 Good afternoon, my name is Jean-Tierrault and I'm very honored to be chairing this panel, a very distinguished panel. I'm not going to spend too much time introducing the participants, they are well known to you. So, the first speaker will be Emilio Calvano and then Philippe Val, President Val will be speaking and then last Bernard Stein for the commission. Paul Belflam unfortunately cannot be with us today, thanks to strikes. But I'm sure you'll fill in with your many questions. So, first we have three presentations for about 15 minutes each. I'll make a few comments and we want to have plenty of time to be discussing with you so we'll open the discussion to the floor afterwards, okay? So Emilio. I'm going to be talking about artificial intelligence and conclusion. This is joint work with a team. Let me take this opportunity to thank you. Can you use the microphone? Where is it? Yes. So, let me take this opportunity to thank the Toulouse School of Economics for financial support and in particular the digital chair initiative, okay? So, let me start with Amazon. As we all know, Amazon is also a marketplace where sellers can go to sell their items. So, this is the marketplace for this particular mop. This is a fantastic mop. Look how it cleans your car. Actually, believe it or not, this is one of the top 100 selling products on Amazon.com, okay? Now, there are actually six vendors selling this mop. And if you wanted to have an idea of our price competition works in this pretty homogeneous product market, then you would just, you know, look at prices. So, this is data from these marketplaces. So, this is high frequency data on the price that was charged by these six sellers over the course of 40 days in the summer of 2015. So, the scale here is in hour. So, these are 40 days and these are dollars, okay? And if you don't understand what's going on here, that's actually normal because what's going on here is what we call a signature of algorithmic pricing. So, this is basically software who is exploring at a high frequency. What would be the actual optimal price, whatever that means for selling this mop? And so, prices here are changing as fast as once every 20 minutes, okay? Let me give you another example of algorithmic pricing on Amazon. This is a, what is it actually? It's a lintrop. I mean, this is one thing that you use for your dryer. It's also on the top 100 selling list of Amazon. And prices look like this. I actually cherry picked this because, so here, you're also seeing algorithms at work, but it's a different game they're playing. They're actually interacting. So, over the course of almost a week, they were ramping up prices around the clock 24 hours every 20 minutes. And then there is some sort of price war. And then they start climbing again and then price war and then some more action here until they settle finally at a price at about $10, okay? Now, the point here is that algorithmic pricing is a reality. And actually, there is an industry on Amazon that sells services to sellers who want to delegate their pricing choices, okay? So, if you type in on Google repricing software, you would get a set of these vendors that claim to use whatever artificial intelligence, cutting-edge technology in order to get the most profits from your goods, okay? And actually, I bet that some of them do, okay? Now, so what? Algorithmic pricing is not news. Airlines have been doing this since the early 80s. But what happened recently is that advances in the field of artificial intelligence and statistics machine learning basically created a new vintage of this algorithm which shared different in so many ways with respect to the one we used to know, okay? Just in a nutshell, they're powerful, they're versatile, and they're accurate. I'll come back to this in a second. And they're increasingly available off the shelf. So, Google, Amazon, and Microsoft, they all sell versions of the algorithms that you can put in production in as little as one day, okay? Now, what we study is a specific class of algorithms which is called Q-learning, okay? So, Q-learning algorithms were designed to solve what we call, what economists call, Markov decision processes or problems. If you don't know what that is, just picture an agent, okay, who has to make a set of repeated choices. And before each and every of these choices is going to observe some information which is relevant for that choice. After the choice is going to observe a reward which is tied to that choice, and now the world around them changes, okay? Now, the agent, it's basically, if you want to, in a nutshell, describe what these algorithms do, it's very simple. At a conceptual level, they explore with the actions, they figure out which one works just by looking at the rewards, and then they reinforce those actions in the future. So, the chances, actions that work well in the past are played with a higher frequency in the future, okay? Just for fun, let's look at applications. So, this was a very influential paper that was published on Nature three years ago now by a group of researchers at Google DeepMind Lab. Basically, what they showed is that a simple version of this Q-learning algorithm can learn to play Atari video games in that application better than humans, okay? Now, here's the choice problem. I can actually show this to you. If Zhang gives me two, three minutes' slack, I think it's fun to see it, okay? I'm buying time. Let's see. You'll tell me later, okay? So, while it goes, this is the algorithm which is learning to play, okay? While it goes, I'll describe the problem. So, the problem here, the action that he has to take is whether to stay left, center or right at each and every point in time. The input here is the screen, okay? That's a visual input. So, the algorithm is told what is the color of each and every pixel on the screen, okay? Now, there are a lot of pixels on the screen, and each pixel can have 120 color. That was the palette back in the 80s. And so, this means that there are 4.3 million possible screens, which means that this algorithm has to figure out what to do in 4.3 million situations, okay? In a way that maximizes the reward. Maximizing the reward here means, you know, breaking this wall. And this is what they show. They show that in about two hours of training, they learn, and they learn even clever strategies. So, in this case, the algorithm learns that if he digs a hole here, then he can put, he can send the ball up, and this is basically the most effective strategy. Those of you who are familiar with this game to actually gather points quickly, okay? So, just another word. These algorithm, they don't know anything. So, they're model-free. So, they don't know what a ball is. They don't know what a wall is. They just choose center, right or left, and they get that reward for doing that, okay? How do we go back to that? Okay, good. So, what we wanted to do, we wanted to understand, we wanted to apply this to online commerce, and in particular, we're interested in whether this intelligent pricing algorithm can actually learn to maximize profit, which in that environment would mean learn to collude. We ask whether algorithmic collusion is any different than collusion between humans. If yes, how do you detect it, and then I'm going to discuss policy approaches to this problem, okay? So, what we did, this is just a probe, okay? Take it as this. It's a probe. This is a long-term project. So, what we did was basically an experiment, a simple one to check if it makes sense to go on, and I'm just going to show you these results here, preliminary results. So, what we did, we coded two of these agents, and we let them play our game, okay? So, our game is what we call a pricing game. So, the idea is that these two agents, they will have to choose prices for a number of periods, and in particular, in this experiment, they have to choose between high price and low price. What they observe is what the other agent did in the last period, okay? Now, the payoffs, if you've been looking at them, are structured in a way that there is a common interest in keeping prices high, but in an individual interest in undercutting