ESMT Open Lecture with Hal Varian

[Music] I'm Molly Rock I had communications and marketing here at es Mt Berlin and it is my pleasure to welcome you today it would have been our president your Kroll Holly but he is out of the country so it landed to me we were very very excited to have Hal Varian here today on the open lecture competition and data common economy in the digital age so if you haven't been to eat some tea before I won't talk much I could tell you talk for hours and hours but I will tell you we were founded in 2002 by 25 German multinational companies and we opened up this campus in 2006 and the room where you are here you can tell is not under historical preservation but much of the building is it is the former State Council building of the former GDR and we are just a pretty you know regular business school we offer an MBA master's and management executive education where we are top in Germany in executive education and I'm very international for example 90% of our MBA class is non German and as you might notice I'm not German either but some people think I am so we didn't ignore high rated this series of lectures in 2009 and since then we've had you know a great number of very renowned speakers and I'm very glad that we have professor Varian to add to this this group today he's going to discuss the digital economy the value of data and competition in the digital age and I'm very much looking forward to hearing more about his experiences at one of the most innovative companies in the world and to hear what you have and you're quite the question-and-answer session we're going to have how variant is the chief economist at Google he started in Google in May 2002 as a consultant and hasn't been involved in many aspects of the company including auction design econometric analysis finance corporate strategy and public policy he has pre met professor emeritus at the University of California Berkeley and has taught at MIT Stanford Oxford Michigan and other universities around the world in 2006 he received the humble it prize for that from the Alexander phone home bud foundation Hal Varian is perhaps best known as the author of two major economics textbooks intermediate microeconomics and micro economic analysis these have been translated into 22 languages and they are the pioneering work for many other textbooks as well and I have to say I'm not an economist I sent his play just like I play a German sometimes I play an economist here at the Business School and I did ask my husband that who took economics he said oh yeah yeah yeah I I used that textbook just wanted to make sure that it seems to be the Bible of economics it's the way I've heard it I can go on and on professor Varian is also the author or co-author of a best-selling book information rules and this was which i think is really really cool one of the 75 smartest books about strategy according to Fortune magazine so our lecture today will be moderated by a professor catalina stefan SQ kuhn so she's a professor of management science and the Deutsche Post DHL chair here at esmd she's also our dean of faculty before I hand over to Professor Berrien I would like to think I'm thank our media partners did hoggish biga and harvard business manager so professor Barron welcome the poor's yours thank you very much for that kind introduction I think the plan is I'm going to talk for about 10 minutes 20 minutes and then I have 10 minute discussion and then open it up for questions so it's a great pleasure to be here this is a paper artificial intelligence economics and industrial organization I sometimes call it the AI EIU paper so it's kind of like Old MacDonald had a farm eieio so organization here I'm going to talk about a bit about machine learning a bit about artificial intelligence and online competition a little bit about startups a little bit about Rd so there's something for everyone in this in this talk this is the triangle that's drawn the first day in the first class of information science at the bottom we have data and as you collect data and organize it lend inside to it you can turn the data into information as you take the information and you analyze it you can turn the information into knowledge and then of course you can turn the knowledge into action so there's a lot of debate now about the data economy maybe the information economy the knowledge economy there are part they're all part of the same process none can stand on its own really and let me turn to machine learning and AI machine learning has been around for a long time as is AI but in the last five years there's been a very fruitful combination of artificial intelligence and machine learning and these AI techniques have been adapted to learning things that five years ago six years ago we thought were really impossible but there's been huge breakthroughs in terms of image recognition in terms of voice recognition in terms of automatic translation and it's a remarkable period for artificial intelligence and machine learning these days because so many techniques have emerged now that just a few years ago were thought to be impossible why have we seen this Renaissance well we've had better algorithms we've had better hardware we have better data and most importantly we've had better expertise because people have learned to apply these techniques to problems that were not possible before the hardware and software they're pretty easy to acquire because you can go to cloud computing at Google or Amazon or Microsoft and you have at your disposal a lot of special purpose hardware a lot of special purpose algorithms that allow you to apply the machine learning data is often easy to acquire data nowadays is generated as a side product of other activities and the challenge