Machine Learning with Susan Nash

VERN STEFANIC: Hi, I'm Vern Stefanic. Welcome to another edition of AAPG's Energy Insights podcast, which deals with all sorts of things that are very important to the world of geosciences. And today is going to be something that is actually very, very timely.

We're talking today-- my guest today is Dr. Susan Nash, who is the Director of Innovation and Emerging Science and Technology here at AAPG. And the topic is machine learning.

Now-- hi, Susan, by the way.

SUSAN NASH: Well, thank you. Hi, Vern. It's great to be here.

VERN STEFANIC: And I'm glad you're here. And Susan, because of your role with AAPG, you are constantly traveling. You're constantly working with emerging technologies all over the world. Actually, this is great for me, because I don't get a chance to see you very much.

SUSAN NASH: Oh, that's good. Same here.

VERN STEFANIC: Susan, welcome back.

SUSAN NASH: Thank you.

VERN STEFANIC: And machine learning, though, is something that in the past six months has become, in my life, enormously obvious. Geologists everywhere, in all the meetings that I went to across the country, wanted to talk about machine learning. And so we're going to do several editions of the Energy Insights podcast dealing with machine learning.

But we wanted to go to you to get this all started so that you could talk a little bit about what it is, why it matters. Even though it's only six months old to me, I know it's been going on for a couple of years. Is that right?

SUSAN NASH: Oh, absolutely. In fact, artificial intelligence has been around for a long time. And even here at AAPG, we've been concerned with it for at least four years, where we've had workshops and information sessions.

But I think what's interesting about machine learning is that either people feel fear or they feel excited. And I've noticed that people feel fear because they think they'll be replaced.

And I don't know if you have looked at people's profiles on LinkedIn, but I've noticed that a lot of people who previously would only say that I'm a engineer, or I'm a geologist, or I'm a geoscientist, now they say, data scientist and engineer. And one can ask why. And the reason is because those analytics are being used in every stage of oil and gas exploration and production.

VERN STEFANIC: Right. And just to back up one more step. So machine learning is dealing with taking data. And so computers are able to take data and recognize patterns?

SUSAN NASH: Oh, let me start-- OK, I'll start with the basics.

VERN STEFANIC: Yeah, help us with that.

SUSAN NASH: OK, I'll start with the basics. OK, a lot of people have heard of big data. So a lot of the machine learning works with big data.

Now, what is big data? Big data are these enormous data sets that are coming in from all kinds of information sources. They could be coming in from smart refrigerators, smart phones. They're coming in from people's choices in social media and online shopping-- just all kinds of information. And they're being housed.

And there are two different kinds of data sets. There's structured. And structured is, like, nicely organized. It's easy to process, and different algorithms and analyze, and see any patterns or categories.

And then the other is unstructured, which is kind of like historical data. And historical data might be well logs on paper. Or they could be PDFs that you do not easily scan, that you can't put into a program.

VERN STEFANIC: OK.

SUSAN NASH: And so that's the first element. And then the second, if we say artificial intelligence, usually we're talking about machine learning, and we're also talking about deep learning. So machine learning is essentially a number of different kinds of-- I'll just go kind of simply-- but like algorithms or equations that are used to process all this information. They're usually, like, statistical things.

And geophysicists have used this for years, and years, and years. And the big criticism was that it took days to process all that information and also that it wasn't necessarily tied to reality. Because what it does is it processes it and processes it. It's iterative things to try to find patterns. And the patterns need to be meaningful.

Now, if they're just simply patterns, you can have supervised learning, which has-- like, if it's face recognition, it'll have a face of a person, not a cat, for example.

VERN STEFANIC: OK.

SUSAN NASH: So your data would be like it process it and processes it, so kind of like it's going toward that thing, that category thing, that you want it to. Or if it's unsupervised, you're just looking for whatever the data will tell you, so whatever the patterns are.

VERN STEFANIC: OK.

SUSAN NASH: So those are different kinds. And then that sets machine learning. So if you think about how you would, like, face recognition, et cetera, that's good. And machine learning is often used, for example, like a well log can be, like, a face. There are a lot of the different types of activities and little data sets and things that are--

VERN STEFANIC: That are identifying characteristics?

