Alexandre Menais, "The Implications of Big Data for Competition Law"

Presented at the Legal Challenges of the Data Economy conference, March 22, 2019.


ALEXANDRE MENAIS: OK. So everyone, I knew that it was not easy to talk the last-- after all such smarter speaker than me, so I'm going just to-- because you know we have indeed talk about the data, so after the thousands of data that-- the previous speaker gave to you, I'm going just to try indeed to make you a link with that, and to adapt my speech-- to avoid any urgency and just to show you the view of a company. So no sexy title as well, except the subtitle, which is from big to gigantic data. Because indeed, that's what--

PRESENTER 1: That's what we were missing.

ALEXANDRE MENAIS: That's what I wanted to talk about.

PRESENTER 2: That's funny.



ALEXANDRE MENAIS: All right, thank you. Yes, OK. So sometimes, to be the last speaker give you opportunity. So seriously-- so the volume of data indeed is exploding, no doubt. In the coming 10 years, the volume produced by all of us will be multiplied by 30-- three, zero-- reaching one yottabyte. I'm sure some of you did not know this word. By 2050, which is a subdivision of bytes, right? So as far as I know, currently, there is nothing by the way that can measure on the yottabyte scale today.

But you know most calculators can also-- even actually, most calculator even display zero that size. And even, as of today, as we were talking together, the current IT environment that you have on the planet, aren't that enormous yet than-- the one yottabyte that I mentioned. So the time for planning of such volumes of data coming faster than most of us realize. And to give some food for thought for all of you-- I see, and we see, an acceleration of this, because of what?

Because we are entering also, for those who are familiar with this world, in the post-cloud area, because of IoT, smart device, machine, and indeed different industry. So IoT is going to transform everything. So when we are talking about big platform, and things like that, now the big topics would be probably more IoT. So example of what we have seen-- because indeed, what is interesting is that now, technology-- of course we have seen that, you know that, you are familiar with that. It all helps you to collect data and generate usage.

But what is new in that now, is that data themselves are creating usage. And indeed-- and particular with algorithm-- algorithm is a way indeed to structure the data, and I will come back to that. So the main driver of digitization, you remember-- new usages and technology. I show you some examples of technology that we see. You know that in particular in my company we are indeed keen also to work on content computing, and I can deluge indeed, it will generate again, again, more and more data. So when I'm talking about the deluge of data is just to say that previously, as we have seen, the volume of processing data were quite limited.

Business rules were essentially-- and statistician, we're quite deterministic in the way that we are structuring the data in the choice of the setting. As the generation of the data becomes more and more important, the volume of data, as you can imagine, become gigantic. New information can be autoextracted from the combination of data-- and random spoke about this. So before, the statistician were sampling to define their algorithm parameter, now the algorithm can be adjusting their setting themselves, so we have the self-learning models.

In other words, the whole presentation now of the data is key, and it's come the point-- for economists and lawyers-- because it is essential now to ask you one question-- what data is relevant for my business? But even in the algorithm, we need to manage also the deluge of data. What I want to say is that you remember that before most of the algorithm were created by hand, but with machine learning-- I take the example of machine learning, because this is, in particular in artificial intelligence, where we are more advanced-- algorithm create algorithm.

So the loop is the following. IT application, as you can see here-- are generating data that are used to create algorithm. Like recommendation engines, that you have. That in turn, they are going to fuel new application, that generate more data, that's allowed to create better algorithm. The example, it has been mentioned before, is clearly what Amazon and Google are doing. So the question is, how to regain-- if I can say, a bit the control on the algorithm. Let me give you the example of the machine learning-- what we have today.

Actually, in fact, in machine learning we have basically more than 10,000 machine learning algorithm. It represents-- every year now, we are creating, generating, 100 new algorithm each year. And in an algorithm, in fact, what is important is that you have three components-- the representation, the evaluation, and the optimization. So concretely, what I'm going to show you with the example of the machine learning is how-- because in fact, machine learning is interesting, because there is a kind of-- and it has been mentioned before-- a kind of maturity in this market, if I can say.

So concretely, we are going to see how indeed the development and the select of those algorithm could have some input on competition. So first of all, representation. So representation basically-- you collect the data, you process, you structure, and you analyze. For that, very simple-- before, we give you many example, on your day to day, if I can say, the multiple businesses application that you can have. So the consequences are very simple-- you can improve products and services, you can multiple business application.

Because now the decision tree and because this is what you do-- in representation, you create the model for the data analysis, the decision tree, or the sets of rules. There is multiple business application, and also you can have some new economic models. It has been said before versus competition law-- very simple. The question of security of the data; the question of-- is or not possible to replicate it; and of course, there comes the question after of scale; scope of data collection matters to competitive performance-- that indeed, things that we mentioned also previously.

