VP Analytics and AI
VP Analytics and AI
Choose your preferred format
SS: Hello and welcome to Lørn. My name is Sylvia Seres and this is a podcast in collaboration with Telenor. Our topic today is going to be artificial intelligence and data analytics and I'm very happy to have Astrid Undheim, the VP of analytics and AI from Telenor Group as my guest today. Welcome Astrid.
AU: Thank you and thank you for having me.
SS: It's a great pleasure to talk to super brilliant ladies in technology that are doing very relevant and concrete work. So we will talk about how you apply artificial intelligence to network problems and basically services that Telenor offers. AI is in my mind, you know it has become this new catch or phrase for something new and fancy, and it’s when we become more concrete about its, not implementation necessarily but its effect that I think things get really exciting, and you are the woman. Before we do that I will like you to say a little bit of who you are and what drives you.
AU: Yeah, that's a good question, well I'm a girl from the countryside and born in a farm in a rural part of Norway, southwest, moved to Trondheim to study and I've been in Trondheim the last twenty years now. So did masters and PhD in telecommunications at NTNU and I've worked in Telenor research all my life professional life, after that, reason for taking that career path for me has been my love for maths since I was a kid and that was always my favorite subject in school, I always knew that I wanted a job that was maths focused, more than a computer kind of things.
SS: Exactly like me, but when I think I love maths and I like it but, maths is so many things so the thing I love about maths is geometry and the visual side of it, what was it for you?
AU: That's a good question, I don't remember exactly from an early age but after joining university I was triggered by very real world problems, so my study was into analyzing traffic data from networks, doing probability modeling of computer systems that was my favorite subject.
SS: You like understanding systems and structures.
AU: I'm an applied person, more than a core methodology person, so that's, yeah, it's a good place to be in a research department in the industry.
SS: But tell me, you are based in Trondheim, so Telenor has a big research department there or how does it work?
AU: No, not very big but we have a research department with the main office in Oslo, we have some people in Trondheim and even some people in the North and in Tromso but the main office is here.
SS: The reason why this is efficient is that Trondheim has the, let's say, admit that it's the best technical university in Norway.
AU: Yes exactly, and that's the reason why we have a research department there and I think also many of the other big industry companies in Norway have industry departments in Trondheim and the link to NTNU is important for that.
SS: So you get good applicants as well?
AU: We do get good applicants, some of the people working in my team have positions at NTNU, we meet them every week, almost every day.
SS: So then you work with managing the research related to artificial intelligence and analytics?
AU: Yes exactly.
SS: What does that mean?
AU: So traditionally the team started from a more traditional data analytics field, statistics et cetera, and then we moved our team into focusing more on AI machine learning and the methods that are more useful and more powerful today, so we have built it over the last four-five years to have more focus on AI. But we look at data, analyzing data in general but mostly focusing on machine learning and the new methods, deep learning, deep reinforcement learning.
SS: What data and what does machine learning or deep learning mean here? I think AI is finding patterns in data but it's when you start applying it that it gets interesting, so what data?
AU: For Telenor we can say it's three main areas traditionally, network data, also data about how many customers connected to a base station, how much data do they use, how many dropped calls have we had et cetera.
SS: So network efficiency and usage, and where do you need to improve it to get better services?
AU: Yes exactly, so that's one use case and the other is customer service and then it's about analyzing text data for instance to understand what problems customers are contacting us about. And for text data analysis it’s really the use of deep learning methods to understand or not to understand but to classify texts. And then it's also customer data, so what type of customers are leaving us, what can we do to keep these customers et cetera. So that's kind of the three main areas and then there is a fourth one that is coming up now which is more into new products, new services where IOT is a really big driver, and then our role is both to offer the best network for IOT services but also we want to see if we can also move into labeling full products in the IOT space. So delivering full applications.
SS: What does that mean? Let's backtrack just one step, IOT (Internet of things) and that is, why is a mobile company interested in IOT and how is that important for your business opportunities?
AU: I mean our network will carry IOT traffic or, so the sensors, IOT sensors will be connected to our network, they will have a sim card that will make it possible to sense something from the environment and send that data over the network to a place where you want to store the data, analyze the data and make use of it. I can give a very concrete example when I say we want to see if we can make a complete applications in this space, we have had a project now in Trondheim with NTNU and then with the municipality, where we have made or put together a quality sensors in Telenor, deployed them on Telenor network, collecting data, we have built an infrastructure to collect and analyze the data and then using machine learning to analyze the data, and in this case for predicting air quality. So predicting what the air quality the next day, next two days et cetera.