is to take that data remember the base of the pyramid and turn that data into information by using the tools that I just described and these days the scarcest factor by far is expertise you're seeing salaries being bit up and people moving from place to place because the necessary experience and expertise is scarce but our universities our schools around the world are turning out students who can use these techniques and I think soon that barrier of expertise will will fall there's a nice little company called kaggle Kangol sponsors competitions and machine learning so if you have expertise but no data and he has data but no expertise well you'd like to get these parties together and what Cagle does is it will work with a company to extract a set of data for training and for testing and for validation and then offer prizes for the parties that can best satisfy this challenge so the money can mount up quite a bit on Zillow for example which is online real estate site they use a machine learning system to estimate the value of a house this is very important to them as an attraction to potential home buyers and sellers and so they're willing to pay where is it 1.2 million dollars to whichever group is able to come up with a significant improvement on the technique that they're currently using and I've listed a number of these traffic to Wikipedia personalized medicine taxi trip donation duration product search relevance etc etc you can go to kaggle and browse through this list they have about two hundred and twenty competitions since they started and get a good idea of practical questions practical answers how much companies are really willing to pay to get these prediction problems solved so it's a nice illustration of real world real problems with real money and real solutions behind them and this is just a little chart the people who are using kaggle who are competing in that those competitions are in fact from all over the world a lot from North America of course a lot from Asia and received more and more from Europe entering into these competitions because every class of machine learning they'll have a little tangled competition as a real world example for the students so it's a very nice way to get a picture of what's going on in this particular area okay so remember I showed you that pyramid the data pyramid let's talk a little bit about data so people say data is the new oil well my response is they have one thing in common to be useful they have to be refined a barrel of oil isn't worth much you turn it into gasoline or kerosene or hydrocarbons and it's worth something and it's the same thing with data raw data is worth very very little it's not worthwhile until it's really distilled into the information and knowledge that I described earlier now in economics unlike oil data is not rival you may have a barrel of oil or I may have the bear oil but is excludable its rival if I have it you don't have it but with data is quite different we can easily have access to the same data at the same time just like all these Kegel competitions there could be hundreds or even thousands of groups around the world that are analyzing the same data so I think it's a mistake to discuss data ownership because it's too narrow a concept it makes sense to talk about ownership of a barrel of oil but when you have data and used by multiple parties really want to talk about rights access licensing regulation contracts permissions all sorts of contractual ways to share access to data it could be an exclusive just like the barrel of oil could be exclusively owned or it could be a license where you could manage it for a specific amount of time many many things are possible so she would think much more broadly about this this concept of ownership it's really a question of contractual arrangements and usually pluralism multiple access to data is important because it allows for a diversity of viewpoints and in many cases allows for a significant improvement in terms of utilizing that data some party may come up with new ideas for how to use a particular set of data and that's all to the to the good data portability well we have something called Google takeout so we provided this for what seven or eight years at least you can go to Google just with a few simple keystrokes download all of the photos you put on Google or all the emails all of the search history etc etc and so the data has become very portable interesting thing is even though we offer takeout there aren't so many take ins that it's rare to find someone who says yes you ask give me your data give me your data I want to use it for this new new purpose so take out as easy take in is a little bit a bit harder where does the data come from I've got a long list here I have to look at it myself byproduct of operations that's by far the most common form everybody's spent millions of dollars beating a building data warehouses they're acquiring data about operations about consumer behavior about marketing all sorts of different things and the challenge is how can you use that data effectively of course they're huge amounts of free data available in the web you can build a web scraper yourself or you can use common crawl which is a free access to the entire web literally petabytes of data they can be accessed you can offer a service several years ago we realized that we needed to acquire expertise in voice recognition because mobile phones were emerging as a way of accessing Google services we had no expertise we had no data so what did we do we hired the best people in the world and they said let's build a mechanism to acquire data we set up a system called goog-411 which was an information service you could dial goog-411 and say I want the phone number of Joe's pizza