Theo Yameogo: I'm excited to bring you our latest episode of Mining Today with EY. My guest is my good friend Pradeep Karpur, who leads our AI and Data Competency at EY, and I look forward to this conversation. It's really timely. Pradeep, welcome to the show.
Pradeep Karpur: Thank you so much, Theo. I'm excited as well. Mining is very special to me as you know and so looking forward to talking about AI and share my perspectives on it.
Theo Yameogo: So, let's get right into it. Can you share how metals and mining companies can adopt AI safely and effectively? Can you also provide examples of what you've seen work so far?
Pradeep Karpur: That's a great question, Theo. It's always good to start with examples because that helps you try to map it to your particular business. When I think of use cases and examples in AI in metals and mining, I tend to classify it into two very distinct buckets. The first one focusing on operational AI. It's quite mature in that space where, when you think about asset maintenance, processing optimization, dispatch haulage, blasting, health and safety. Those are some of the mature areas when it comes to operations. And AI is heavily impacting those areas.
The second bucket is the back office. When you think about A/P, A/R invoice handling, contract compliance, fraud detection, procurement intelligence, cost variance analysis, board investor reporting and just productivity using Copilot. So, I'll break them as these two big categories.
For one of our Australian clients, we implemented an AI-driven machine learning ecosystem for rail maintenance. Optimizing your planning workflows and improving asset reliability and improving your operational efficiency. This is a multi-year program involving IoT sensors across your supply chain and making sure that you are using the insights coming out of those sensors to drive operational efficiency.
The other one, I will say, is something which I was personally involved in developing. The health and safety assistant examples. You know, especially in our large mining clients, they have operations across the world. And in each place, regulations are different, and you have health and safety manuals that are very specific to their jurisdiction. So, what we are now able to do with AI is to make this health and safety information accessible in the office, on the field for pre-planning, real-time decision-making and post-job analysis. Think about, let's say you are going into a mine in Northern Ontario, and there is a heavy hailstorm happening and someone wants to understand what precautions they need to take before undergoing a certain function. Leveraging a health and safety AI bot and have a conversational interaction with it to get those specific recommendations so that you can operate safely. That's being used heavily across the industry.
Similar kind when it comes to technical asset documents, how can you do a semantic search. The business ask here is to improve knowledge management. How can you use these search technologies, which goes through thousands of documents and gives you information at your fingertips. Imagine you're doing a large turnaround activity and you're doing a planning, having access to that level of information very quickly, where you don't have to call 10 different people to get that information. That's a game changer. Next example here. You may remember that we worked together five years ago for a mine in Africa using health and safety compliance monitoring using video analytics on CCTV. Are people wearing their hard hats? Are they putting on the right boots? Are they wearing the right gear? And if there is noncompliance, flag it right away so that predictively you are preventing an incident from happening.
Theo Yameogo: These are great examples, Pradeep. What did these clients do or what are the lessons learned from these clients that are successful in designing or strategizing and implementing these kinds of great productivity assets?
Pradeep Karpur: That's a great question, Theo. One of the challenges I see in the industry is people jump straight into AI, do fantastic pilots, and then they figure out that they can’t really scale. It gets stuck in a pilot quagmire and it's because it needs a little bit of planning. The first thing I would say is focus on high-quality, accessible, well-integrated data in mining. As you would know, the landscape, technology landscape in mining, it's not all in one system. You have multiple systems focusing on different areas, there’s ERPs, there’s plant maintenance, asset systems, IoT sensor systems, and the data is all over the place across all of these systems.
So, the first thing you want to focus on is how do I integrate the data and bring it together. Then I can start unlocking a lot of these AI use cases that I just spoke about. Second thing I would say is, how do you pick the use cases that you want to work on. Focusing on high-value, well-defined use cases, picking something very specific and economically meaningful, that's very important. And then move from pilots to real scaled workflows, don't stop with the pilot. Your pilot may be great to impress your boss, but that's not where the real transformation happens. Real transformation happens when AI is part of how work gets done. Embedded. We call it as built-in AI. You build in AI within your process, and not just a bolt-on to an existing process that doesn't work. AI is a transformational technology, and you should think about your processes differently when AI is at the middle. The last thing I would say, it's more on the human side of things. People, change and operating models matter more than technology. The program succeeds when the workforce becomes owners of this particular AI use case you're building, and not just an observer of an AI driven tool. The companies that are getting real value from AI aren't just better at algorithms. They are better at data. They're better at integration. They're better at change. AI success is much more than just a technology. It's an operating model in itself.
