19 Sep The relevance and impact of AI in mining
by Francine Harris
Artificial intelligence will be a hot topic at the upcoming GMG-CIM Edmonton Forum on October 2-3. It was also the focus of a lively panel discussion at the GMG Sudbury Forum last June. Panel moderator Marcus Thomson, UDMN commercialization manager at CEMI, introduced AI as “a subject that we are all very familiar with in our daily lives,” citing as examples Amazon’s Alexa and Google Home. AI has different applications in mining, and the mining industry must consider how it can add value and meet their specific needs.
The panel included AI experts and mining industry leaders: David Dickson, Global Industry Strategy Consulting Leader at IBM, Larry Johnson, VP and Technical Director at Object Management Group (OMG), Michelle Ash, CIO at Barrick Gold and GMG Chair, and Larry Clark, COO at Newtrax.
The panel was first asked to describe what AI is and its relevance to them.
To Dickson, AI is about enhancing the understanding of information’s context. AI can receive and structure vast quantities of data to provide a very full and accurate understanding of the environment that can lead to “better ways of working.”
AI has been traditionally correlative, Johnson explained, but it has transformed in the past five or six years to incorporate machine learning. He described this transformation with the example of gaming, describing how learning paradigms can make computers difficult for human players to beat. This learning capability is valuable. In the next five to ten years, he predicts, “AI is going to become much more relevant.”
Ash recommended two books would help those, like herself, who do not come from a data science background learn about AI: Predictive Machines, which is about the economics of AI and The Sentient Machine, which explains various AI concepts. For her, AI adds value through its predictive capabilities that can be applied to a variety of processes, such as maintenance planning, process control and geology.
Clark described AI as a useful tool that “allows us to match patterns and recognize patterns in multi-dimensional data sets that have the richness of contextual information.” As for AI’s value, he said that “the horizon is quite open,” noting some “early wins” in their use of AI for predictive analytics, traffic control and motion metrics.
AI in practice
Mining leaders Ash and Clark were asked about the approaches they are taking to get buy-in for AI initiatives in the mining industry. Ash explained that theirs are business case related. For example, their business case for predictive maintenance is a traditional one that presents the current maintenance costs and how they believe predictive capabilities will reduce those costs. As for “more futuristic” use cases, they start small by using technologies that are low in technology and innovation readiness levels, basing their business cases on what they think the technology’s future will be. At the moment, they are focusing on discerning what will and will not work before making large expenditures.
Clark added that it is essential to get support from those who are progressive and eager to innovate. Having support at both the executive and operational level is ideal to demonstrate how value is delivered. He further emphasized that encouraging openness to new technological changes is crucial.
Johnson and Dickson were asked to describe the best practices in other industries that may enable AI in mining. Johnson explained that AI initiatives depend on semantics. Those involved must come to a consensus on the universal meaning, or ontology, of concepts, which can be challenging because concepts do not always have universal meaning. To exemplify, Johnson explained how “pond” has two translations in French. Computer languages, too, have similar translational challenges. These challenges all need to be considered when developing a vocabulary so that meaning is consistent.
Dickson also discussed semantics. He explained that the oil and gas industry has been using the ontology of the W3C standards of semantic technologies, a set of standards for describing and organizing consistent and contextually-driven data. He stressed that large-scale interoperability needs to be supported by the data and by standards for managing it. AI requires consistency if it is going to provide useful information. Ultimately, continually working to optimize vocabulary, analytics, and machine learning builds a structure for important data architecture that uses the best possible data for its context.
Issues surrounding shared and proprietary data were also raised. Dickson described IBM’s C-Suite study that, in 2015, found that C-Levels felt new competitors were the greatest disruptive factor in their industry. This year, innovative industry incumbents were the greater concern because data is mostly proprietary. In response, Ash made a case for data sharing in the mining industry. Having a fleet of only 500 trucks, they did not have enough failure modes to enable predictive data. This issue could be resolved by sharing data with others in the industry, in turn, providing all parties with better predictive data capabilities.
AI and people
As the discussion turned to the relationship between humans and AI, it became clear that this relationship is complex and interdependent.
When asked how to know if what the computer provides is accurate, Dickson explained that AI provides a list of ranked answers based on semantics and evidence. Then, it is up to the people involved to validate them.
“You’ve got to understand what is going on,” Ash added, “you can’t just blindly let any of those tools calculate stuff.”
Clark, too, emphasized that “AI is a tool and not a panacea solution” and that people need to make judgements. He further noted that those involved should be aware of the possibility of not only making mistakes in the inputting process but also of introducing biases into the data.
“So, I guess the clear message at the moment in the world of AI is that this isn’t a closed loop system, this is something that is augmenting the capabilities of the expert,” Dickson added, “particularly in industry, heavy industry, that loop is always closed by an expert.”
AI, because it relies on human input at a variety of levels, differs from autonomous systems. Ash explained that automated mining equipment, for example, does replace at least part of a human’s job. AI, in contrast, “is augmenting people’s decision-making, rather than replacing it.”
The issue of AI and health and safety was raised by an audience member involved in research with Laurentian University using a deep learning system for data analysis to examine miner mental health. A Chilean man involved in mining approached him and told him that this research is very important in his context because many Chilean workers do not have a strong reading ability, so forms and surveys were frustrating. Having other options, like the option to mine insurance data or other human factors data, would be very helpful for them..
Ash responded that they have been working on similar health and safety initiatives to see if data can predict employee health and well-being. Though these initiatives have the potential to be helpful, it is also necessary to be aware of the security regulations that are involved with using people’s private data.
Ash also stressed how important it is to “really talk to people about the impact of AI or the impact of automation.” Many have anxieties about job loss. In the short-term, AI and automation are not cutting jobs and, though jobs are admittedly changing, it is happening at a slow enough pace to allow for reskilling and reorienting. Though job prospects may decline in the long-term, the workforce is also declining in developed countries with aging populations.
Developing the ecosystem
The discussion then turned to how to improve and grow the AI ecosystem. Clark proposed that there needs to be increased awareness. AI needs to be demystified in a way that builds confidence in the technology. “We need examples of success,” he suggested, “before people actually get on board.” That said, the ecosystem is growing very quickly. Failure, too, he explained, is necessary to “learn from our mistakes” and “improve our system”.
Clark and Johnson both also stressed the importance of not overhyping AI. Johnson explained that people’s unrealistic expectations for AI resulted in a period in the 1980s – known as the AI Winter – where advances stopped.
When asked what is being done to grow the ecosystem, Dickson answered they have been doing this by “working in a collective way” and taking a “cohort or consortium-based initiative in that early innovation stage.”
To this, Ash added that they must also ensure that “juniors are also able to access these types of ideas and concepts and technologies” so no part of the industry is left behind. From a supplier perspective, she added that growing small and medium-sized businesses can help develop a healthier ecosystem in the long-term.
The future of mining
When asked what the future of mining would look like in 10 years based on the impact of AI, Dickson and Ash both responded by saying it is safer. Ash added that it is cheaper and less risky. Clark suggested that they may be starting to see improved find rates, productivity and capabilities, but thought that 10 years would perhaps be unrealistic.
Johnson concluded: “The future of mining is in the groups we collaborate with.”