your rival and charging a low price, okay? Now, this is a compact game, okay? So, we basically want to test whether these algorithms can learn collusive strategies. Now, a collusive strategy is a strategy in which there is a punishment phase and a cooperative phase, okay? So, basically, what they have to learn, they have to learn to punish. Punishing means sacrificing the shortened payoff in order for the long-term gain of getting the collusive monopoly payoff, okay? So, how do they did it? Now, what I can tell you is that they did quite well, actually. We were surprised by this result. In 52% of all these, oh, sorry, I forgot to tell you that we do this for 1,000 experiments and each experiment is 1 million games each. 52% of these experiments, these guys, they actually learn to play a collusive strategy, okay? So, another way to say that there is a lot of high prices in this game is that in 64% of 1 million times 1,000, it's 1 billion iterations, these algorithms were charging high prices, okay? And the convergence to these collusive strategies was actually pretty fast. Now, another nice thing of doing simple experiments is that we can actually understand exactly what's going on. So, here I want to show you a bit how they reason, okay, in a way. So, what you're looking here is time. Okay, so these are the number of repetitions or interactions, so to speak. And what this graph shows you is basically what are the incentives to defect when you are in a cooperative phase, okay? If this is negative, then it means that the algorithm actually understands that the present value of doing so is lower than zero. So, they'd rather cooperate when your opponent is cooperating than defect. Now, what's interesting about this graph is that it shows you that in a sense first of all it shows you that there is fast convergence at the beginning, okay, towards these negative values. And second of all that from time to time there are some price wars, okay? So, these price wars basically are the price that they have to pay because they're experimenting, okay? So, remember that from time to time they won't do the collusive outcome because they want to check what happens if they do something else. But when they do something else, sometimes they trigger price wars and so they have to start learning again. So, you know, this is just one way that we use to try to understand how they work and what can we do about it, okay? Now, the key features in the last three minutes are so there is no instruction in the code whatsoever to coordinate or collude and this collusion is achieved without any form of communication, okay? Now, again, this is a simple game. We're not drawing any conclusion for this. The only conclusion that we're drawing is that we should spend more time investigating, okay? But suppose that we find evidence that indeed algorithmic collusion is a problem. How do we tackle it? So, my last three slides are on a policy approach. So, one thing that we could do is just stick to the current policy, okay? What would happen then? Now, clearly here, there is no intent, no mutual agreement, no meetings or mine or mutual understanding whatsoever on cooperation, right? The algorithms don't even know that they're playing against each other. And whoever chose those algorithms, they did so because they were promised to maximize profits, okay? They were not necessarily seeking any sort of collusive agreement. So, the current legal definition of collusion should be in a way updated. But also, the evidentiary standards have a problem, right? Because right now, what we're looking at, you know, what we're trying to find out is evidence, communication, explicit agreements. But again, here, there is none. The presumption that is challenged here, in a way, is that it's very difficult to achieve collusion without communication, okay? And so, sticking with current policy is a bad idea because under current policy, this would be perfectly lost, okay? Now, the second thing that you could do is what we call the sandbox approach. It's a regulatory exandee approach. And the idea is, you know, whenever a new product is on the market, say, the Q-learning epsilon-gritty deep algorithm, then we try it out. We test it in a synthetic environment. We see, we don't see anything in a code that suggests collusion. But what we see is the kind of strategies that emerge from having this code interact with other codes, okay? And if those strategies have a flavor of something that we do know, collusive strategies, then we blacklist them, okay? This is something that we could do. Joe Harrington is one of those who proposes this approach, and then there are also legal scholars who would favor this. Now, the problem, if you want, is not only that this is very intrusive, but it's also problematic to implement because, you know, something that works well today might turn out bad tomorrow when it's interacting with some sort of new innovation on the marketplace, but then how do you retroactively blacklist something on which, you know, there's been heavy investment in production? Again, it's an option. Maybe we should discuss it. Another thing that we could do is just stick with the exposed intervention, but sort of rebalance the, reconsider the balance between explicit and passive collusion accounting for the sort of higher rate of false negative. So finding better tests that would feed this problem, okay? And I guess I'll stop there. Thank you. So it's a great pleasure and an honor for me. Remind me that I was an economist one day to speak in front of you. My first point is that I think that for economists like you, La Post is a perfect Schumpeter case. It does mean that in our group we have business which are in attrition, business which are slowing down, business that are on birth, and business that are in acceleration. So we are a perfect Schumpeter case of business. All our objectives is very clear to reinvent, to refund the old postal business model. And in doing that, we are more and more crossing platform in our business life. That's why we're saying that working with platform, working for digital platform is clearly a big strategic issue for us, which is leading us to this question, should this platform be regulated or not? My second point is that we are living with a specific platform named Amazon. I don't know if you heard about it, but you will see that our position toward Amazon is something very complex. Let's take the time to think that through. First point, Amazon is our first client. Thank you. Which does mean, which does mean that Amazon is bringing us million and million of parcels, which is just giving us the revenue to employ 253,000 people in the world. So thank you Amazon. And at the same time, Amazon is becoming more and more our first competitor, which is not so often seen in the traditional business to have your first client becoming your first competitor. And from the Amazon case, we can see very positive impact and some questions. What are the positive impact? It is the convenience for the clients, for million and million of clients, maybe you. The second one is the quality of service, a huge level quality of service, which is stimulating us, which does mean that we have a room of improvement for our postal services and also this fantastic ability to innovate everywhere. That are the very positive effects of Amazon for everybody, for the whole society and also for the post. At the same time, there are some questions. And as it is a client, I will not speak about negative impacts that are questions. What are the questions? The first question is the financial power of those platforms in the world. They are becoming those five or seven platforms, the biggest companies in the world. And in five years, it will be the case, I think. Chinese and American companies, the digital companies will be the biggest companies in the world, which is not a concern, but which is something like a question. The second thing is the market power, which is in the hands of the platform. From my point of view, it is a big issue not only for the company I am in charge of, but also for economists like you. Because when there is a