Theo Yameogo: I like the fact that you talk about people. What have you seen companies do successfully and what would you advise metals and mining companies for training the workforce and also things around governance, making sure it's ethically done and that the training is also rightly done so that you can do your own AI program, but you can also have the movement of the workforce participating for their own productivity. What have you seen in that space?
Pradeep Karpur: I think that's a great question. If you really think about AI today, the models are commoditized. Maybe a few years ago if we had to do AI, you need to have a lot of data scientists who can build the model specific for you. But now there is less need for those types of in-house models to be developed, because there's a lot of models from out there that you could just leverage, and you can just subscribe to it and start using them. So, what is the bottleneck now? The bottleneck is change, adoption and knowing when to use your AI, how to use AI, and how do I integrate AI into my processes. So there comes the people side of things and there comes the change management and training side of things. If you ask me what differentiates the most successful organizations compared to others is having a persona-based training program across your organization starting all the way from your board. Educating your board on how AI matters for metals and mining, right? Which is very different compared to AI for a bank or AI for a government. So, sector-focused AI education for your board and your C-suite. Starting there so that the right questions get asked to the management below. That's very important. What are the use cases? Where is the value? How do I quantify value? How do I set up my organization in this agentic world that's coming up? What kind of skill sets do I need in my organization moving forward? What is going to be my upskilling strategy for my people as AI is disrupting the work we do? Having that level of training and literacy at the executive board level is very important. Then moving below into some of the middle management, that's probably where the largest friction happens.
People are more used to relying on intuition and experience. I'm not saying you don't have to, but you need to have the right mindset to blend your intuition and experience with what the AI and models are saying. And how do you do that? In many cases, if AI is just a black box where AI is giving you an answer, an engineer will be hesitant to just blindly follow that.
If you are bringing in AI with an explainable set of reasons why an AI is actually making that decision and give the engineer the step-by-step broken-down thinking of AI that will help the engineer understand how AI made that answer, and they will be more willing to accept AI and adopt AI. So, when I think about that middle management training, we should focus a lot on what we call as responsible AI. Explainability and lineage of the data. Helping them understand the mechanics of AI. I think that's very important in getting people to start believing in it and using it and basically make their life easier.
Theo Yameogo: Yeah, that's a very good point. And I know you've been invited by many boards to come and talk about AI for metals and mining in general, here, in the US and also Europe. I think it's very important to do that education, because there's also a generation difference between board members and people on the floor. Somebody might say, yeah, it is all good, but we do not have the technology infrastructure to run this. It's too complicated. What do you say to that?
Pradeep Karpur: Yeah, a couple of years ago I would have accepted, but these days where the way you have simplified infrastructure in the cloud, where setting up an environment end-to-end from your sensors, data integration, to your models to building applications in front, which leverages these models, you can do pretty much do the whole end-to-end lifecycle without writing a line of code. These are containerized technologies that are available whether you are using a Microsoft stack or any other stack of your choice. These capabilities have now become so easy and democratized. AI has played a huge role in simplifying it. The challenge here is how do you use this technology in a responsible and safe manner?
So having those proper guardrails and how you implement your technology infrastructure is so important. So, this is where we talk about this terminology called responsible AI, which for me, especially in our industry, in mining and metals, I think it's super important because we are a safety-focused industry. Safety is part of every single thing we do. And when it comes to AI, responsible AI is exactly that. I would strongly recommend that we take those responsible AI principles to heart and focus on using AI in a way it builds trust. You build AI in an explainable, transparent manner. You embed governance into operations, put some tight controls on your token usage and have a proper reporting mechanism and try to map the tokens you're using to the value you are generating, so that you are always having value in your mind, and responsibly scale out the using of AI so that you are running a profitable, safe and successful AI operation.