SUSAN NASH: Teaching people how to build algorithms. Yeah, exactly.

VERN STEFANIC: OK. OK, and so then the machine can look at it. So for the industry, that is very helpful. Because I'm thinking that a lot of calculations and a lot of recognition can be done perhaps a lot faster because of this.

SUSAN NASH: Yes, exactly. And so you might be looking at it from a historical point of view. But you could be looking at it from a live perspective too, so some of the things-- like, if somebody's drilling and they want to stay in the zone, they don't want to porpoise, or whatever, they can build in some of those recognition things.

Let's think about face recognition. Anything that would be recognizable through a visual image is perfect for AI and machine learning.

So deep learning, many times, has to do with predicting, so predicting flow, for example, or predicting your behavior.

VERN STEFANIC: OK.

SUSAN NASH: Yeah.

VERN STEFANIC: OK. OK, so that's interesting. And so that helps me and perhaps others understand why so many of our members and other geoscientists across the world are really becoming intrigued and interested by this. Has machine learning become something that's real right now across the industry, or is it still conceptual?

SUSAN NASH: Oh, no-- it's been used for a long time. In fact, Devon is one example. They've been using sensors downhole. And they use that for in-production. Well, these sensors are sending information. And they can do things like register the flow rates and predict flow rates or predict equipment failure-- pump failure, et cetera-- based on patterns.

VERN STEFANIC: Well, and that is actually a perfect segue way into a reminder that, as we continue with this podcast series and deal with the emerging technology, we have episodes planned where we're going to talk to some people very specifically about some of the success stories that they've had. And I would imagine that you were aware of many throughout the industry--

SUSAN NASH: Oh, yes.

VERN STEFANIC: --specifics that have happened that people can point to and say, yes, because we had machine learning, we were better and we did something-- right?

SUSAN NASH: Absolutely. And I think that where we've seen the biggest improvements have been in things like pipeline safety or in transportation, in a production-- in gas processing and gas gathering. And also, we've seen it really, really dramatically in imaging and being able to create 3D images-- images where we integrate seismic and we integrate well log information. We integrate all these different kind of heavy things-- even geochemistry.

And so we've been able to image sub salt and also image things where we're able to see that the parallels between, say, one side of the Atlantic-- and that's in deep water-- and then the Africa side of the Atlantic. And actual fields have been discovered, because they've been able to image and see those patterns.

VERN STEFANIC: And that upstream application of machine learning is what I think is really intriguing, maybe, for most of our members right now, because that's where we deal. So the excitement and the potential is clearly there. But you spoke in the opening that there's also two kinds of people. And one of those kind of people fear any kind of change--

SUSAN NASH: Yes.

VERN STEFANIC: --that's coming. And I guess I would set up-- I want to hear your comments on all of this. My setup is that, as I traveled around the country in the past half year and our members mainly-- but geoscientists-- would talk to me about machine learning, for every one person that talked about great potential, there was maybe somebody else who talked about-- we don't know what this is going to do for our profession and our career.

SUSAN NASH: Right-- and there are two things that I would look at. A-- the people who are afraid or nervous say, well, OK. We know that jobs are being eliminated for the traditional geologist or engineer or geophysicist. Why? Because people need to do more with less. And the well planning, the drilling, et cetera-- geo-steering-- can be automated. So that of course, they're going to be resentful.

However-- however-- the same companies that are laying off traditional geologists are hiring, like, maybe 20 to 50 hybrid data scientist engineers. And I'm not saying that every person needs to go in and be able to set up their own Hadoop network for big data or be able to program extensively in Python. But they need to know what it is. And so then they work in a team. And then they need to know where to look.

For example, Python is a coding language-- a programming language-- that people use. Well, it's not like you have to code everything from scratch. You can go to a repository. And it's open source. So it's like, there are a lot of free cookbooks.

VERN STEFANIC: OK.

SUSAN NASH: Yeah.

VERN STEFANIC: OK-- great. Well, let me ask you. Geographically speaking, are there parts of our society or parts of the world-- I don't know-- where machine learning is more prominent than others right now? I mean--

SUSAN NASH: Oh, yes.

VERN STEFANIC: --the application is right there.