concentration of power, it could, in the human history, lead to an abuse of this concentration of market power. That's why, leading to my third point, we think, and I am telling you and sharing our view on that, that those platforms should be regulated, not by ourselves, not by ourselves, by the European Commission, by the judges, but I think that we could not just rely on very positive impact of those platforms to say, there is no need of regulation. There is, from my point of view, obvious needs of regulation. Because of the market power in the new digital paradigm that those platforms have reached. How to regulate, we don't have to say how, because we have a clear, not common interest in this question. However, ex ante, ex post regulation, there is a need of regulation. And I think, and thanks to Jean Tiron and his team and the team of TSE, we are working on that. Key question for the future, what is the best regulation for common good for this platform? I think it's something which is really not only a big issue, but intellectually something which is very, very interesting. Very interesting. I don't know how, but there is a need of regulation, clearly. My fourth point is that to regulate or not could be questioned. I am in favor of regulation, but I am leading La Post, so I have clear interest for a good regulation. There is a point where I think it is no question, it is the tax element of the regulation. Because as we are competing with Amazon, we have to see that they are not paying any tax, not tax, but a very low level of tax in comparison with La Post. And as they are becoming more and more our competitor, you think that you can see that this is a big issue for us. We are paying taxes and we are happy for that. And more taxes are more happiness for us, because it would mean that we are making more money. Which is describing us as a really sustainable development company. But at the same time, some competitors are not paying taxes and clearly that is a big issue. The good news is that not only the French government, but also the European Union, is going in a direction to set up a tax framework for those platforms. Because that is really hurting the competition and the fair competition. So in those four points, I explain and I try to share with you what we are doing in changing our business model. And between in the digital era, and not only in the digital era, but in the digital business, those businesses need to be regulated, especially on the fiscal side. In my conclusion, I will not quote an economist, but a very young man who in 1548 wrote from my point of view a key speech. His name is Etienne de la Boécy. I don't know if the non-French of you know Etienne de la Boécy. And if you don't know, clearly this night you have to read what he wrote at the age of 18, and which was the speech of voluntary servitude. This is a key speech, a key note in the human history. What is the relationship with the digital world? You will see. Etienne de la Boécy is telling us, the human beings, that if there is a servitude, it does mean that people have chosen this servitude. And the digital platform has given us convenience, a lot of convenience. My question, and please I ask you to follow my advice to read these speech for voluntary servitude. My question is about the price of these convenience. We have more convenience thanks to all those platforms. We have more services. What is the price in terms of freedom, sharing intimacy, sharing data? So is there not a voluntary servitude in the digital world? That will be my question. Thank you. Thank you so much. So now we know what the European Commission is going to do. Verner Stang is head of unit of e-commerce and online platforms, if I'm correct. And he's going to tell us about what regulation is going to look like. Thank you very much and good afternoon everybody. Online platforms, that's my passion and my profession. It's just fascinating, I mean, what platforms have been doing to our lives and will be doing is just fantastic and scary at the same time. Now in my short intervention, 15 minutes I was told I'll do three things. First, a few words about the platform economy, some characteristics. Second, some challenges that arise from that and what the Commission or Europe is doing in general about it. And thirdly, very specifically on a policy proposal for regulation that comes out next month in the April where we're trying to set some framework conditions for relations between business users and platforms. So unfair trading practices in the platform world. Now starting with the platform's economy as such. I mean, this is a room full of economists that don't have to spend much time on this. But we all know what the platform's economics looks like about the multi-sided markets, the speed and ease with which they grow. The network effects, the importance of having more users each time on every site to attract even more users. The extent to which this is further accelerated by data, network effects and the results that we all know is of course that often referred to as the winner takes all principle that you would in specific sectors of the economy have one, two, three maximum really successful platforms. Now this is just an observation. It's not a problem yet because as also Mr. Vaughn has rightly said, there's of course numerous advantages arising from that. So also in Brussels in spite of probably even an avalanche of policy initiatives that one way or the other tackle the platform phenomena, we don't have an anti-platform reflex or rhetoric here because platforms have immensely contributed to welfare if you want in many different regards already from pure market language the way in which they have made many markets much more efficient by bringing supply and by matching supply and demand much more efficiently than has ever been the case in the past. I mean when I grew up in Austria quite some years ago there was no internet and even when I lived in Vienna in the biggest city I just had the choice to go to three or four or five relevant shops where I got there or I got otherwise I didn't even know what existed elsewhere nor would any supplier from Toulouse have ever gotten me as a customer sitting back home in Austria. Now as we all know in many regards in many different fields of the economy and society those platforms have been bringing together people. That's good for businesses. Again talk about this small Toulouse-based, small-sized enterprise could never actively go out and serve ten other countries in Europe let alone across Europe. Doesn't speak the language, doesn't know the market, doesn't have the logistics in spite of La Post doesn't have any ability to deal with the after-sales services elsewhere whatever returns and so on and so forth. Now jumping on eBay or other platforms allows those SMEs to go out and sell to many more. Consumer we don't have to talk about the consumer. Also Mr. Weil referred to the convenience. How easy it is for us to buy on Amazon and to communicate on WhatsApp and to go on Facebook and see our data share. But who cares about data sharing? We discussed this over lunch when I talked to my teenage daughters they don't give a damn about Cambridge Analytics and then I don't know where Cambridge is nor what the analytics means. All they care about is the ease with which they can communicate with all their friends. So lots of advantages. But at the same time of course there's lots of challenges and we are fully aware of those. Again starting with economics. On the one hand those platforms have increased competition through this matchmaking and greater transparency of supply and demand. But then of course the two biggest issues are market power on the one hand and the information asymmetries on the other that put at risk that markets function as efficiently as they should. Now just a few challenges in what Brussels is doing. Of course you have this concentration. Winner takes it all, concentration, increasing dominance of some players. Yet they may not reach any thresholds on the competition law. What is a relevant market? When are you really dominant? How much are data worth? There are lots of open