Theo Yameogo: Yeah, that's very good to really grasp and is super important. Now I want to pivot to something that is somewhat newer and it's physical AI. How do you foresee the integration of physical AI in mining operations, especially when they are remote?
Pradeep Karpur: Yeah, that's a true game changer. What's different now is the way our hardware and the chips have progressed, wherein you can embed that intelligence and those computer chips in these particular dogs, and you can run localized AI models within these devices themselves. Why is it important? When we are operating deep inside a mine, in a very remote site where you do not have network connectivity, typically running AI models has been a challenge for us. That has changed now because our infrastructure has improved so much. The chips’ technology has improved so much. Our models have become so much smaller that you are now able to deploy those models into these physical devices and let them operate in areas where humans generally struggled to go into. And the way you are training those models, and they are running autonomously within that device, for me, that's a game changer.
Theo Yameogo: Yes. Now that's a very good segue into the intersection or the convergence between AI and cybersecurity and the risk of cybersecurity. What would you like to share with the audience around this?
Pradeep Karpur: Very timely conversation, especially with Mythos and everything else we are hearing in the market. Sometimes it is scary, right?
What we are seeing is that AI and cyber aren't separate topics anymore. They are becoming one combined risk and opportunity. I see there are three things happening. The first one, AI, is expanding the attack surface. The reality is AI systems themselves can be targets.
And they can also amplify the impact of a cyber issue. We just spoke about the dogs doing some critical activities. And yes, they are now vulnerable to a cyberattack. The second one is AI is making cyber threats faster and more sophisticated.
Attackers are now using AI to automate reconnaissance to generate attacks or exploits faster, scale phishing, malware controls. So, all of these are real. So, I'm not going to sit here and say that's not an issue.
But at the same time, the third one I'll say is, AI is also the best defence. AI is becoming critical for real-time threat detection, anomaly detection in OT environments, automated response and resilience. So, the big shift is this: in mining, cybersecurity is no longer just an IT issue. It's an operational risk. And as AI gets embedded into operations, securing AI becomes as important as deploying it. AI can be your best friend in defending against those cyber risks as well. I'll give you a story here. I was with one of my clients here out here in Alberta. There was an attack happening where AI agents were attacking a certain infrastructure. It was fascinating to see the defence agents of this organization getting activated and fighting against the offensive agents. And in real time, you see an attack vector and a defence vector fighting against each other and protecting your organization.
I was having goosebumps watching it, but at the same time it works.
Theo Yameogo: Wow that's cool. So, based on what you what you're telling our audience, it seems to me that the critical piece seems to be a change in mindset and how things can flow. How would you leave audiences with messages around just the mindset change piece?
Pradeep Karpur: I absolutely agree with what you're saying, Theo. To lead in AI in the mining industry, it really comes down to having a fundamentally different mindset, not just adopting new tools, but rethinking how the business operates. I will highlight five key mindset shifts. First one from being physical-first thinking to data-first thinking.
Imagine all these machines over years have generated millions and millions of terabytes of data. Let's start using them. Let's have data-first thinking. That's a mindset change. From being experience driven to data and judgment driven. The third one is from pilot mindset going and trying to 25 different little pilots. Let's think about the scaling mindset. That's when the real exponential value gets created.
Fourth one from siloed thinking within a small part of your process or one system, move more towards integrated system thinking. Think end to end across the value chain. Again, that's where exponential value gets created.
The last one I'll say is from risk avoidance kind of a mindset to a learning mindset. A digital mindset that encourages curiosity, learning and adapting quickly. I’d say these are probably the five big mindset shifts which are really important. To generate transformational value, exponential value out of AI, you need these five mindset shifts.
Theo Yameogo: That's fabulous! And on that note, Pradeep, thanks for being with us today. Was great to have you on.
Pradeep Karpur: It's been my pleasure, Theo! Thank you for having me.