on University Avenue in Palo Alto and it would say do you mean John's Pizza no I mean Joe's pizza and so people trained this system by using this service efficiently it learned how to recognize voices that learned how to recognize accents that learned how to recognize all of these different things and there's very limited domain of directory information but that was enough to get started and now I think we have one of the best voice recognition systems in the world you can share data by data data from governments and NGOs cloud providers so you look at Microsoft Google Amazon they're not only providing you with the algorithms and with the hardware but they're also more and more providing you with the data so it could be census day it could be traffic data it could be weather data all sorts of data repositories are available via the cloud services because it makes a better product for the users they can hit the ground running and they can do the analysis they want by merging their own data or their own analytic techniques with the public data available on the cloud so here's some example of data training sets Google released nine point five million images labeled with their contents to the public at large to the open image project video video understanding eight million labeled YouTube videos so right now image recognition is a solved problem you can do very good image recognition using the tools that I alluded to a few minutes ago but identifying video content is still rather difficult you can see a group of people and ask are they fighting are they dancing are they exercising are they marching what are they doing the machine systems are not sufficiently trained in understanding activity video activity but I think this will become a solved problem in just a few years and this 8 million labelled YouTube videos will facilitate that voice data crawl data set I mentioned that click prediction OpenStreetMap the cago data sets six thousand seven hundred and eighty-four data sets available european parliament proceedings now one of the really good things a european parliament has done is produce these multiple transit translations of all of the activity and speeches and bills that go on and that date has been really invaluable in terms of producing automated translation machine learning systems canada has also been very helpful in this regard since everything is done in french and english or you could look at belgium you know on and on these translation documents are extremely useful for this kind of analysis let's see how important is data so I said it forms the base of the pyramid but then the question is there are other issues the expertise the hardware the software and so on so how important is data well here's the imagenet challenge where they had 20 million labelled videos and each year for the last how many years is at six six or seven years they've chosen 1.2 million images for training your machine learning system and then a hundred and fifty thousand images for testing and everybody who did image recognition would compete in this image net challenge because they wanted to show why their system was better than the other guys interesting thing about this is they kept the size of the training set fixed because they were interested in the algorithms and the hardware improvements and here's what the error rate looks like back in 2016 for the first time they surpassed humans in terms of image recognition and it just keeps improving since then I remember this was a fixed set of data so having more data was never an issue here it was really coming up with clever algorithms and better hardware and of course the accumulation of expertise so data is important nobody's going to deny that but it's not the only thing that matters and this is a picture of a image cottage neck competition on a subset of data namely recognizing dog breeds okay so they're about 200 different breeds of dogs and now you can take a picture of a dog using google photos and he'll tell you what breed it is it's pretty pretty magical 120 different breeds of dogs 20 thousand images and they looked at the number of training images and how the accuracy improved as they increase the number of images obviously gets more accurate as it has more data but there's diminishing returns just like any other to reproduction and here's a few examples if you go to cloud.google.com slash vision you see this empty box and you can drop a picture into that box and the machine will tell you everything it recognizes in that picture so here's an example this is a cat our cat in fact and guess what it recognized that 99% sure it's a cat it's a small to medium sized cat it's a cat like man mammal but here's rag doll well that's the breed of cat so not only did it recognize it was a cat but it recognized the kind of cat it was and it does the same thing for dogs and horses and all sorts of other things of that sort here's a picture of Wikipedia it says that's a ragdoll cat and this is a picture of our ragdoll cat so you see it did a pretty good job of recognizing that particular breed which is quite remarkable now you can do it with plants all sorts of things you photograph a plant and it will tell you what the plant is and how much water it needs how much sunshine it needs all of those things you don't have to memorize these long lists of cats or dogs or plants your mobile phone will tell it for you now what happens in terms of providing all of these great services that I just described well it's a very very competitive market for example here in the left wash what do I have I have the Google pricing for image recognition and here on the right I have the Amazon pricing for image recognition and it's basically a tenth of a cent per image so you can feed in