Blog & White

© Amazon.fr

Étant données une probabilité a priori qu'une théorie est vraie et une observation, le théorème de Bayes permet de calculer une probabilté a posteriori qui prend en compte cette observation. Il permet donc d'affiner la crédence (un anglicisme, on devrait parler du « crédit accordé à ») d'une théorie, sa plausibilité, au vu des observations. Bien que simple, le théorème de Bayes à cependant un grand nombre d'applications même si des statisticiens célèbres l'ont par le passé honni parce qu'il fait appel à un a priori alors que la recherche d'objectivité en sciences devrait bannir tout a priori pour ne se concentrer que sur les faits. Dans la pratique cependant, un scientifique demandera un grand nombre de contre-exemples avant de rejeter une théorie fondamentale sur laquelle reposent un grand nombre d'autres théories et d'applications : il s'agit bien là d'un a priori, d'une crédence très élevée de cette théorie fondamentale.

Depuis ce temps, le bayesianisme a lentement fait sont entrée dans les sciences, en particulier dans les domaines de l'intelligence artificielle dont nombre de théories sont fondées sur le théorème de Bayes, et dans les sciences cognitives : il semblerait que cerveau soit une machine à faire des inférences bayesiennes. Étrange donc que lorsqu'on fait un raisonnement explicite, on ait tellement de mal à appliquer ledit théorème.

En dehors des cas très simples, le théorème de Bayes est impossible à appliquer, car il requiert de prendre en compte les crédences de toutes les théories alternatives à la théorie étudiée, et ceci est impossible. La complexité de Solomonoff (ou de Kolmogorov) d'une théorie étant définie comme la longueur minimale du programme informatique qui permet de calculer la crédence de cette théorie (en acceptant la thèse de Church-Turing selon laquelle tout l'univers peut être simulé par un machine de Turing universelle et un programme adéquat, et que toute théorie peut être représentée par une suite de bits selon la théorie de l'information de Shannon), certaines théories ont des complexités comparables à la taille de l'univers visible. Il est donc nécessaire de calculer des approximations des crédences de ces théories, et c'est ce que font le cerveau et certains algorithmes d'intelligence artificielle.

The economic value of a Youtube video

The Youtube platform has been designed for the purpose of making all of its videos are available to anyone at anytime. Given its large amount of content, its suggestion algorithm and the ability to automatically play the next suggested video, a user of Youtube can consider that the stream of videos he wants to watch is limitless. This makes the stream a nonscarce resource i.e., a free good which, from an economics point of view has no value.

Asking users to pay for watching videos would indeed be a bad business move for Youtube, as many user would refuse to pay for watching content they currently get at no cost. Even the Youtube Red subscription, according to an Ars Technica article and related user comments, seems to be valuable not for it exclusive content, but for the extra features provided by the mobile player (download for offline viewing and background playing). Interestingly, these features are already provided by non-standard software, which is freely available. The exclusive content being apparently of little interest according to the article, the subscription seems to rather be “tax” on people who don't know they could legitimately get the same service for free.

Youtube being a business, it needs to get revenue, at least to support the cost of its large IT infrastructure. Since users are not charged any money, Youtube's revenue comes (exclusively?) from advertising (I guess that the collection of user behavioural data and their profiling is also sold to advertisers). Youtube's content therefore has value only insofar as it attracts potential consumers and exposes them to advertisement. The content itself does not matter to the platform, as long as users are drawn to it.

This is thus the foundation of the Web platforms and the Attention Economy, where the users' attention is the actual product being sold by the platforms. User profiling and Big Data is only a tool used for maximising the amount of attention being captured.

Advertisers are gamblers, you don't need to let them win

In 2004, Patrick Le Lay, then CEO of the French TV channel TF1 said “what we sell to Coca-Cola is available human brain-time.” One aspect of this is that TV channels sell advertisers the opportunity to reach the channel's viewers and attempt to influence them into buying the advertised products.

From the point of view of the advertisers, it is a gamble: they bet money (the cost of producing the TV commercial and the price paid to the TV channel to show the commercial) and hope to gain from it (when viewers buy their products because they have seen the commercial and have been influenced by it). This gamble is two-fold: the viewers may or may not see the commercial, and they may or may not be receptive to the influence techniques used in the commercial.

In social media, advertisement can be considered from the same angle: the website sells advertisers display space on its pages, and the advertisers gamble that users of the website will see the commercial and be influenced by it.

While everybody agrees that you don't have a moral obligation to buy a product after you have seen a commercial, it seems less obvious that you don't either have any moral obligation to view the commercial. For example, you could close your eyes and plug your ears to ignore the commercial; you could even use tools that automatically hide the commercial from you.

Putting the advertisers' money to uses I approve of

As I want to protect myself from the influence of advertisers, I normally use automated tools that prevent me from seeing advertisement. I record television programs and skip the commercials (automatically when the tool works as intended, otherwise manually), and I use ad blockers in my Web browser to prevent commercials from being displayed on my screen.

The difference between TV and the Web is that the TV channel gets paid for broadcasting the commercials and cannot control whether or not I skip them while advertisers on the Web pay the websites only if the commercial is being fetched i.e., only if I allow the Web browser under my control to display the commercial. In that case, my preferred solution would be to fetch the commercial as if it would be displayed, without actually displaying it. And since advertisers not only display commercials but also track the users across websites, it is necessary to isolate each commercial so that the tracking is not possible.

I would not normally care about websites not being paid by their advertisers, but in the particular case of Youtube, I use tools that allow me to watch content without having to view any commercial, meaning that the content's creators cannot hope to get payment from Youtube. I therefore dream of a tool that would allow me to channel advertisers' money to the content creators without having to view any commercial, thus letting the advertisers gamble, but strongly shifting the odds of this gamble in my favor. I consider this to be retribution for the advertisers' attempt at influencing me for their profit.

© Amazon.fr

Second tome du trosième cycle, publié en 2018.

Les enfants découvrent qu'ils sont passés dans un monde parallèle, royaume du Père Fouettard. Le professeur, qui les cherche dans le palais du Père Noël mange les mêmes bonbons et finit par les retrouver. Menés devant le tribunal du Père Fouettard, Il reconnait la voix de ce deernier, qui n'est autre que le Père Noël. Ils repartent tout vers le premier monde, retrouvent Jan et Blanche, et repartent ramener les enfants à leurs parents.