SS: And how do you manage traffic, maybe to optimize air quality in the city?
AU: Yeah that will be the next step, so the first step is just to get the data in place and analyzing and then next step will be how to manage traffic, maybe how to manage the cleaning of the streets, when do you have to clean, where should you prioritize your cleaning vehicles et cetera.
SS: Very nice, I asked you about, well actually let’s go into machine learning versus deep learning, could you say a few words about that?
AU: Yeah I mean I see deep learning as a sub branch of machine learning.
SS: If you will first explain what each of them are for people who don't know.
AU: That's always a bit tricky but machine learning in general is about having a mathematical model that can be used to do predictions or classifications, and machine learning process or training is to set the right parameters for that model based on the data that you have.
SS: But a human has to kind of define the model and then it’s optimized by the machine right?
AU: Yeah exactly.
SS: And deep learning?
AU: And deep learning is just one field of machine learning that is based on a neural network, and a neural network slightly resembles how our brain works with the neurons connected together but each neuron is just a mathematical function as well, and machine learning algorithm or the training process will be the same as with another machine learning model or can be the same, the difference is that these deep learning models are bigger so they have many more parameters.
SS: And maybe they can find patterns that humans were not able to put in the model.
AU: Yes exactly, they can find much more complex patterns than traditional machine learning models have been able to do, the biggest breakthroughs on use of deep learning has been within image recognition and text recognition and these are the kind of problems that you haven't been able to solve before, that deep learning can solve. But then we also see that it's also relevant for other types of, for instance IOT data, right? A deep learning model is able to predict air quality better than our traditional model is able to do.
SS: I use to work with AI, a long time ago it was only machine learning at a time expert systems really only. And I was shocked when I saw the transition that happened within one week last summer when Google went from natural language processing before deep learning and your networks and after, the quality of translations improved overnight to a level that was shocking. And it has to do with these kinds of tricky patterns that work much better.
AU: Exactly, so and it happened in 2012 and I think just as you say one week and I think over a couple of years the accuracy in these models for instance in tracking machines recognizing an object in an image went from being 75% correct to being 95 or 96% correct in over just a few years. So it's a true revolution that has happened and it's because these models are, you're able to build bigger models, you have more available data and you have more powerful computers and these three things together just make this work.
SS: Yeah, and it's a matter of scale, it's the Chairman Stalin who said in a quote quantity has a quality of its own and I think with AI we underestimate this because we think it's some new magic but really it's somewhat old magic but with extreme computational quality.
AU: Exactly, so the methods themselves, most of them were developed during the 90's and then of course there has been progress also on the methods of last year and what is happening now is of course all or not, all but a lot of talent flocked to this field that has struggled to recruit people over a long period.
SS: Are you able to recruit?
AU: In Telenor?
SS: Yeah for AI work? Must be a cool place to work.
AU: Yes I think so, I say that I have the most interesting job in Telenor or most interesting job in Norway, but I mean in general there are not that many AI talents in Norway and with AI talents I would like to recruit the people that studied AI when AI was not hot, so those that finished their PhD ten years ago.
SS: It really, we were passionate about AI because it was just a..
AU: Yes exactly...and I have to say we have been lucky to find some of them but they are not many, and I think we have done in Telenor we were quite early to start recruiting, which means we now have a good team which makes it easier to recruit again.
SS: If you have any secret AI personnel internationally in Telenor, this is a good chance to contact you.
SS: You said something that I was really fascinated by but I think it was a great point. I asked you about what you think are the most important controversies and you said this thing about AI and ethics, I think that we will see the next twelve, twenty four months lots of events discussing AI and ethics, my worry is that most people will not know very much about AI. And you made a point that it's really important not to try to regulate AI as a technology but to regulate the effect and the usage. In each in-health as a health service, in-traffic as a traffic thing, or I don't know safety as a safety thing, what do you mean by that?
AU: Yeah I mean, AI is a technology, its maths, it's a tool, but it is a very powerful tool. But it's not the tool itself that should be regulated but the use of it and I can come up with many examples where the use of AI doesn't have any either ethical or safety challenges at all for instance using AI for recommending songs in Spotify. It doesn't really need that much regulation, and then we should focus on where it action needs regulation where we talk about face recognition, where we talk about mass surveillance, autonomous weapons which is maybe the most pressing at the moment, we should regulate the use of AI for these fields, I really believe we have to. But I feel that when I say that there is a debate and I feel that the debate is a bit immature as you say maybe the technologies are not involved enough in that debate but what I'm most afraid of is that we discuss AI as this a general AI that will become dangerous on its own.