SUSAN NASH: OK-- yeah, we've got some hot spots. One is Houston. Just off the top of my head, I can think of, say, 10 startups that are doing really cool things with creating work benches and things. And a lot of people are doing things with voice activation too. So they're using it not just for analysis but for actual productivity-- improving productivity. So I see a lot of people in Houston.

And then they usually have, say, some team members in India, doing a lot of the coding. So that's pretty cool. And then another spot is Austin. And Denver has quite a few.

VERN STEFANIC: OK-- and actually, I'm looking at this in three different ways. What about the veteran petroleum geologist who's been around for a long, long time and is used to doing things a certain way? What about the mid-career petroleum geologist who's lived with technology, maybe, all of their lives? And then you have the students and the new hires that are just coming in.

Are there different approaches that they should be taking about this?

SUSAN NASH: Oh, yeah. I think that teams are the best. And I really like the whole idea of yin yang. So the field experience is just absolutely indispensable. And knowledge of what the actual rock looks like and what it does-- and also, we're digitizing a lot of core repository-- things like that. So the person who's been doing a lot of work with cores, with the field, the structure, the pore architecture-- all that-- can work with a person who can do the programming.

And then the person in the middle could be-- I just think-- all three, because you have a bridge. And knowing a little bit about engineering as well as geology always helps.

VERN STEFANIC: So let me ask you-- so the era of the loner working alone-- we may be in trouble for that person. Is that a true statement? Or maybe I'm overstating.

SUSAN NASH: I don't think that that person, the loner, ever existed. Because I know that, even in the small shops, they would go to the log libraries and get information and then get the books and things. Now, you can definitely work from your cabin in Colorado. But you need to have a good Wi-Fi. You have good Wi-Fi and be able to work with the cloud.

And that's another thing too. I will say that the things that made things possible in the last three or four years-- cloud computing and also high-processing chips. But I would say, be able to team and use WebEx or GoToMeeting or whatever. And team that way.

VERN STEFANIC: Yeah-- yeah, actually, the advances in computer technology is what's made all of this possible, right?

SUSAN NASH: Yes.

VERN STEFANIC: And especially cloud--

SUSAN NASH: Yeah.

VERN STEFANIC: Yeah.

SUSAN NASH: But there's a little cautionary tale too. And I think you were talking about people having fear or skeptics. I think that one of the big problems with any of the models is that it privileges what you already know, what's already there, and the patterns that have already happened. And it doesn't-- like, if you're doing a heat map, in risk assessment, and you aren't incorporating something-- like, the red red is high impact, high probability. But then there's always the low probability, high impact.

And it's almost like, OK, there's never been a tornado in So-and-So. So I'm not going to plan for it. So you have to be willing to introduce things that have never happened or are potentially weird in there. Otherwise, it's just replicating what we already know.

VERN STEFANIC: Right.

SUSAN NASH: And it's not letting you grow.

VERN STEFANIC: Right.

SUSAN NASH: And it will never find anything, unless it's just a complete analog.

VERN STEFANIC: There's something very special and unique that makes a petroleum geologist very valuable.

SUSAN NASH: Mm-hm.

VERN STEFANIC: It's that experience that they can come and they can look at patterns and see, maybe, not what everybody else is seeing, but something special and unique. And I just wonder how machine learning is going to fit in. Are they going to take over that skill for us?

SUSAN NASH: I think that that's where there's going to be-- it's very challenging. Because again, how do you transcend what's already in your paradigm and what you're using as a paradigm? And how do you break out? And so I think that machine learning has that potential. But it almost has to do it in ways that an expert needs to, like, collide it with other data sets or make something imitate something else, just for a creative thought, and using a lot of mind mapping-- things like that-- and then, also, maybe sharing or doing things that you would do, like imaging for medicine. Do it.

At the same time, one of the big critiques of geophysics and using analytics-- it's like, OK, that looks really pretty. But this could never happen in nature. But it makes a pretty pattern. So that's where you get the experience of the person who's been out there in the field. And I would say that what it does is it makes it more important for a geologist to do more field work.

VERN STEFANIC: Oh!

SUSAN NASH: Yes-- because you can then start taking information. You could do photographs with drones and map the outcrops. But then you have more reality to compare the subsurface with. And then especially, if you can connect it with core and then say, OK, this is similar to this and this and this-- so you have a broader sense.