issues there which I've heard a very interesting presentation from Emilio on the limits of competition law when it comes to collusion where nobody actually intends to collude. That's already where competition law enforcement ends. You saw Brussels going forward against platforms. I mean Vesta, the commission of Vesta has been quite active with the Google case and Apple taxation and Amazon and so on and so forth but there are limits for competition law. There's a lot of talk about level playing field fair competition and there are some and I will come to this proposal of next month in a minute. There are others as well as the copyright reform which is about the fair remuneration of rights holders if their stuff is being sold on YouTube without them getting any money for that. There is the issue of audiovisual media services that strictly regulate the old television, what they can send, what they can show and what they can't show and when and how much advertising is allowed and when whereas all the social media of course can pretty much do what they want. There's the taxation discussions. Very, very important digital taxation is being discussed by member states as we speak. Should the taxation be due where the value is created or where the company is established and taxed and some other examples. Also society of course, we had interesting presentations this morning on fake news or on illegal content, hate speech, terrorism and so on and so forth. So how do we deal with those issues at this stage? It's much more the soft law side of things in Brussels because someone mentioned this morning's censorship and things like that. Very, very delicate issues. Data protection was mentioned here. We have the GDPR and the ePrivacy Directive. We have consumer laws that are also meant to protect consumers as a revision of consumer laws as we speak. For instance, when you go to a platform as a consumer you should be much clearer from whom you actually buy from the platform, from the third party seller what happens in terms of liabilities and so on and so forth. So there are lots of things going on and I'm skipping lots of those. What we are doing next month is looking into the relations between the business users and the platforms because those platforms obviously with the growth of the digital economy, generally speaking the digital world is growing and within that the platforms are growing and with that the dependence on those platforms is growing to reach your customers and that is either you are an eShop or some shop some of these selling stuff eCommerce and you sell it through a platform like Amazon or eBay or you're a hotel and to be seen you need to be on booking.com or Expedia. You're an app developer and you need to be on Google Shop or on the App Store but also Facebook increasingly turning into a marketplace and for many businesses to reach customers they need Facebook to reach them, that's where the customers are. So there is this dependency and at the same time of course this imbalance in power and size and compared to those platforms even large companies are dwarfs but when I worked on postal issues a few years back I thought La Post was big and I was defending your small competitors against you but now if I deal with the Google and Amazon and Apple and Facebook of these worlds all of a sudden La Post becomes not so big anymore or Spotify against Apple because they are dwarf and highly dependent on Apple so we are looking into these imbalances and dependencies and we're starting from a contractual relationship point of view because those platforms have terms and conditions and you are now a business that uses, that needs that platform you're not negotiating terms and conditions with them Apple is not changing their terms and conditions for you when you come as a new app developer nor would booking.com do that for a small to lose space hotel so you have to sign up to those terms and conditions that don't give you any rights, lots of obligations and if you don't like that you can go to court in the US so we are starting from there and what you will see next in next month's proposal it is still probably not going far enough for some who would want a much more muscled intervention we are starting from the basis of that we know the problem but we're not entirely sure of the solution because these are two very different things establishing an issue out there but then coming with a policy response that does more good than bad I mean that's the art of lawmaking and we constantly fail in this I guess inevitably so we're cautious we're starting with transparency I told you at the beginning it's market power which is the driver and transparency which is part of the solution so we want to increase predictability that for business users signing up to a platform it is much clearer who that platform is what that platform does, how it interacts with you so they have to have clear terms and conditions while it's obvious, well it's not actually but that's part of it they have to be very clear terms and conditions once they change them you have to be notified in advance not overnight but you have two weeks to adapt to this you have to have some basic information on how ranking functions of course we're not asking platforms to reveal their algorithms that would be counterproductive for many other reasons for not doing so but some basic understanding of what is it that determines ranking if I'm a hotel and I'm on booking.com what are the key parameters used by booking that I know especially if they're paid for mechanisms for instance there's a preferred partner program but if I sign up to that and pay a bit more commission what does that actually mean to be found and then these days it's need to be found on the internet we want basic transparency on data policies of the platforms we're not asking them to share data many ask for this, we want data from the platforms I don't have the time to go into the reasons they should be very delicate and dangerous but still we want platforms to be crystal clear about what data they share don't share on what basis they share them same for discrimination one of the biggest concerns were platforms are not only intermediaries but compete with their third party business users there is of course a big risk that they could use their privileged knowledge data artificial intelligence we've all heard of that to the detrimental business users again here we want platforms not if we don't want them we force platforms to be much more clear in what ways they may they may not give preference to their own services compared to other services so we're not saying you're not allowed to do this we're saying if you do give any type of preference that has to be spelled out so that it can also be watched more easily the second pillar and I'm coming to don't worry to the end in two minutes the second pillar of our intervention next to transparency is redress what if things go wrong you suffer as a business user from some practice of a platform you're not going to go to Silicon Valley or to Seattle to go to court there especially not as a European SME I guess so we want first a better internal complaint handling system there must be a mechanism whereby when you face a problem that you can turn to that platform and within 48 hours or so a solution can be found that's very important take one example which is the most harmful practice we've experienced that's delisting that means you're no longer on that platform that can be deadly of course if your turnover depends 80% on platform X and platform X closes your account doesn't really tell you why doesn't give you any remedies how to address this it happens four weeks before Christmas and this is not an invented example it's a real example that can be the end of your business if you're an app developer without an app store you're also dead so therefore we are saying within the terms and conditions you have to spell out clearly the rules of the games welcome to my shop but if you do one, two, three or four you will be out that must be clear up front now when you do delist me I took your shot on your shop because you sold counterfeit products