AI has revolutionized countless industries, yet it has not yet discovered a Tier 1 mineral deposit. What are we doing wrong, and what is the opportunity to get it right? In this article we dive into where machine learning works, where it doesn't, and why geologists are critical to the process.
Learning from those that got it right: Digital Advertising.
In digital advertising, companies like Facebook and Google have achieved remarkable success. They utilize billions of ad-clicks and purchase outcomes to train their models. Each time we click a Facebook ad and make a purchase, their ad targeting becomes smarter. This is because they gain another positive outcome to refine their algorithms.
For example, consider how Facebook predicts who is likely to click on a real estate ad. Their algorithms identify people whose life events suggest a move. They track families expecting a second child or preparing for a job change. By understanding these life events, Facebook can anticipate when a family might need a new home. The key here is access to massive datasets of positive outcomes.
The challenge of training data availability in prospectivity mapping.
In mineral exploration, we face a different reality. We don't have millions of Tier 1 deposits to train our models on. The processes that lead to large mineral deposits are complex and data on Tier 1 deposits is scarce. Machine learning alone can't solve this problem due to the limited training data available.
How can we use machine learning in prospectivity mapping ?
This limitation is also an opportunity. We can leverage mineral systems to bridge the gap. Mineral systems are the science of understanding the critical processes that must occur for large mineral deposits to form and the clues in the geological data that reveal them. Geologists use these systems to map out where these processes have taken place. Mineral Systems are still in infancy in mining industry, their analogue, Petroleum Systems have been used now succesfully for over two decades by oil & gas explorers leading to 1/3 success rate.
Usage of Mineral Systems is limited in the industry. Exploration process typically starts from mineral occurences, instead by identifying locations where all processes required to form Tier 1 deposit aligned in time and space.
Combining geologist and AI.
Mineral systems allow us to ask the right questions about the geology. Machine learning and AI can then help answer these questions. How?
1) For instance, large language models like ChatGPT can help geologists quickly access relevant geological papers containing important geological data.
2) Hyperspectral data and AI workflows using computer vision can identify specific rock formations crucial to our prospectivity models.
3) We can extend our bedrock map interpretations into areas under the cover using advanced algortihms. This can help us to extend our search space of exploration and reduce costs of under the cover exploration. 80% of Australia, 70% of Arabian Peninsula and 60% of Finland are under the cover - all with potential for the next Tier 1 discoveries.
Fig.1 AI Bedrock map in Finland in the area with higher till cover vs original bedrock map. Higher level of detail can be extracted from AI Bedrock maps.
4) We can now compile and visualise regional geological datasets with billions of geospatial data points and query them interactively. This allows geologists to querry the data in seconds vs days or weeks and unlock new insights not accessible with traditional GIS and exploration tools.
Fig.2 Equivest Platform: Example of interactive filtering of 2Bln geospatial data points for mafic rocks, elevated Cu and strong EM signal in 15 seconds. Data Cube has over 80 layers of geological and geophysical data.
Building Mineral Systems and AI powered prospectivity maps.
Once we have identified critical process of mineral systems and mapped their geological proxies either using traditional or AI based workflows. We build prospectivity maps showing where these critical processes have occurred. When we score these maps together, the areas where all processes align for the formation of large mineral deposits become clear. This approach leverages the best available science and technology, curated by geologists and supported by AI workflows, to map out these critical processes.
Optimizing Mineral System based maps with Machine Learning.
The final step is to run prospectivity machine learning models on these layers and use existing deposits as training dataset. This method combines the expertise of geologists with the optimization power of machine learning. By identifying discrete patterns across multiple datasets, we can pinpoint the most prospective areas.
In conclusion, while AI alone hasn't yet found a tier 1 mineral deposit, it has immense untapped potential when combined with the expertise of geologists and the science of mineral systems. By leveraging both human and machine intelligence, we can make more informed and accurate predictions, driving innovation and success in the field of mineral exploration.