I don't know if you've seen pictures of fields as they're developed or maps.

VERN STEFANIC: OK.

SUSAN NASH: When they first start out and there's two or three wells, the whole thing is like-- oh, this big simple anticline or whatever. And then the more information you get, the more complex it is. And so we have the computing power to deal with or accommodate complexity. But now we need to have appropriate knowledge to propose that type of complexity.

VERN STEFANIC: So you're saying, as good as machine learning has become, the individualist factor is still going to be necessary.

SUSAN NASH: I believe so-- yeah.

VERN STEFANIC: You were talking just a bit, before we went on camera, about universality.

SUSAN NASH: Oh-- universalizing one's experience?

VERN STEFANIC: Yeah.

SUSAN NASH: OK-- OK, so here's another area where people are fearful and geologists are fearful. And it's a phenomenon that happens in the industry. If you have a lot of experience in the Utica, for example, or the Marcellus-- you're in the eastern part of the US, are you likely to get hired in the Permian or in the Bakken? And the answer is, generally, no. Because they're not clear analogs.

So one of the things that machine learning can do-- and with skill sets-- is it can give a person additional skills and analytical skills. So it doesn't matter where they worked, it can apply everywhere. And I'm not sure if it would go as far as to say carbonates and sandstone. But I don't know. I just think that it gives people tools to be useful everywhere.

VERN STEFANIC: Well, that's very encouraging, because that addresses specifically what some people have as a fear-- that they're going to be replaced, that they're not going to have the opportunity. But maybe we have been thinking too small when we have that thought. Maybe we're trying to define the future, in terms of what we have now, rather than seeing the potential of what's ahead.

SUSAN NASH: Exactly-- exactly. And I think too, as we come into situations where we're like, oh, our success is our enemy, because we have over-supply, we'll have to figure out ways to work with well planning, especially in horizontal, where they have just multiple wells going all at the same time-- developing that play. And then there's a steep decline curve-- just really coordinate and get the most to minimize the decline curve and then get the most out of the each well.

But plan it, so that you don't kill the market with over-supply.

VERN STEFANIC: Right-- yeah, exactly-- exactly. Susan, we will continue to talk more about this in weeks to come. And I know you're going to be back with us, talking about it. Let me ask you-- just before you go though, because of your role with AAPG as director of innovation in emerging science and technology. You're already active in AAPG. Because of you, it's already active in this area, right?

There's some opportunities?

SUSAN NASH: I'm happy to say that we're doing all kinds of things. In the last four or five years, we've had workshops that've included deep learning and machine learning. But then we're also having things like hackathons. People can hack into Python code to predict flow, et cetera-- better predict flow. And we also have workshops.

In fact, in January of this year, we have a two-day geosciences technology workshop on machine learning in the oil industry-- just upstream-- also used incorporating with seismic and also in selling and decision making-- financial decision making. The last two years, we've had something called U-Pitch, where we help startups and companies at the commercialisation phase find investors or partners. And the majority of those are in analytics.

And that's just the tip of the iceberg. We've had special sessions at AAPG, at ICE, URTeC. It's just, we're really, really embracing it.

VERN STEFANIC: Right-- URTeC will be coming up. And then ACE this year, the Annual Convention for AAPG, is going to be in San Antonio--

SUSAN NASH: Yes.

VERN STEFANIC: --in May. So if you want to know something about machine learning, that would not be a bad place to be.

SUSAN NASH: We have a course. And we'll also have lots of sessions.

VERN STEFANIC: Well, OK-- thank you.

SUSAN NASH: And U-Pitch.

VERN STEFANIC: Yeah-- so we were able to pitch that. But the important thing is, machine learning and technological advances are revolutionizing the industry and creating new potential for our geoscientists.

SUSAN NASH: Absolutely.

VERN STEFANIC: That's great. Susan, thanks for being with us today.

SUSAN NASH: Oh, thank you. It's been a pleasure.

VERN STEFANIC: Thanks for joining us. My guest today has been Dr. Susan Nash, Director of Innovation and Emerging Science and Technology for AAPG. We hope you'll check back to the AAPG website, because we're going to be continuing our conversation of things that are important to petroleum geologists all over the world, with energy insights. But for now, thanks for listening.