Velkommen til LØRN.TECH - en læringsdugnad om teknologi og samfunn, med Silivja Seres og venner

Silvija Seres: Hello, and welcome to LØRN. My name is Silvija Seres, and this is a podcast with Telenor. Our topic today is going to be artificial intelligence, different kinds of predictive models, and my guest is Johannes Bjelland, who is a senior data scientist at Telenor group research. Welcome, Johannes.

Johannes Bjelland: Thank you.

Silvija: Johannes, you have super exciting projects at the moment. I just overheard you at a breakfast seminar talking about the use of AI to optimize power  and performance of electricity grids and mobile networks.

Johannes: Yeah, on the mobile network that we operate in Telenor.

Silvija: Right. And then you also can talk about using mobility analytics to understand how people move, and even things like poverty analytics. Understanding the causes and correlations behind people's impoverishment. Super exciting.

Johannes: Yeah, we had a big rain show project from ranging from the, first one is maybe more business, but then last year we actually did some more Bigdata for social good. I guess you can call it. To try to see what can we do beside that those topic that are directly-

Silvija: What drives prosperity, and the opposite of it.

Johannes: Yes, and actually benefit developing countries with analytics so we can help people have better lives.

Silvija: Very cool. We are going talk about how you do this, and what data science is for Telenor. Before we do that, would you mind telling us who Johannes is and what drives him?

Johannes: Yeah, my background educational wise is physics. So kind of engineering physics. Been kind of a geeky background. I've always been interested in technology. My gift for Christmas was a chemistry play tools and stuff like that.

Silvija: Don’t do this at home, but you still did it?

Johannes: Yes exactly. So always been into technology so I started in Telenor after being some two years as a physics teacher. Then into business and especially into the data field. 

Silvija: So, you have a master’s degree in physics?

Johannes: Yes, that’s right. 

Silvija: And you go over to do mobility data science. I’d love to hear more about your experience of that transition. I want to ask I little about a little personal hobby. You mentioned you do love music and collect LPs.

Johannes: Yes, good old vinyl records. So, I think I have like 6000 or something.

Silvija: Wow, you listen to them as well? 

Johannes: Yeah. 

Silvija: Why?

Johannes: Yeah, something about having, or listening to music or the physical records. Or the music is the same as on Spotify of course. But owning that piece of vinyl. Hunting it down and buying it, that’s kind of a collector’s passion

Silvija: That’s a different love?

Johannes: Yes, different passion all together than to just browse on the internet. Hehe.

Silvija: I think it is a really important kind of signal on what you think related to the virtualization of everything. Because I believe that some things are still good in their physical forms

Johannes: Yeah, and I read now a couple of weeks ago that vinyl sales are actually increasing, on CDs of course are…

Silvija: Driven solely by Johannes

Johannes: Yeah, hehe. But yeah, owning the music and actually having a physical, tangible thing that is really valuable.

Silvija: I’ll tell you one of my biggest regrets. I moved a lot. And at some point, I decided I need to minimize my belongings. I didn’t minimize my clothes; I minimized my cd´s. At first, I threw away all the covers, and I put all the CDs in a big, you know, one of those leitz albums, and then I lost the album at some point. And I really miss that, I never missed any piece of clothing. But I miss the feeling of going to the shelf and looking at the cover edges and deciding what mood I’m in.

Johannes: Yeah, exactly that’s what you missing. Now with Spotify, you have a playlist, and you end up just playing the same playlist.

Silvija: Someone else telling me what mood I’m in.

Johannes: Yeah, but you act like you’re safe. Physically just going down to the basement and finding an old box of records you haven’t seen for a while. That’s really something.

Silvija: Reminds you of who you are, I think. So, a physics guy, with master’s degree, couple of years as a teacher and then you go into Telenor. And you go straight into Telenor research and you decide you do Data science.

Johannes: Yes.

Silvija: How did it start, and when did it start?

Johannes: Long time ago now, ten-eleven, eleven years or so.

Silvija: So, you were a data scientist before it was cool?

