineral wealth.
GoldSpot Discoveries. (2018). Homepage – GoldSpot Discoveries. [online] Available at: https://goldspot.ca [Accessed 12 Nov. 2018].
Marr, B. (2018). The 4th Industrial Revolution: How Mining Companies Are Using AI, Machine Learning And Robots. [online] Forbes. Available at: https://www.forbes.com/sites/bernardmarr/2018/09/07/the-4th-industrial-revolution-how-mining-companies-are-using-ai-machine-learning-and-robots/#64bf0e3497e5 [Accessed 12 Nov. 2018].