thousands of images it will get this labeling I just showed you and it does have for a $10 a very low cost tenth of a cent per image so this process which took let's say many years to develop and a lot of expertise a lot of hardware a lot of software a lot of training went into it but now it's been essentially commoditized anybody has access this at a very very low cost and in fact this is not unusual you'll see the same sort of thing happen with other kinds of machine learning because of the competition in the industry I just described so there's competition in pricing as soon as Amazon changes a price Google follows or Google changes the price Amazon follows the standard services that are offered everybody is offering images transcriptions boys translations again if you go to the cloud google.com slash translation you can drop in a text from some other language and will translate it for you you could do similar things with voice transcripts and so on on the cloud computing services not only you getting the data sets I described the algorithms I described you also get some training assistance consulting etc so it's made it far far easier to begin a business in this area because you've got all of these services available just from the from the beginning for the first day and in fact I think this general theme of online competition which has become a very contentious issue in in Europe in particular if you look at this chart you can see the competition going on between Amazon Apple Google Facebook and Microsoft in advertising autonomous vehicles browsers digital assistants ebooks etc so all of these services are now available at a very low price often zero at a very high quality compared to what we had available five years ago and they're continually innovating and improving and offering their services so the reason that you see so many low-cost highly available technological advanced services is because of competition because all of these companies are competing with each other to provide those services to potential users this is a picture of R&D spend which is pretty interesting according to this chart Amazon is a leader in global R&D spend with Google close behind the red dots are the technology companies the green are autos and you wouldn't have seen this a few years ago why are all the auto companies investing so heavily in R&D well of course it's autonomous vehicles a huge opportunity and let's say an opportunity and a threat at the same time because if your competition moves ahead and you fall behind it's going to be very problematic but of course for users the advances in these areas are really staggering where we're seeing lots of spend on R&D but lots of outcomes that are again something that was inconceivable just a decade ago people would find it very hard to believe that you can have fully autonomous vehicles but now it's definitely a reality and one example of this competition I referred to this is a competition in in-home assistance where you have smart speakers like the Google home Mini or the Amazon echo this first came out from Amazon within nine months Google announced its home speaker now we're seeing a very intense competition going on in that area with prices being computed down and the number of services rising very dramatically so from our perspective from the company side we're seeing huge amounts of of competition going on in pretty much every area from cloud computing to smart speakers to home video and the venture capital funding is robust this is a picture of what venture capital funding looks like in the US and in Europe over the last decade the peak of course was back in 2000 that comm boom as they say but now we're seeing expenditures on VCS and vege capital investments that are the highest we've we've ever seen since that 2000 period so we've got robust activity going on in the commercial venture side as well and in fact that's partly because it's so easy to start a firm because so many business processes can now be outsourced so this is not entirely serious how to start a startup well you can fund your project with kick-starting you can hire employees using LinkedIn you could use cloud computing from any of the providers I mentioned open source soft software is widely available cetera et cetera et cetera so all of these services and capabilities are now available it's like a construction kit for starting a business you can handle HR here legal here R&D here and start your business going really from scratch in a much more efficient way than was possible five or ten years ago and that's the talk AI is here to stay hardware software data lots of practical applications very intense online competition a lot of entry going on in this area so we really live I think in a golden time in terms of innovation in the digital context and well I'm interested to hear your opinions on this so let's turn to the conversation thank you [Applause] salsify thank you very much indeed Fadi very timely and interesting comments now one thing that crossed my mind as you are talking about the many applications of artificial intelligence of course big data is we all realize they're huge tremendous potential in your view are there any limits to AI to its usage and applicability of course I'm not talking necessarily technological limits because with computing power and storage capacity increasing this is no longer a concern perhaps not for even expertise limits because we're getting better and better at it I'm thinking more in terms of behaviors psychology mmm-hmm well I I do think there are limits there we've had this tremendous breakthrough so that's made people very optimistic about these fields of image recognition voice recognition translation and of course there's a great deal of interest