Johannes: Yes exactly, it was not even called a data scientist, it was called data analyst or something like that. It was just before the hype of Big data. So, it was called, when I was doing this, it was called data mining. I mean when you did machine learning and predictive models, it was data mining. It was just starting to catch on. And we actually had some of the first in Telenor, pilots, on using machine learning to fine tune and detect what products to sell to which people. So, we were early to have some proper concepts there. Now it’s quite standard actually, to use the data we have, and to sell the right products to the right people at the right time.

Silvija: But I think having the length of that experience is a super strategic point. Because the reason why Google is so good at making value out of their data, is because that they had all these years of figuring out what the right models are. And you having worked on this for eleven years, and not just the past few years when data scientist were the hottest profession you can have, gives you this long intuition about how you ask the right question, how do you aggragate the right data? How do you find the necessary patterns?

Johannes: Yeah, Google is a good example. Because when I started, I started with the page rank algorithm, that was kind of Google. I feel Google is the first Big data company. I mean they scanned the whole internet, and the actually managed to pick out the relevant pages. And when I started, that’s actually what we tried to do in Telenor as well. I had a colleague that made a spin off company, trying to compete with the page rank algorithm. But there you learn... I read a lot about Google and thinking this data, first kind of thinking, that you need data before you make a hypothesis.

Silvija: I think it’s a cycle. A vicious cycle. You need the data, but in order to get the right data you need to have a hypothesis about what you need the data for.

Johannes: Yeah.

Silvija: And you’ve been doing this in many really really exciting areas, so I was hoping we could go concretely into some of these. This is kind of a meeting point of tel code data and a cool statistic. 

Johannes: Yes, that’s right.

Silvija: And you’ve used this. I would like you to tell us about the performance and optimization of mobile networks. At big scale, and why it's important. And why maybe, I don’t even know if you have to wait for 5G to do, or can you do it already?

Johannes: Yeah, it’s a pilot we run in Denmark. 

Silvija: Doing what?

Johannes: When you’re making a mobile network, you usually scale it so it always will be available. It’s so annoying when you’re a customer if the network is congested, that means it’s no capacity left. And then they call, and we will have complaints. So, then we upgrade the network and add capacity or radio power basically. But the thing is that’s very peaky behavior, during peak hours there’s huge loads. But doing off-peak. Imagine a beach area and the weather is nice, there will be a lot of people and demand of mobile network, and then you need to upgrade. But then, during bad weather and wintertime there is no one there, but the capacity is still there and it's consuming, because we have to have the radio on, so it's consuming power. So, what we want to do is to actually predict when we have to activate it. So, we can actually turn it off to save money, kind of forecasting the next day and the next month. So that’s sort of a no brainer, but you need data. All the towers have different patterns, so you need to ultimate this thing.

Silvija: But also, I think it’s a good development for these changing behavioral or usage patterns of people. I have four kids and they all want to try to use the mobile network for Netflix streaming. And suddenly you have peaks, that are probably quite predictable. It’s when we go to the cottage, it’s in the weekend, it’s in the car.

Johannes: Yeah, we look at the data and it’s so regular, usually it's quite easy to predict. But then you have those spikes and that you need to take care of.

Silvija: So, you can redistribute the capacity in one way or the other as well? Can you borrow capacity from one-

Johannes: That’s actually in some cases possible. We had earlier some projects trying to do that. But that’s not what we’re doing here. First, we just want to just shut off and save power. And in Denmark that’s one thing, but in Pakistan we have many towers that run on diesel. So if we manage to shut them off, we will save diesel, the environment, less CO2 pollution. So if it works in a small country like Denmark, we imagine we can just scale it up to the whole Telenor group, which have thousands of cell towers.

Silvija: You said the usage is actually quite regular and predictable, was that a surprise? I mean it’s an interesting learning on how people actually use the network.

Johannes: Yeah, I guess I have learned that looking at mobile phones after all these years, that people are quite regular, actually. And when it comes to mobility, for instance, old people, there are research studies looking at this type of mobility data. Most people have three places in their lives, that's home, work and one more place, that they spend most of their time. And of course there are some more places, that's pretty general.

Silvija: And the thing in between, is that the car or more social settings?

Johannes: Yeah, you have some travel, but it's amazing how regular our routines are.