now if you did sell counterfeit products you're probably going to shut up and try to sell them somewhere else but if you didn't then at least you have access to this quick complaint handling mechanism you can go back to them and not just sometimes you receive an email triggered by a robot decision taken by an algorithm there was not even somebody sitting at the board of Amazon and says where are we going to take her down it's much less spooky or more spooky so therefore finding a mechanism to discuss this quickly and show them evidence no it didn't do that and then you're back up again it's in the mutual interest second stage we promote mediation so if that doesn't work out that there is some independent mediation possibility in Europe to do this out of court because access to court not only whether it's in the US or here but the court is always expensive takes always a long time and there's a huge fear factor the number of businesses and associations that have talked to us saying we don't dare to complain because we are dependent on that platform we don't even want to know but them to know that we are complaining so therefore they also won't go to court but probably they won't go to mediation if they don't we're going to introduce a mechanism of collective redress whereby associations on behalf of their members can take a platform to court for non-compliance with our regulation so not to sort out an individual contract issue so there's some enforcement element and the third and with that I'm closing that's also relevant to anybody in this room and to this sort of events where we're trying to increase knowledge of what's going on that's why I welcome such events very much it's an observatory it's monitoring, it's watching not only in implementation of our regulation but the platform economy as such what's going on out there even the things that we only regulating at the transparency level today discrimination, data algorithms and so on that there will be much more monitoring of what actually happens on that market what the effects are which would then allow us in a couple of years time to come back with much more certainty on finding the right answers that we already know today thank you very much thank you so much Veranera the three presentation have given us a lot of food for thought let me comment a little bit on that and then we'll open the discussion to the floor let's start with Emilio's paper it's there is very little work on algorithmic pricing and the impact on collusion and what it implies as you know the theory of collusion in economics is kind of rational collusion we you know the assumption we make actually is that the firms actually if there is an opportunity for collusion they find a way of colluding so they already reach this level of optimal collusion actually if you look at the guidelines for example the merger guidelines and you read what might be conducive to collusion actually to tacit collusion they actually take a rational perspective from that and actually a number of people from Toulouse actually have contributed to those guidelines we know much less about what happens when people don't find a way to coordinate and that's where it steps in now there are two things to be distinguished the fact that you monitor in a continuous time almost of course you can react much more quickly and also the fact that you have statistics called algorithms that help you solve the signal extraction problem because when you see your demand decrease it might be the case that there might be some secret price cut from the others in the case in which you don't observe prices or it might be a demand shock and so on and all that you can do that very quickly and more accurately so that should favor collusion but it has nothing to do with AI in a sense it has to do with computer but not with AI and actually I would say it's a little bit of a conjecture it's going to bring us closer to the standard economic analysis in some way now in practice those games are played by humans or algorithms so humans have only limited capability limited intelligence and their behavior actually so for example you could have hyperbolic types who actually want to make money right now as opposed to revengeful types who actually are willing to bear a loss for a long time so it's complicated to predict human behavior in some way and they probably use a simpler strategies and algorithms also they might use tit for tat or very simple things like that which actually may apply to your game it's easier to learn but they also have a harder time to learn themselves we know very little so I found that very exciting that you and others are working on the topic because again we know rational collusion but this kind of stuff is still a new world on the presentation by Philippe and Werner so you start of course from the point of view that we have more and more monopolies nowadays I mean that's the technological fact direct and indirect network studies and that means that actually much of what we knew about regulation is actually obsolete unfortunately what has served us well for 100 years now you know the kind of market definition, price regulation profit regulation and even structural policies have gone down the drain structural policies I mean it's actually a difficult thing we have to reinvent that because it's easy to break up some kind of electricity company right you know there is generation transmission and distribution roughly you know and still it takes years but you know nowadays it's very hard actually for most I mean I'm not saying it's impossible but for most to actually break up things technology is moving very fast and it uses the data so all the services use the same data also so it's not that easy to break up the firms also we could also monitor acquisition a little bit more I mean we are talking about Amazon Instagram, WhatsApp I mean those are social networks in some way they were not defined as social networks but you know we might think about there might be future competitors to Amazon of course contestability is very important and that has been mentioned by Philippe contestability is basically the idea that you know we have no choice we are going to end up with monopolies or tighter legal polis and what's important is not to try to create competition, artificial competition but rather you should try to make sure that if you have efficient entrance they will be able to enter and they won't be blocked by the incumbents and if that the case actually Drew Foodenberg and I wrote a paper on that 20 years ago it might actually induce actually the incumbents to charge low prices so as to build a big install base of consumers and also to innovate now this is a very idealized thing so we don't really believe in it of course and let me add one thing often if you look at how people enter, how firms enter market they don't enter the entire set of segments they are not like Amazon or Google today actually Google started with just a very market niche which was a search engine or Amazon was selling books it's not the Amazon of today so you enter on the market niche and then you expand and Uber is just trying to do that of course and they all try to do that so it's very important to attention to tines, to bundling or in a milder version preference for all services it's important to check possibilities for multi-homing and there are lots of things we have to look at and economists have looked at a lot at best price guarantees the most favorite nation closes which almost every platform uses right which looks good on paper when you teach your students watch out intuition is not always the right thing so you give them your best price guarantees booking is going to tell you I have all the hotels on my website and you will get the lowest price on the platform and you cannot imagine a better deal accept that of course that once you can offer that because you have asked for a best price guarantee from the hotel then the consumer is going to be totally loyal to you and you are going to have unique customers and then you can go and see the hotel and say you have to pay 25% if you want to reach my customers and