I recently gave a talk at the SMU Steel Summit. Here’s a transcript, edited for length. If you’re a glutton for punishment, you can see the full video below, including the Q&A.

Screen Shot 2018-09-21 at 10.51.49 AM

 

I’m excited to be here today. It’s a real honor. What I want to talk to you about is something I think is going to transform every one of your careers. Hopefully your one takeaway is that you need to learn something about this topic, get experienced in this topic, and then figure out for your executive teams and for the people around you how to get exposure to these new technologies.

 

The New Human + Computer Partnership

The topic that I’m going to talk about is this how humans may team with computers in a new, more powerful way. It’s something that wasn’t possible even three years ago. I like to think of April 2016 as the time in which these things became possible. And it’s been remarkably rapidly applied in the steel industry, in the aluminum industry, in the scrap recycling business, and in a lot of other manufacturing companies and consumer products companies. That’s the topic of artificial intelligence.

Just for purposes of definition: we say artificial intelligence, machine learning, and data science are all the all the same thing. I will also say that there is nothing artificial about what we’re doing and what other companies are doing this area; it’s just advanced mathematics. This is applying advanced mathematics to complex operations using supercomputers, then helping you be much better leaders in your company.

The Beginning of AI

Look at how we–most of us–became aware of artificial intelligence; think of Deep Blue, which is the IBM computer playing Garry Kasparov. This happened initially in 1996, when IBM researchers came up with the idea that they thought that they could build a computer that would be better than the world’s best chess player at the time, Garry Kasparov. So, they challenged him in 1996, and Garry won. After that things were progressing faster in the area of computers, mathematics, and data. They’d collected a lot more data on historic chess games. IBM basically rewrote the algorithms and challenged Kasparov again in 1997. This time Deep Blue won. This was seen as a seminal moment in what computers could do.

There was no way a computer could calculate every possible move that that you could make in a chess game by force, but what Deep Blue did is to figure out probabilities. It learned from all the historic chess games, it learned what happened in the past, it predicted what Garry Kasparov would do in his next moves and anticipated those moves, and then made better moves than the world’s chess champion. This was really a big deal. Why am I talking about this? What does this have to do about your business? Your business is not chess.

The Power of Probability Predictions

In fact, your business is actually a lot more complex than chess. Your company is playing something more like 10 games simultaneously, all with different rules. And your business is not just playing multiple overly complex games, it’s playing them with more than two players and without fixed rules. This is really a vexing challenge for a business leader.

What happened in 2016 is that Nvidia came out with their DGX1 computer, and the DGX1 computer is 1000 times more powerful than the fastest supercomputer in the world in the year 2000. In the year 2000 there was a computer called the ASCII Red, it had crossed the teraflop threshold of computing power, which means it could it could process a trillion floating point operations per second. That was a big deal.

How many of those computers existed in the world? There was one, and it was owned by, of course, the Department of Energy. The U.S. government was doing all kinds of spooky simulations about what happens when bombs explode. It really wasn’t practical, and of course no companies had access to this. This brings us to today. In our data center we have something that is 1200 times faster than that. Finally, we can take all the data that is that is available in very complex operations and use it to think many moves ahead and do and predict things that you couldn’t do before.

Finding Patterns

This includes sensing patterns in your operations that you could not quantify before, making more accurate predictions than ever, and generating recommendations about what moves should make in your business. This is bleeding edge stuff that will very quickly become table stakes. I will make a forecast that certainly within 10 years, maybe within 5 years, if executives don’t understand how to use these technologies they will be left behind.

The company that is at the forefront of this is Amazon and the way they figure out: what products need to be in what places, at what time, for what customers, at what price, with what recommendations, how many drivers they have, what the customer might want next–all of those things are driven by machine learning algorithms. You can feel it in the way that you interact with the company.

Industrial Data Science Meets Steel

Operations systems have traditionally been focused on two ways of modeling industrial operations; theory-based models and heuristic (rule of thumb) models. In theory based models, companies would have someone understand the physics of how your steel operations work and generate a model. These models are very slow to develop and quite static. Your environment is constantly changing, but these theory-based models do not change accordingly and are very difficult to do (but very powerful).

The second way of modeling is how most companies run. In fact, how most steel and aluminum companies, recycling companies, and aftermarket companies work. They base decisions on empirical models and heuristics. These are fixed business rules, kind of “rules of thumb” – the “way we’ve always done it.” I will tell you that the difference between averages, rules of thumb, and the future (which is very precise predictions based on the data and mathematics) is that this difference will be the difference between the winning companies and the losing companies very quickly.