as saying how far these can be pushed how far can they go I come from the machine learning background so the AI techniques of using neural nets and deep learning and so on are novel to me but it's it's amazing to watch what people have been able to do and there's a lot of interest these days in what's called transfer learning I mentioned it on one of the slides but I didn't talk about it it's how you can use what you've learned in this area to carry over to some adjacent area so you've learned how to recognize all these breeds of cats can that help you recognize breeds of dogs or horses or monkeys or whatever and it turns out with humans we do that we extrapolate our learning and now machines are beginning to do that as well so at this point I would say I'm sure there are limits but we just don't know where they are and that's why there's so much excitement surrounding this area and do you ever feel that they might certain resistance from the point of view of the end-user to do these techniques yes I would say the businesses you said at the end-user the end-user yes well of course the end-users I think will find a lot of these services and these capabilities very very helpful I like to say if you want to see what the future looks like look at what rich people have today in five years or ten years that will be available to everyone so now rich people have human assistants who help them organize and plan their lives but within a few years everybody will have a digital assistant that will provide the same functionality that will make life certainly easier now speaking as a statistician of course we're talking here with big data a primarily pattern recognition and prediction a lot of that now of course from the point of view of classical inference the Holy Grail is causality right how do you think of causality in the context of Big Data in machine learning well of course if you can as you indicate if you can design an experiment and run an experiment that's going to be the best the best way to understand causality and at Google we have built platforms for doing this kind of experimentation and not only for ourselves but also making those available to our users so I think we'll see a lot of exciting activity there in terms of using these systems to detect causality for everyone I mean just as just as you democratized as so that lots and lots of different parties can use these systems to understand the causal connections in marketing or production but there are also some techniques you can use that are using quasi causal data that is they're looking at situations where you're trying to infer causality in a way that you to do this without using a classical randomized control problem so things like difference-in-differences or regression discontinuity design or instrumental variables or other things of this sort that are finding their way into machine learning not as good but better than nothing so we think we can learn a lot from those those statistical techniques applied in this in this context now you have talked a bit about the competition in the online marketplace and of course there's a large academic literature looking at a link between competition levels and innovation outcomes at what level on that curve do you find that we are now well I think what's interesting about online competition is compared to offline competition is how easy it is to switch to alternative providers so if you don't like Google's results you can type a few letters and go to Yelp or go to Bing a go to any other kind of search services so using alternative providers is far far easier online than in any offline business so that's why it's so competitive and that's why we're seeing so much online innovation and technological improvement because of that competition then you expect that this is going to continue so again you expect this is going to continue yeah I think so partly because the entry costs have gotten down so low so somebody with a new idea can often get funding and can get the cloud computing environment and apply these state-of-the-art algorithms so there's no shortage of entry in this area at this point um it's of course we all know very soon in Europe we're going to have the new data protection recognized in fact what impact do you expect this to have on the general online activity and the way in which the online businesses so if you look at the in the so-called industry Giants you look at the googles and the Facebook's and Microsoft and Amazon they've all spent 18 months or 24 months working to be compliant with gdpr requests that might occur so we feel pretty confident that when the magic day rolls around what 10 days from now or pretty pretty pretty soon that we'll be in in compliance but I think going forward it will be more difficult for small companies to take on this extra obligation to be able to produce the reports and the user requests that they come in so I do have some worries about whether this vibrant competitive eco system will continue to thrive because you've added some more friction I think that's undeniable that there's a bit more friction there and just how the industry as a whole would deal with that it remains to be seen slowing down the pace but still keeping them perhaps yeah we hope not but it's possible what well I'd like to take to turn the floor over to the audience and take a couple of questions I am I'd like to know but you said that artificial intelligence and consists on expertise and data but I think the real and realistic data will only come from the real professionals working in a certain area working in a certain job having the everyday experience and having the psychological connection to consumers and I think those professionals will be where we interested in giving data