Silvija: Yeah, how we live our lives. It would be fun to see some of your, you know, when you get this scaled up internationally, to see the international cultural differences.

Johannes: Yeah, we are actually lucky in Telenor to operate in many countries, so often when we looked into collaborating researchers etc., they are interested in discovering these social patterns. So we are actually lucky to be able to compare Norway, Sweden, Denmark with Asia, Bangladesh. We have worked a lot with Bangladesh and Pakistan, for instance. So for researchers, I think that is very amazing, as with my academic work as well, using this data to actually discover things about culture differences and human behavior. For the first time, we actually have data on millions of people in Bangladesh. I mean, it's quite amazing.

Silvija: And you also have worked on a very exciting project. I'm on a bord of public transportation companies, and I know that you were trying to help us some time ago to work with mobility analytics. Can you please explain what that is?

Johannes: Yeah, when you use your phone, you connect to a base station. And that base station is of course.., we have the location of that one. So every time the phone is used, we get a location for a customer. So what we did, was to aggregate that and look at how many people actually move, for instance, between south of Oslo and to Fornebu. So we can count the phones, basically, traveling in the morning-

Silvija: You don't need to know who's who, you just need to see the aggregate data.

Johannes: Yeah, exactly. For privacy reasons, we don't care about the person, we just want to know the patterns. So the meetings with Ruter was of course that they can use that to optimize their routes. Maybe they need a direct bus fro, say Grorud-

Silvija: Again, predicting peaks?

Johannes: Yes, and improve their... I mean, they move people, and we can actually measure how people move, so it was actually a win-win.

Silvija: I think with time, this might have an effect on payment services. Because eventually that phone could be paying as you go in and go out.

Johannes: Yes, I would imagine they are working on that. So also, regarding mobility, we also saw that a big opportunity was to look at foreigners coming to Norway, because we know if it's a German sim card. So we could actually measure how long the tourists stay in Norway-

Silvija: And where they actually go.

Johannes: And where they actually go - is it Berge, Oslo and how long are they staying in these locations. It's a huge industry, tourism, and we know little about how international tourists are actually using Norway. It is very fascinating, this mobility, because it is kind of a side effect of running a mobile phone operator, that you get this data, and now we explore how-

Silvija: I imagine there are a lot of interesting privacy issues. And you have a very strict privacy policy around this?

Johannes: Yes, so that's why we have worked a lot with lawyers and this GDPR specialist, to find good ways of anonymizing it. Because that's the trick - if you manage to get the data anonymous, it's much easier to use it. So we need to anonymize it as quickly as possible. That is possible, but it's a challenge to do, of course. There are some things we wish we could do, but we can't because it would be a risk for the customers that they could be identified. 

Silvija: I think that's a very interesting area where Norway actually can show unique approach to how we deal with data, very different from the one in the far east, or the-

Johannes: Yes, yes, for sure.

Silvija: You also mentioned to me that you do something on poverty prediction. What is that?

Johannes: That was really fascinating. We collaborated with some researchers in the UK, and they are working with the UN and others, to really understand how developing countries are doing, and how their economy is improving. Because now, the UN, when they want to know if an area is getting richer or poorer, they rely on the ground. People go on the ground and observe, and it can take years to collect data. And these researchers in the UK, they have found a way that you can use satellite imagery. To take a picture of the country, and you see if it is a lot of light during night time, for instance, tends to be a richer country, because they can afford more electricity. And they can see different types of crops, what color of crops you're making, and they can say something about the economy in that area. So what we want to do-

Silvija: Help me understand that. Because in the poor areas, you have not enough fertilization, or the wrong kind of seeds, or the wrong kind of-

Johannes: Yes, it could be that poor areas tend to grow certain crops that are easier, or that require less fertilizing or things like that. And also the roofing. If it is tin foil or very basic roofing, you can infere that this is a poor area. So that's really fascinating that we got in touch with them, because we have mobile phone data, and we did some earlier studies, showing that how you use your phone actually can indicate how rich you are. Poor people, they tend to use a phone... They have to buy scratch cards with the calling minutes. So poor people would often top up, for much less a month. Because they don't have much money, so they can'r really afford to buy a huge chunk of calling minutes, so they will just go through the hassle of top up more frequently, for instance. 