SS: But it’s the special applications that we need to be very wary of.
SS: And the way I think about it is if we were trying to regulate say cars like we try to regulate AI we will say you can only have four seats in every car and you have to have four doors and you have to use only this kind of petrol and you have to have a speed limit of 100km an hour, I mean nobody thought of cars as such, you know car is a technology you have to regulate the traffic rules around it, and it’s the traffic rules around AI and its different applications we need to think about.
AU: Yes exactly, and I think there is a big challenge now that AI as a technology is immature, and I really don't believe that AI is mature enough to be used for the safety critical applications, we are not there yet, the transparency is not there, the trust in the models are not there to be really used for critical applications, but we should start with the simpler applications, we should start where the use of AI have a low risk and then use that to learn how the technology works and then I think really important to support research into how to make sure that we take the next steps so we can also use AI for these more critical applications.
SS: Astrid, you mentioned open AI labs to me before, is this collaboration between Telenor and the technical university in Trondheim, can you say very briefly what it is and how can people get involved?
AU: Yeah absolutely and it was initiated by Telenor, NTNU and SINTEF in 2016 and opened in 2017 and then we invited in, also other Norwegian companies in 2018, so DNV GL, Equinor, Kongsberg, and the intention of the lab is to build competence for Norway. And I believe if you want to build...
SS: ...with concrete applications ?
AU: I will come to that because if you want to build competence you need to start with professors, because one professor means five or ten, five to ten master students, and you cannot increase the number of master students without having these professors in place. So that was from Telenor's side what we wanted to push, we need to have more professors in AI, and then we said we want students, researchers at the new university to work on problems and data that are relevant for Norwegian industry. Instead of working on data from Google or from the large Internet companies, they can...
SS: ...this is super relevant because we have some industry data that they don't have.
AU: Exactly, and I think that if we can both push that data and problems to the students and give them more interesting topics to work on, at the same time as building competence at this relevant for Norwegian industry, I think that is really the way to go, and what we see there are two important learnings. First of all, students really like to work on concrete problems within the industry with problem owners, and second if they are exposed to real data, real problems they can do amazing things.
SS: And this is where Norway can be uniquely good.
AU: Absolutely, and I think this collaboration and this openness that we see and that we want to push in the lab and inviting all companies to come with problems to the lab, to students, I think that is quite unique in Norway, there is very little bureaucracy, very much openness, these partners come with their problems and are open about it, and that's what I think is a unique selling point.
SS: Honestly I think these are probably the coolest masters degrees to do these days in Norway, especially when you combine them with the way the social sciences have been moved to Gløshaugen and now you can combine the social side of things with the concrete applications for these new traffic rules.
AU: Exactly, and I think there is a large potential to do even more in that space to combine these different expertise together.
SS: Agreed. We talked about what you do, you do it from Norway, what are the biggest advantages from doing this from Norway and what do you learn from Telenor's international arms?
AU: That’s a good question, I mean the biggest advantage of doing it from Norway is probably that the Telenor brand since we are a Norwegian company is stronger here, so our possibility to link with universities to recruit is maybe better here. I'm not sure if I have any other very good suggestions but at least that, what we have seen also with this open AI lab initiative is that in Norway when Telenor go out and say we believe AI will be important we want to take part in building competence on AI for Norway, that is is listened to. Other companies listen and universities, and I think it has created a movement that is much bigger than Telenor and NTNU and Open AI lab, it creates an inference to the rest of Norway as well, and I'm not sure if we will be able to have that same inference in another country.
SS: What will you recommend people to learn if they need to learn more about AI or read?
AU: Yeah to learn more about AI, I think I'll try to go a bit into the details of the technology, because I think it’s a field where the devil is really in the details. You need to understand the basics of the technology.
SS: You need to learn some statistics?
AU: Yeah some statistics and if you want to experiment and do a bit of yourself, you need to do a bit of programming.
SS: Now with all these open tools it's become surprising easy to put together your little AI tool in one day of the workshop.
AU: Yeah and that maybe also very, can be a bit dangerous because I think it's important to also understand the background, and the theory behind so that you understand the possibilities and limitations.
SS: You know the image I have in my mind, you know that good book in Disney cartoon with the sorcerer's apprentice where Mickey gets this, you know the sorcerer is out and the apprentice tries to be the sorcerer and he gets the broom to dance around and he can't stop it, that's where we can quickly get with these tools that are a little too powerful.