to some enterprises who are willing to substitute those professionals by their self so how are you going to get this data from the professionals right well there are lots of examples out there you may have seen the famous cucumber sorting example there was a student in Japan who worked on machine learning he went home to visit his parents who had a small form for a small farm for cucumbers and realized what here was a place he could apply his knowledge and indeed he was able to set up a machine learning system that made the family farm much more productive and all of these cattle data sets that I showed you this 2220 data sets are in fact available for researchers to apply their skills to and all of these 9.5 million still images and eight million moving videos from YouTube they're available so I guess my experience is that typically it's possible to acquire the data you need and if you're an ongoing organization the data is typically available as a byproduct of actual operations so I understand that particularly in Europe people were talking about the data economy and data monopolization or data competition whatever but in practice I haven't really seen it being a problem that usually people have the data the challenge is finding the expertise to extract the value from that data so Mike I mean the machine learning allows you to correctly quote who your agent is and there is basically no isometric information and in this regard do you believe that the information economics is still relevant well yes I believe economics is still relevant actually information well it is interesting because one of the possibilities that computers offer us is they offer better monitoring and compliance if you agree you and I make a contract that you will do some service and I will pay you to do that service there's always the question of how do we verify whether you're really providing the service at the agreed-upon level and having computers available to monitor those performance and contracts that's actually quite helpful so I'll give you a it's kind of abstract let me give you a very concrete example it used to be that the contract between an advertiser and a publisher was pretty simple I'll put your ad up in my publication and maybe people will see it and maybe they'll come to your store that was what the old world looked like in advertising well now online is I put your ad up on my publication and only if people come to this door do you have to pay me that is just seeing the ad isn't enough you're only going to pay for that ad if there's a click and the visitor that is associated with it so you've got a much much better measure of performance in that contract and you had prior to through the online world being a primary source of advertising now so it's a big big change and of course we've seen this explosion in advertising online because of the fact that it is so easy to monitor questions oh here we go you present eight you went to great lengths to convince us that there's lots of competition and just now you did sits as well so I'm wondering under what circumstances could you see yourself arguing in favor of breaking up your Google or any of your competitors yeah thanks well I I've gone a great links but let me go another length or two before I address that question if you look at an online search general purpose search my argument is it's a really tough business because you can only sell 6% of what you produce because if you look at clicks 94% of clicks or on organic results and only about 6% of clicks or on ads but ads are where all the money is so in fact there's really intense competition on commercial searches right commercial searches if you search for travel there are many travel providers Orbitz Expedia a trip planner etc etc if you look at shopping there's Amazon there's eBay there's many comparative shopping services Google shopping and so on if you look at insurance or you look at automotive or you look at any of these areas where the query has a commercial value there's typically a large number of competitors and if you think about it firms that enter in to this to this area are focusing on the commercial side of things because that's where the money is so actually even though Google as I would say obviously a strong position in terms of general purpose search it's really that 6% of the searches that are commercial that are really contested so that's why that's where all the competition is on that relatively small segment way in the back there you're probably all aware there's been a lot of political controversy around some of the services that have become available because of the internet and these huge platforms so I'm wondering what your analysis is of how soon these tools that you've just shown us will be able to filter out some of the content that you might not like to see now whether that's competition related people uploading the most recent football video from a game that's just finished or whether it's related to political events it would be really interesting to hear how how soon you think that will work because at the moment as far as I know Facebook are still employing humans to actually filter out stuff well on YouTube we're working very actively in that area and a substantial number of YouTube videos are filtered using machine learning techniques and often at this point reviewed by humans for the final decision but the this technology is getting better and better part of the issue is you might have exactly the same image and in one case the voice-over could be very positive and in the other case the voice-over could be very negative so you see some battlefield situation is this a movie is this a documentary is this a news is this Isis recruitment film in lots of cases there's still