that has nothing to do with market share I mean people think it's market share that counts in that case but not at all actually if you have market that's enough your monopolist on all your customers so that's the kind of thing you teach to the students just so that they get the intuition that intuition is actually not always the right thing to use but you know this is very important we have had cases in Germany for example in France Amazon has had cases of they all use most favorite nation closes so unless you intervene they are going to keep using that and ransom basically all the merchants and that's what they do and they also bring a very nice service I mean it's not like we should expropriate them either but it's going to be all over the place so how many of you have a personal assistant Alexia or Google Home okay wow this is lots of Europe I guess because in the US it's very big Alexia is not sold yet now but Google Home is and you it's pretty scary by the way in terms of privacy but that's another matter but of course the most favorite nation close is going to play a big role because you say Alexia find me a doctor for my foot or whatever deliver me that meal or something like that and you know they will probably you won't have Google Home and Alexia at home you won't multi-home right you'll just have one of them and you'll be a unique customer and then you can ask 35% from the doctor to be referred by Alexia that's the kind of thing you have to think about and that's important and let me just say a little bit about data I mean as an economist I found it overwhelming to think about privacy I have no clue what it tells and I have no time to read and those terms and conditions are pretty awful if you look at them and what I do honestly is I click I'm a bit hyperbolic too I want the information I click and of course I sell my entire family in home but you know I think there's a point where the rational consumer I think of myself pretty rational actually but you know at some point I have only limited time and limited knowledge so you know Libertarian paternalism appeals to me I need help from the regulator somehow I don't want to kill innovation but I need a set of simple policies I can choose from maybe by default I don't know but at the same time it should not be free so for example if I'm using Waze and I say I don't want to be recorded okay then nobody you know it's a free option then nobody is going to be recorded by Waze and then Waze will be totally useless because there will be no data in it so you have to find the right the right solution somehow it should not be a free lunchizer but we need help we need help from the regulator so that you know you keep the innovation but you get rid of the bad aspects and I think that will be my conclusion I'm very excited about this round table about this conference because you know there is so much we have to do for the next 50 years if we want to it's a new world and we are not quite ready for that we could talk about fiscal matters as well actually Philippe mentioned that we could talk about privacy we could talk about all those things so I have to get to grips with all those questions so let me first ask the three panelists whether they want to reply to each other I was very happy with Werner's orientation for regulation because I think that we need we need help and we need the regulation I don't have to define it because I am a member of the play game we need help and I feel very happy that the commission is entering this new exercise of setting new regulation I think it's very positive we also need help but we can't admit it openly no questions at this stage okay so I suggest we open the discussion to the floor I'm sure you'll have lots of questions could you speak out please so what is the name of the assistant in Google Home what's his name, her name it's okay Google it's generic and neutered to the French it might be difficult there's no law or law in this case I had a question for you based on the algorithmic complexity in some sense the deep learning algorithms are intended to maximize a multi-dimensional space if it's the case the collusion maximizes that why should we ever be surprised the collusion wouldn't be the result matter of fact I could even envision cases where there are incredibly sophisticated collusion strategies that are extremely difficult to detect imagine for example that you have an Amazon AI working with an Alibaba AI now I don't have to experiment on an expensive device like signaling with an iPhone I might signal on a penny candy item that I'm willing to collude in which case my costs of learning or probing the other AI are very small but it's such a high dimensional space that ever testing that probing could be extremely difficult but then now they start colluding across lots of different dimensions I'm surprised that we would be surprised that these algorithms wouldn't find the collusive outcomes I think we need to think more carefully about these algorithms yes I think at the moment the technology is not yet there so what I showed you is so these algorithms they don't know anything they're versatile because they're model free they don't know they don't even know that they're competing against somebody else and so my take is that maybe 10 to 15 years from now when whoever is coding these algorithms realizes that more contextual information is helping them then in that case I would expect these kind of outcomes to emerge for us it was a surprise because it was a simple it was a very simple naked version of the technology but I think we're getting there we'll get there at some point question for LaPost so if LaPost is said it's like Schumpeterian organization is there any presumption that LaPost were to compete with Amazon without regulation is there any presumption that LaPost will lose the competition to Amazon in other words why LaPost can be more innovative than Amazon and win the competition is there any efforts LaPost is doing like alternative method delivery like drones or other in fact the need for regulation is not a need for protection and I like your question I think that even in Toulouse as Amazon has now organized delivery platform here in your town we will win the competition against Amazon but the issue is not on delivery where we will win the competition the issue is that Amazon is connecting a lot of market not only the delivery market but also the market platform market but also the Alexia at home but also Amazon web services and so on and so on so we don't fear the competition and as mentioned we will win in Toulouse and in delivery clearly because we have a good organization of that the issue is the market power and the concentration of market power that is why at the same time I told you and I sincerely believe it they are stimulating us which is in fact the role of competition we like competition we wear an ancient monopoly and we are very happy to be out of this monopoly situation however the concentration of market power is a big issue not only for La Poste so on drones we have a drone commercial service in the south of France it exists today with La Poste we are working on that for some of the small French island to use drones for that so my need for regulation was a need for a fair competition not a demand for protection to be as clear as possible I have a question for Verneuchten you in your speech you mentioned that regulation should go in the direction of avoiding situations in which someone is kicked out of the platform for no reason but just for clarification would you mean that the regulation should establish a minimum security for the participants or it's more that it should be transparent from the beginning and yeah it is the latter it is about transparency and predictability we are not interfering at this stage with business models per se we're not saying these are the rules we impose the rules on the game on you these are blacklisted clauses and this is grey and this is white I don't know how to do that frankly a platform can freely decide and should be free to decide who it wants on its platform it just has to be crystal clear from the outset I'll give you