Silvija: So what can we do for good with these kinds of data?

Johannes: What was to be better at building and monitor development in these countries. And know what regions, for instance in Bangladesh, if you see certain regions have below 3 USD a day, you know that, and you would know where to actually invest, and try to stimulate economy. And it's kind of almost real time, because the satellite data, cell phone data, is coming there every day, so you can actually learn much quicker. So I think that is fascinating because it is big data, it's used for good in a way.

Silvija: Can I ask you... You were a physicist and then you became a data scientist before it was data science. But if you were to give advice to people who want to learn about data science today, where would you get them started? 

Johannes: I think it is very important that you have to read some statistics, and some maps.

Silvija: Where can we read that in a fun way?

Johannes: That's... I mean, I think YouTube is brilliant. So you will probably find something there. I learned in university, of course, but get the base statistics right, and then you have to learn some python, some programming skills. 

Silvija: You also mentioned a concept that is not pure data science, but social physics. Can you explain what that is, because I think that is a super cool thing. 

Johannes: Yeah, before this I was thinking about what I actually like to talk about, and that was exactly...  It's kind of a new term. We had a collaboration at MIT, and one of the professors there wrote a book called Social physics. But it is an old term, you take these physics methods, really scientific and experiments and being data driven, and you use that for social sciences.

Silvija: You try to calculate and predict the movement and dynamics of the social systems.

Johannes: Yes, you model the society in kind of the same way as you do. When I was at physics, I was studying molecules and electrons moving about, but actually you can use the same method and tools to understand how people move around, and try to predict where they will end up. So when I think about it, I really like that thought and using this approach.

Silvija: You also mentioned Laser Lab, what's that?

Johannes: Yeah, that is a collaboration we had that was really good, a couple of years ago with North Eastern University in Boston. So they were one of the first ones to really understand that telco data is a treasure drawer into human mobility, and try to really understand social networks. Because if I text and call you a lot, that means that we have some kind of social relation. So they were early, and we worked with them to study these huge social networks, and how our social networks develop over time, and how some people are really well connected with a lot of friends who also have a lot of friends, while others are odd fliers. And doing that, in a global scale, and comparing Asia and Scandinavia, that was quite amazing. 

Silvija: We're heading towards the end of our conversation. I asked you what you recommend people to learn, and as you said, they should learn some stats and machine learning, and perhaps play a little bit with python programming.

Johannes: Yes, that is...

Silvija: It is easier than people think. If you look at a couple of good YouTube videos.

Johannes: Yes, because nowadays you have modules for everything, and it is not that hard as writing Java, as I did when I started programming. That is much more hard.

Silvija: You have these lego blocks that you can just combine to create a module relatively easily. 

Johannes: Exactly, with a few lines of coding you can actually do something useful. 

Silvija: Do you have a quote you would like to leave to our listeners as a parting gift?

Johannes: Yeah, I was thinking about it before I came. I don't really have many quote, but I have a fun one hanging in the office. What was is...

Silvija: "If you do not change directions, you may end up where you are heading"

Johannes: Yes, exactly. That is the only future quote I have.

Silvija: Having a sense of direction is an important part if you want to have some idea of what your future is, right?

Johannes: Exactly, and you have to be open to change in order to get out of the track you're on.

Silvija: We talked about a lot of thing. If you were to choose one that you would like people to remember, what would you like it to be?

Johannes: Maybe the thing with bigdata, and I mentioned it can also be good. It doesn't need to be Facebook being evil and collecting too much data and squeeze money out of it. It can actually be used for understanding poverty and understanding how people's societies are functioning.

Silvija: Help societies live their optimal lives, in a way.

Johannes: Yeah, and understand it and actually learn.

Silvija: Johannes Bjelland, senior data scientist from Telenor Research group, thank you so much for coming here to LØRN and helping us see really practical applications of data science for social physics.

Johannes: Thank you.

Silvija: And thank you for listening.

Du har lyttet til en podcast fra LØRN - en læringsdugnad om teknologi og samfunn. Følg oss i sosiale medier, og på våre nettsider LØRN.TECH