That brings us to a third category, which is advanced data driven techniques. What this does is look at all of the data around what’s happening in your steel mill, the chemical composition of the steel, whether it rained in the last week, it looks at the caster speed, it looks at all these things and makes predictions about energy consumption–will there be a caster breakout, will there be a longitudinal crack in your operations–and allows you to act on these things. What’s special about these models is they are constantly learning. They learn and improve every day, every time you give the model more. Every time you give these models more data, they actually do tighter and tighter correlations and regressions. And again, that’s just math, you do very high-dimensional correlations, regressions, classifications, and clustering, then you get a superior product.

How AI and Predictive Analytics Can Help You

You might ask, “what’s the big deal?” There are many areas of tremendous value. Let’s start with energy savings. Electricity is a huge cost component. For a lot of electric arc furnaces, you can do one of two things: you can either change the production schedule and smooth it out, or depending on your contract, possibly sell electricity back to the electric grid. Just putting in this one application around electricity for prediction is saving one client more than $5 million a year – just this one application! That’s a big deal when you talk about profit per mil our and of course, this has allowed them to be incredibly successful and actually expand their operations.

This also applies in maintenance and refractory. If you look at when the interiors of these need to be replaced, typically, replacements had been done according to number of heats regardless of wear. The app we build looks at what the temperatures were when that happened as well as what the impurities were, what the timing was, and how long was it in there? Then it figures out all of these different things to actually come up with a much better approach to your challenge.

So, how do you get started? How do you actually do this? There are a lot of different paths to this. I would encourage you to dive in!

You know my favorite quote, often attributed to Mark Twain, is “the secret of getting ahead is to get started.” I think that that that applies here. None of you should feel bad that you didn’t do this five years ago, because you couldn’t have done this five years ago. This is a very exciting time and I think it’s fantastic that data science is actually happening in the metals industry. The metals industry could be one of the real pioneers in the application of these learning techniques in your business. What this is all about is actually thinking five moves ahead and saying, you need to fix this problem so it doesn’t cause these other 10,000 problems. It can fundamentally change the economics of your business.

This is the path that some of our other metals companies have taken. They start with demand prediction, or with the hedging application. Planning starts with having a good demand signal like what is the going to be the demand for, for what kinds of what kinds of skews in the future and then moving to things like inventory management: “how do you optimize the sourcing of inbound materials, both inbound logistics and outbound logistics?” There were a lot of fines that were being incurred by these steel companies, for example when a barge would come and they’d have to wait. That’s very expensive! So, actually doing predictions for when the steel is going to be ready, when should the barge arrive, will save a lot of money and is so important.

AI Thrives on Complexity

Then moving to things like production scheduling, which is enormously complex – data science is a differentiator there. The one thing about these algorithms is that they actually thrive on complexity; the more complex the better. If it’s something that the human mind can figure out and calculate precisely, whether you can do it an Excel spreadsheet, or in one of your other systems, then you wouldn’t need these techniques. But if you’ve got something that’s really complex, this can figure out precisely what the answer is; then you go into production, the production process itself, and this is where you get into energy production, production, preventive maintenance, predictive maintenance, all of those types of things, which are really exciting. And then at the end of this, you connect these things together and you run an optimization process on top of all of them and it and it will fundamentally change the economics of your business, it’ll lead to higher quality.

If you ever wondered why we get these seemingly random defects in our steel, predictive data science can figure out why you’re getting those seemingly random defects in your steel. If you’re wondering why the width keeps varying, what led to that, it can probably fix that, too. If you’re wondering why we have these random occurring incidents, even safety, it can likely do a very good job of helping predict those things. So, my call to action for you is to is to get started using these techniques! You can start small, I realize this is a new thing for people.

I understand that when we talk about math a lot of people cringe. But, I can tell you that this is really the future. If I went through and asked you how many of you are using artificial intelligence today? Probably very few of you would say yes. But the answer is: almost every single one of you are using artificial intelligence every day. If you’re using a navigation system in your car, you’re using artificial intelligence. If you’ve ever used Siri or Alexa, you’re using artificial intelligence, if you’ve ever tried to search for pictures on your phone, that’s artificial intelligence. If you’ve ever ordered anything from Amazon, well, there’s so many artificial intelligence applications embedded in Amazon that you can see the result, so I would say you’re already using it more on the commercial side. It’s time for this to make its leap over to the business side. And I think that the metals industry could take a leadership position in traditional industrial manufacturing in this exciting new area of technology. Thank you very much.

How Can We Help You?