one example you probably know the Etsy website Etsy ETSI American based it's for handicraft products that's huge and that's sort of the Lithuanian lady that's with handmade stockings could sell it through Etsy through some wealthy American lady or what have you of course Etsy can say we only want handicraft so the moment you start selling Chinese made mass manufactured stuff as Etsy I am and I should remain to be entitled to kick you off the next moment yeah so that's why we don't want to say what is allowed and what is not allowed it's a platform decision like any other business any other business in sector of the economy if we were trying to tell them what they're allowed to do you're crazy you know this is a market economy so what we're trying to ensure is that from the start it's clear what the rules of the game are that you're properly and immediately informed on the grounds for the listing and that you have a chance to challenge that decision that is what we are doing just to follow up on that there are always things you cannot specify exactly right there are some characters you don't want to have on your platform there are products you know are not going to sell and are going to encumber the you know the landscape in your on your platform so you want to kick them out even so somehow there's nothing wrong with them it's just that they won't be profitable at all or they may even discourage users from joining the platform so that but that then you need a dispute resolution mechanism which is what you said how do you envision this dispute resolution mechanism who is going to do it is that going to be the European commission no thank you very much starting still from the first part of your question because we cannot anticipate each and every situation that is clear even the platform itself but there are two things first is sort of the policy of the platform which of course it can describe more or less precisely but the quality criteria would be one of those I don't want this type of business on my you could say well if you don't consistently reply to consumer requests within two days or deliver within two weeks you know you can specify all sort of quality criteria to protect your reputation as a platform all of that is very legitimate to the extent that the user knows that or if it is coming back to the fake news it's Facebook that decides what it wants and allows on its website it's the terms and conditions of course there's always what is illegal that needs to go down but there's a lot of legal stuff that Facebook removes every second because they say we don't want naked bodies and this that and the other you know it's still their choice their business now in terms of redress as I said before first it's really internal to the platform they need to set in place a mechanism it's much more defined in terms of objectives than in how exactly they should do this you're not saying the human interface that interacts within 24 hours you know we don't want to micromanage a fast moving environment we want to be technologically neutral but we're saying that it needs to be a complaint handling mechanism that allows to solve and discuss those situations in a timely and effective manner so it's more principle based and results oriented legislation then prescriptive on the how the other two levels are as I said outside the platforms that we promote mediation and there's a strong obligation on platforms to engage in mediation because that has proven to be very powerful where it exists and to certain it's really the real enforcement threat will be that associations can take platforms into court in Europe because they are not doing what this regulation requires them to do so this is the escalation if you wonder of redress I have a question for Werner there's a lot it seems to be so large this field of regulation between with the aberration of platforms like privacy like fiscality like level playing fields or respect of different laws do you feel you have enough means to regulate or to be able to construct something around this regulation isn't there a problem of means for the European Commission? two years ago we published a policy document on online platforms setting out our general strategy if you want it was in April or May 2016 at that stage we had already described the platform economy and how it works and what it is the challenges arising from it one of the policy conclusions was that we refrain from proposing all encompassing platform regulation that solves all the evil in this world so in that sense we have opted more for a targeted issue by issue problem by problem response to that the challenge arising from that is between the different things but for instance if you look even into platform definitions they differ between different initiatives because we are addressing different issues so when I am talking about those between sellers and buyers I have a very different scope and I have as I said before for market places e-commerce market places to app stores to Facebook it's more functional definition of the intermediation function between business and consumers but as other initiatives that aim for digital taxation for instance for them this entry point is not relevant because the underlying problem is different so we are always starting really form sort of a problem analysis of what is the issue who is concerned and then we try to find a policy response rather than to say there is this uniform body of platforms they are all the same and all bad and let's now regulate them to death but it's very tricky because it also means that we have to make the case intervention by intervention and of course discuss with all the member states and with parliament but that's European lawmaking I can't change this for John you mentioned that much of welfare economics was down-drained and I was wondering if you meant that because of the predominance of platforms or some other reason and if you could give a couple examples of that and also what are the implications of that of worth of welfare economics thinking for regulation because I have trouble imagining regulation without some kind of something like that yeah of course we are not throwing everything that we built over the years but but it's true if you if you look at the market definition for example we have been trying to convene the antitrust authorities to think in terms of two certain markets right and not because on one side of the market you'll be charging monopoly prices and the other side you'll be charging velo prices you might think it's exercise of monopoly power and on the other side predation when you might be perfectly competitive in the end you know that kind of thing you know the most famous nation stuff also tells you market shares make even less sense than before so it's more like a matter of degree I think the price regulation same thing and you know what price cap do you want to impose I mean you know all those platforms charge zero on a number of services right how do you deal with that I'm not even mentioning profit regulation of cost of service regulation I mean we don't know how to do it and then the international aspects how I did up to that so there is a very strong need for regulation but I'm very much in line with you have to find ways of having a regulation which is both powerful and non-inclusive right we don't want to second guess the companies but we have to make sure even that we are going to have all those monopolies and we have to live with them let's not try to create fake competition let's try to make sure that only the efficient and transcend but for that there are some conditions and we have to design this so that's a challenge for us I would think okay Verano sorry back to you if on cross-border parcels activity I prefer exposed regulation for sure we need an example one for the for the platforms and you started commenting about the time it will take I mean the timing of legislation in Europe but there is kind of emergency here because when we talk to our customers the platforms the non-Amazon platform they are scared they are really feeling that the competition is too tough and some are pushed