Now after more sophisticated, the evaluation and prediction. So here, indeed you have several algorithm which are in competition. So evaluation-- it is essentially how you judge-- or prefer one model versus another. So in that case, you can have also specific businesses. We'll take the example of fraud, detection of fraud and things like that. You know consequence-- in that case, you collect, you analyze statistic, and after-- as I mentioned-- you start to model. So this is the modeling, indeed, of your output of data.

Indeed, you can improve your product and services-- this is what we do for certain of our customers. In fact, targeted business application, because in fact, you are capable to target more precisely the application. And of course, it can also set up some new and targeted economic models. Example of consequences for competition law-- of course, market transparency. In that case, you can attenuate the competition by reducing uncertainty and behaviors-- favorable to price competition. You can also tacit collusion that could, of course, also as a result of sophisticated machine learning.

I can give you example of what we do for instance with aircraft, where usually-- that's one of the cost effect that you need to avoid to make sure that you optimize-- your use of aircraft is of course the maintenance of the different components of an aircraft. Of course, it's easy now to predict-- by pull up all the data. But after indeed, as it is so easy to predict, you can see that after you can have, in certain organization, some collusion that could also result of some interesting machine learning that you can get.

So that's an example-- the last one is optimization. In that case, what you do is prescription. So the idea is that it's about the operational usage, and we talk about that as well-- the ability to manage the instantaneous needs, in fact of the user. In that case, you may have also some different type of application. We see that of course in the scale and in super exchange. So the consequences-- of course what you can do, you can improve the product and the services. You can be, as I mentioned, with more precision-- more adapted to the instantaneous needs of the user.

Versus competition-- here this is also probably where the whole of competition law, in my view, is very close to of course privacy. And this is where at least-- I don't know who should start if it is competition or privacy, but indeed it's important that we address the things with a mix of indeed elements. The fact that some specific legal instruments serve to resolve, indeed, sensitive issue on personal data doesn't intend that company law is irrelevant of personal data. Give you an example-- you are on the Airbus 380 on Boeing 777-- not sure-- may be two 7. But anyway.

You know that for your seat, you need to have a maintenance of the seat, after let's say-- it's just figures that I invent. 25-- no, let's say 50? 50 flights, OK? Well, it's probably more. That's what I was mentioning-- previously, in the second example that I gave regarding prediction. You know that after normally 50 flights, maybe you need to have a maintenance of the aircraft, of the seat of an aircraft. Let's say that Mr. Menais, who has a nice IoT program with France, with the alliance SkyTeam and [INAUDIBLE] for instance.

And you know when he's flying to New York, surprisingly because he's all the time sit down at the seat 12, 20, or 45. A surprising, after 10 flights, and all the time you know, these type of seats. That's because the morphology, you know the type of-- because my weight, my size, and things like that, pull up intact will help of course the company, and will help, as you can imagine, either in the IoT program, but of course also the aircraft for the maintenance of the plane. That's when we are talking about example of prescription in supply chain.

This is where we have a lot of application already. So now what's next? So it just-- and again, here, I'm not trying to find a solution-- I'm just trying to raise problems, if I can say, but what I see is that indeed regarding the big data and the innovation, you have seen the deluge of data we will have. In fact, the level of privacy for me is the true competitive parameter between competitors that will need to be to be addressed. I believe, by the way, that data collection, the storage, and the processing-- and all of them, need-- I know that it's not only the topic here, but legal analytics.

And that's why I believe there is a lot of opportunity, because follow years, because we should switch probably to a legal by design algorithm-- that's really my feeling, and this is at least the way I believe. Because the thing is that even if algorithm-- we need a triple cooperation. With data scientists, for sure-- lawyers and economists, and also of course the customers. Because in fact, you cannot give to your customer a black box-- it will never work, right? So that's why we give the algorithm, but after, indeed, you need to set up between with them.

I see that's more and more we will have commoditization of the algorithm, and that's also very important in terms of topics that we see for the future. One final word, we know, because I believe that there is a good initiative that we are taking in France, linked to big data and analytics. This is the initiative, of course, of [INAUDIBLE] to create an ethics committee for artificial intelligence to audit the algorithm,

And I think economists and specialists in competition will have probably indeed to be part of this committee, because it will be probably needed to make sure that we operate the future of idealism. However, to finish up with something much less scary than what I said before. You know that actually believe that 80% of the project of algorithm in artificial intelligence are pure alchemy. So thank you.

Big data