University of British Columbia (UBC) professor and engineer Ilija Miskovic, who was born in the former Yugoslavia, has an easy way to distinguish the oft-confused domains of artificial intelligence (AI) and machine learning. Think of them as Russian dolls, he suggests. Artificial intelligence, a branch of computer science focused on creating intelligent machines that can emulate human beings, is like the biggest wooden doll. Inside that is machine learning, which involves the ability of machines to continually improve performance with minimal human oversight. Deep learning, a subset of machine learning, is the third, smallest doll.
In a conversation with CIM Magazine, Miskovic explained how AI is currently being applied to mining and where it might go in the future. The implications of AI to the mines and workforce of the future could be profound, he said, but humans will not become obsolete any time soon.
CIM: What was your career path leading to your current position at UBC?
Miskovic: The area of Yugoslavia I was born in had, by the time my schooling was complete, become part of Serbia. I have degrees in process, energy and environmental engineering. In 2007 my wife convinced me to go to the U.S. and continue my education. She told me, “Pack your stuff, we’re leaving in two months.” I came to the States and started my PhD in mining engineering at Virginia Tech, with a focus on geological carbon sequestration and enhanced coalbed methane recovery. I joined the Norman B. Keevil Institute of Mining Engineering at UBC in 2016. In addition to my academic and research positions, I also spent 11 years in the power industry. I worked as a senior power systems designer and consultant on over 60 projects in power generation, oil and gas, and mining sectors worldwide.
CIM: What got you interested in AI and machine learning?
Miskovic: While at Virginia Tech, I was working on SECARB’s (Southeast Regional Carbon Sequestration Partnership) field demonstration study in the Central Appalachian Basin. My work was focused on designing an integrated suite of sensing technologies for deep-well, near-surface, and surface monitoring, and their implementation at a remote test site. During the design phase of the project, we had to analyze tremendous amounts of unstructured geologic and production [coalbed methane] data, and then come up with the safest way to inject and sequester CO2 into the subsurface. So we had to [find ways to] correlate, analyze and make sense of very large heterogeneous data sets in order to predict the behaviour of the subsurface in advance of the actual CO2 injection. That’s when I started exploring and using data mining, machine learning and predictive analytics in my research.
CIM: How are these technologies most commonly applied right now in mining?
Miskovic: Machine learning and AI can be applied from the beginning [of mining] to the end of the mine life-cycle, from prospecting to production to closure and mine reclamation. For example, we can use machine learning to answer questions about where to explore and where the best areas to target are. We can train our algorithms and computers to automatically identify rock faces or other salient features we are looking for, then merge different types of structured and unstructured data to create a fuller picture of what is happening on the surface or in the subsurface, all of which will allow us to narrow down an exploration area, analyze mineral prospectivity and ultimately identify the locations of high-grade ores. AI is also used within the domain of smarter planning and operations where its advanced master data management, workflow management and predictive analytics capabilities are used to analyze different types of data in real time and optimize short-term planning, scheduling and transportation.
Related: Loader monitoring system employs deep learning to keep operations running smoothly
CIM: What are some specific examples of how it’s used?
Miskovic: Perhaps, the first and best example of how machine learning and smart machines are applied to mining is how Rio Tinto is using a fleet of driverless trucks in Pilbara, Australia. They’re running their whole operation remotely from the control room in Perth, which is 1,200 km away from the site. To run those trucks, you need to analyze enormous amounts of real-time data collected by embedded sensor systems and learn about the behaviour of the machines to enable predictive maintenance and maximize efficiency. It’s been about 10 years since they started using driverless trucks, and over this time they’ve seen about a 14 per cent improvement in terms of operational efficiency, mainly because those trucks can be monitored and run 24/7.
Another, more recent example is Goldcorp’s collaboration with IBM on their Red Lake project in Ontario, where they are attempting to train IBM’s Watson cognitive platform to analyze huge amounts of unstructured historic data and identify the best regions to mine within an existing brownfield project area. A lot of that work is still confidential but, if successful, this project will definitely have a great impact on the mining industry.
CIM: How useful is AI and machine learning when it comes to interpreting vast data sets generated by satellite imagery, particularly in places difficult to access on the ground?
Miskovic: Using machine [deep] learning-based image classification is the most developed field of AI right now. When it comes to exploration and interpreting satellite imagery, we are collecting huge amounts of data, and every day our spatial and spectral surveys are getting higher in resolution and more accurate. The downside is that with these improvements our data sets are becoming bigger. When it comes to prospecting and mineral identification, it would take forever to target and analyze the pixels with the highest likelihood of occurrence of a specific mineral. By applying AI and high-performance computing, we can expedite our analytical workflows.
CIM: What is exciting to you in terms of future applications of AI to mining?
Miskovic: In the mine of the future, everything will be digitally connected. I expect this connectivity will allow companies to monitor their operations centrally. A company might have one team in Canada, one in Europe and another in Australia, but they won’t have all experts in all places. They will collaborate in an immersive visualization environment, where all exploration and operational data will be shared, consolidated and analyzed. This virtual environment will be fed by hundreds of thousands of embedded sensors and data streams in real time, including market data. By adding AI as a top layer, companies will be able to gain strategic insights from their operational data, run multiple “what-if” scenarios based on market conditions, and make more informed value-driven decisions.
CIM: In the future, what are the workforce implications of AI and how do we prepare for that now?
Miskovic: There is a deep-seated fear of AI and automation among many people in the mining industry, especially when it comes to their job security. An increasingly popular concern is that robots will completely replace ordinary workers. There is a grain of truth to this because mining operations have a high technical potential for automation, especially when it comes to jobs that involve activities in hazardous environments. Also, AI and machine learning can significantly improve and expedite mineral resource estimation, and mine planning and production activities. Who knows, maybe within the next 10 to 15 years mining companies could employ more data scientists than geologists. We definitely need to think about the workforce of tomorrow and how to get ahead of the curve on this. In my opinion, the first obvious step is to expose the workforce to new technologies and adapt mining engineering education as early as possible.
CIM: What skill sets and occupations will be needed to run the mines of the future as this technology continues to advance?
Miskovic: There will be more data scientists, statisticians, geo-statisticians and machine learning specialists. Machine operators will also need to understand new technologies. So, we will need to upgrade the curricula taught in mining engineering programs to reflect the changing skill requirements. To stay competitive, mining engineers will need to diversify their skill sets and learn about topics such as robotics, edge and cloud computing, and augmented and virtual reality. They will also need to understand how predictive analytics works as well as how to develop, deploy and troubleshoot Industrial Internet of Things systems.
CIM: We talk so much about the potential, but what are the limitations of AI?
Miskovic: We cannot expect that AI alone will do anything and everything for us. AI and its spectrum of technologies are not really useful without combining them with technologies from other emerging fields, like the Internet of Things, virtual reality, etc. So by combining these technologies, we will get the most accurate and timely information about the [mining] system performance and proactively respond to operational challenges. Another challenge with AI is if algorithms are allowed to roam too freely through all the data we have, they will find only insignificant patterns. Today, finding patterns in data isn’t that hard but finding those that work reliably in real-world mining operations is. If we automate the process without careful supervision by geologists, mining engineers or data scientists, algorithms will not be that effective, or find and tell us anything meaningful.