to find some alliances to try and resist and the said already sometimes that if we are not quick enough in some years Europe could be the place where Alibaba and Amazon fight together but no more European operator in the middle second question at the end of your intervention you said we need help and we know each other for a long time I know it's just not a chance you said that so here what could you could you expect from the economies or from the post to help? First on the timing people are afraid and competition is tough first of all we are not against competition per se and as Mr. Walbright said you don't want to be protected you don't want to be empowered you want others to play by the rules rather than be protected our first instinct is not to stifle competition and innovation because people are complaining I'm not downplaying it with my own story but you saw this also in the collaborative economy space whenever something new hits something old the old protests and the old usually has a stronger lobbying power and so if taxes are threatened by Uber by Airbnb that pretty much quickly tends to come up with responsive response very often at member state level by the way faster than we do that so we want to be careful as well in that we don't shoot from the hip before knowing actually whom we are going to kill that's just a normal reflex and a healthy one I think now on this specific proposal we are approaching the end of the mandate that's very much Brussels talk now but whenever the parliament must end and that's next year middle of next year that is usually when ongoing files need to be closed so starting negotiations now beginning of May gives me a year to negotiate that if that goes through that's pretty fast because also regulation which means directly binding and will intend to force after a few months after that so if I get parliament and member states on board and there is a likelihood of that happening because what you said was true and this is when I talk to members it's really now to many of them to a greater and lesser extent they are agreeing with our analysis same parliament so the negotiations will be more a bit more here a bit less there but it could be for once that Brussels comes and it's the first worldwide first platform regulation of this kind looking into so I know we are always slow but I'd rather be slow and get it right than kill anything else what was the second question again that is a very good question we talked a bit about a lot actually about artificial intelligence and algorithms and this that and the other and I also ended my first intervention by saying we first need to monitor, we need to need research, we need to understand what's going on and this is precisely where people like yourselves in this room come in and it was encouraging research really presented this morning which of course only looked at as you would have to do in an individual project it would always be narrowly circumscribed and then there is 25 assumptions that limits its value at the end of the day to draw policy conclusions but the more of this there is the more you build on previous colleagues works and try to bring light into this because that's the main problem we are having is nobody understands what's going on and that's also what scares people by the way also citizens if you think into algorithms increasingly they take decisions they take decisions on how much we pay for things but I use an iPhone or an Android or what have you where I live they decide whether or not I get insurance cover I mean I manipulate that whether I get a loan they will increasingly take many many decisions for us and people don't know how these decisions are taken now we will not be able to open up algorithms and you know like a dead animal so we need to think about how we can make things more transparent and more accountable and the transparency how does this work actually and what can the results at the end be explained and how can they be explained so the more research in all of these issues is done the more confident will we feel when just last remark on this Sean you mentioned MFN clauses now repeatedly even there no matter how intuitive it may be or not currently the evidence is still inconclusive about the welfare effect of this competition authorities all over Europe including the commission have looked into those hotel price parity clauses they are exercised by booking until today there is no clear conclusion what the ultimate welfare effect on consumers is actually good or bad it's not clear although people may think well that's obvious so that's one reason by the way we're also in this regulation we have not banned MFN clauses but force platforms to be more explicit about those so there's almost no area data including where not more research would be absolutely wanted well let me just respond a little bit on MFN I think it's also our fault as economists not to have design policies and we are we may think 25% maybe too much but is that 14% of 12% or 20% we have no clue honestly for credit cards we did that we had a rule which is called the tourist test which basically gives which is used by European Union which actually tells you how much you can search out for example you know an upper bound and MFNs are not completely inefficient either they also try to prevent show roaming you know you go on to booking and then you know it's going to be cheaper on the hotel website and then you go to the hotel website and booking that's a reverse and it's completely expropriated from its investment so you know and I think it's up to us to understand the help here to actually come up with rules so far we have used you know limited MFNs or no MFNs like in France but those are not perfect rules and I completely agree with that let me just push a little bit on transparency so you mentioned Alibaba but Alibaba landing is landing a lot to SMEs it's landing a lot to SMEs because Chinese state-owned banks basically are instructed to land to state-owned firms and Alibaba does its SME landing mostly through artificial intelligence there is no people actually granting the loan it's actually the way it works and often those loans are about $2,000 or less or $3,000 or less millions millions of loans actually I don't remember how many SMEs they have but it's a huge number what's bad about that I mean if you make it transparent then the entire business model I don't want to defend Alibaba but the entire business model is actually becomes public and then it loses a competitive advantage so that raises the question there must be some market failure in your mind saying if it's a machine who does it I said who something wrong and that's something that might be cases but it may not always be wrong no clearly not and I don't have anything on my mind probably also because it's Friday evening on algorithms for instance we're just launching now as we speak a one and a half year project sponsored by the European Parliament it also said algorithms are very important and there we're going in without any preconceived ideas or solutions with the chosen consultants and the large inbuilt stakeholder involvement process because we don't expect any single expert or group of experts to have the answers to look into the space and to see what are the challenges the issues are raising from algorithms and algorithmic transparency and from that gradually then there's still a few issues and then see whether some response is needed artificial intelligence in the same end of April package I was referring to there will also be a paper on artificial intelligence but it's not coming with solutions but starting a bit the debate and raising the awareness on some of the challenges and issues so this is on all of these issues at least at the European level we are very cautiously trying to find some light in the dark no preconceived solutions I understand it's Friday afternoon but I'm sure there are more questions because we have those three great participants and I'm sure you have questions for them no? so we have to call it a day I first would like to thank a lot Emilio and Philippe and Werner for this great contribution thanks so much