ambiguity that you can't resolve until you look not only at the image itself but also what people are saying about the image now this is can be solved this is not an impossible problem by any means but it's just the the effect can be quite different given almost the same content so that's why it's challenging to really really analyze this these completely by machine we think that it has to be people plus machines to be able to be really effective I'd like to return to this question of whether there are limits and I'm actually interested in technical limits is there any particular area where you say this is gonna be really hard to crack and you'll never get it to be done and I'm kind of building on this would the limit not be the very expertise that you're saying they're currently lacking so the people who develop artificial intelligence are they going to be replaced by artificial intelligence at some point on well there are several challenges in this area one is what's called adversarial AI so you can deliberately design images to fool some of these AI systems so just as human beings are subject to optical illusions computers can be subject to optical illusions but they're completely different optical illusions so what fools a computer wouldn't fool a person but what fools a person that wouldn't fool a computer so there's this whole issue of how do we deal with adversarial artificial intelligence and another challenge is the so-called right to an explanation in some of these AI systems they make decisions or they can they can come up with a good predictive model and we don't really understand how they came up with that model so an example of this is there were a number of medical images that were examining the retina and the eye and using this as an indicator of health and those were quite successful in doing that and along the way they discovered that the images of the eye could also be used to identify gender so it was 85% accurate and identifying whether it was a man or a woman and the the medical literature had never noticed this before and in fact just a human looking at the images could not understand why this was classified as from a male and why as a female well after a few weeks of work they were able to understand the difference but it isn't something that's immediately obvious in many cases and there are a lot of requests or a lot of proposals that there should be the ability to explain the decisions made by these AIS again I think that's a solvable problem but it's just we aren't there yet that's a challenge that we have to face in the coming months and years there's somebody back there I got a question about the role of government in the data economy and basically it's it's twofold question the first one is in terms of regulation I think if I interpret you correctly you're slightly skeptical or at least hesitant when it comes to the regulation you're in Europe as in terms of the complexity it will introduce but do you think something like this might happen in the States and the second question is in terms of resources I I think you're absolutely right there's a ton of resources online already but where do you see the role of the public sector or the government to pry more data or resources to sort of accelerate the data economy well I obviously though there'll be regulation of course both in Europe and in the US and of course let's not forget about China where the regulatory environment might be quite different I think that there are many attractive opportunities in medical studies and health that will benefit from machine learning these areas are both very very important to people and at the same time they're often somewhat sensitive in terms of privacy issues so in my view I think we should be very careful about constructing legislation in these areas because the potential benefits are so great so I'm not opposed to regulation I just think that regulation has to be applied with a great deal of care so we understand what the harm is why the legislation or the regulation can mitigate that harm before we run out and start passing laws we'll see we'll see how that works out by the way in the u.s. there's now a referendum on the California ballot that is very similar in spirit to the gdpr so clearly these were those kinds of regulations will be will be arising so we have time for maybe one more question there we go um well you just raised the issue of China and I realized that in the slides you presented none of the Chinese Internet giant's are there and of course that makes sense because the markets are as of now fairly separated but I would like to know what your opinion or maybe even Google's view on the whole issue of Chinese Internet giant's are and yeah maybe just some insights perspective well I I will only state my own opinion I am NOT going to try to guess what Google's opinion is in this area but of course doing business in China is very different than business in the rest of the world it's very highly entrepreneurial there's this very important role between the private enterprise and state enterprises and I think it's going to be difficult to export that Chinese model to other countries in the world so the Makai nays are happening there how quickly things are moving but will those be able to become global companies I think there's considerable skepticism on that particular issue so maybe that's a good place to end thank you very much for joining us today and thank you for that [Applause] [Music]

Channel: ESMT Berlin

Views: 1,483

Rating: 4.7894735 out of 5

Keywords: Bschool, Berlin, ESMT, ESMT Berlin, Open Lecture, Google, Data Economy, Digitisation, Digital Economy, Economist, Chief economist

Id: n3UZfvnczKA

Length: 51min 22sec (3082 seconds)

Published: Wed May 23 2018