AU: Yes and I think the reserve value of having these people that have been part and know the whole feel of AI and know the history and know the pitfalls, and yeah.
SS: Astrid, would you like to leave a quote to our listeners as a parting gift?
AU: Yeah, no I was asked about a future quote, and what I would like to do is quote Nikolai Astrup actually, our new minister of digitalization and I've heard him several times talking about the Norwegian AI strategy work and he says, I really believe it, it is not a true quote because I don't remember exactly the wordings but the message is really that a Norwegian strategy should focus on grasping the opportunities of the technology, and then also handle the challenges, and I like that mentality, it's optimistic and pragmatic in my view. So I just hope that that will be the path we take, I mean I'm a technology optimist myself, I believe also AI will do much more good than it will do negative things but we need to focus on restricting those negative impacts.
SS: If people are to remember one thing from our conversation, what would you like it to be?
AU: That's a good question, yeah I think if we want to really succeed with AI in Norway, we need to understand this link between competence on AI and problems and data, and that needs to be a close collaboration, because it's not about dumping a set of data to the experts and they will get back with something amazing. You need to have interaction to maybe change the data, to make sure that you've collected the right data et cetera and the AI experts need to understand the main, to be able to do the amazing stuff so this type of collaboration industry, academia research, that's what I think you need to succeed in AI in Norway.
SS: Very cool. Astrid Undheim, a VP for analytics in AI, thank you so much for coming to Lørn and teaching us about how to use AI to optimize both mobile networks but also our future society.
AU: Thank you.
SS: Thank you for listening.
Hvem er du, og hvordan ble du interessert i teknologi?
Kommer fra en på bondegård på Jæren og har studert IKT/Kommunikasjonsteknologi ved NTNU (MSC og PhD). Har jobbet i Telenors forskningsavdeling siden 2009 og har helt fra jeg var liten elsket matte. Interessen for matematikk var inngangen min til teknologi, snarere enn at jeg hadde noen spesiell interesse for duppedingser og datamaskiner.
Hva er det viktigste dere gjør på jobben?
Vi er opptatt av å ta kunstig intelligens/maskinlæring videre som forskningsfelt. Vi driver anvendt forskning på AI, men jobber også tett sammen med NTNU, NR og andre rundt metodeutvikling. Det som er morsomt med AI, er at det ofte er kort vei fra forskning til anvendelse.
Hva er du mest opptatt av innen teknologi?
Jeg er opptatt av dataanalyse generelt, og maskinlæring spesielt – med fokus på dyp læring i kombinasjon med andre metoder. Den viktigste jobben min er å sørge for at forskerne får friheten de trenger for at de skal få brukt kompetansen sin og vokser.
Hvorfor er det spennende?
Jeg syns generelt at automatisering er spennende, og jeg er fasinert over hvordan man kan bruke data og maskinlæring til å gjøre ting mer effektivt og samtidig tilby kundene bedre tjenester.
Hva synes du er de mest interessante motsetningene?
Det pågår nå en stor debatt rundt etikk og kunstig intelligens. Det er etter min mening en stor tabbe å prøve å regulere kunstig intelligens som teknologi. Det er bruken som bør reguleres og begrenses.
Dine egne relevante prosjekter siste året?
Jeg er mest fornøyd med at vi har kommet godt i gang med å analysere data fra mobilnettverkene våre ved bruk av nye metoder og verktøy. I tillegg har vi fått på plass en IoT/AI-plattform som vi kan jobbe med sammen med NTNU i Open AI lab.
Dine andre favoritteksempler på din type teknologi internasjonalt og nasjonalt?
Favoritteksempelet er DeepMinds gjennombrudd innenfor Deep reinforcement learning, til bruk for spill (blant annet sjakk), men også for å styre datasentrene til Google for å spare strøm.
Hva tror du er relevant kunnskap for fremtiden?
Jeg er teknologioptimist og tror at fremtiden er teknologidrevet.
Er det noe vi gjør her i Norge som er unikt?
Jeg synes at det vi har startet på Open AI-lab, er viktig for å kombinere forskning og næringslivsproblemer.
Har du et favoritt-fremtidssitat?
Nikolai Astrup om Norsk AI-strategi: «Vi må gripe mulighetene, og håndtere utfordringene».
Viktigste poeng fra samtalen vår?
For å lykkes med AI i Norge er det viktig å få til store satsninger der vi kobler næring (data og problemer) med kompetanseutvikling og forskning. Blir alltid positivt overrasket over hva man kan få til når data, infrastruktur og de gode problemstillingene er på plass.