Trump and the Netflix effect

Trump won the election. Aside from one in the Los Angeles Times, no other poll predicted he would win, no survey, no other newspaper. According to him and his supporters it was because there was a media plot to silence them and that the polls were rigged. Not long before the election, the conservative analyst and Donald Trump supporter, Scottie Nell Hughes, argued for a conspiracy on CNN: "The only places where Trump is losing is in the media and the polls. He is not losing with the people who go to the rallies, he is not losing on social media where Donald Trump has more than double the followers of Hillary Clinton on each social platform."

Is it true that Trump has more influence than Hillary on the net? Let's look at a few numbers. On November 9, the date we knew that Trump was the winner, Trump had generated twice the number of Google searches as his opponent in the previous year, he had 13.5 million followers on Twitter compared with Hillary's 10.5 million, 13 million likes on his official Facebook page compared with 8.6 for the Democrat candidate, and on Instagram they were level at 3.3 million followers. In general, the messages generated by Trump and his followers caused more reactions —retweets, favorites, responses, commentaries, wall shares— than those of the Democrat candidate.

Yet that is only a qualitative measure and not even a necessary condition for winning the election. To be clear, I could have 1 million followers on Twitter (it is easy to buy them, I can assure you) who would give me a ton of likes and retweets and who could even click the link to my article and turn it into the most-read of the day (also easy). As long as you do not leave the digital universe, I, my ego and the head of VIA Empresa would all be very happy. Our problems would begin the day we decided to publish the articles in a book (it is just an idea, boss) and asked my followers to buy it: not a single copy would be sold.

It is true that both candidates had bots to pretend to be followers and to amplify the messages of each candidate. A study by Oxford University estimated that a third of the tweets in favour of Trump were generated by bots and a fifth of Hillary Clinton's as well. Yet, Trump did not win because he mobilised his followers —bots included— on social media but rather he won because he mobilised them on the day of the vote.

We thus have millions of messages, reactions, photos and videos that the followers of one or the other constantly share at all hours on the net. How are we to separate the wheat from the chaff? How can we know whether a user who retweeted one of Trump's messages is a supporter of his or did it just to make fun of him? Here the big numbers do not help and we need a certain intelligence, in this case artificial. An example of Artificial Intelligence applied to decision-making is the MogIA algorithm, which predicted Trump's victory based on 20 million samples from platforms like Twitter, Google or YouTube. MogIA knew how to find the electoral behaviour patterns in the Big Data that the voters for one or the other candidate had voluntary shared on the net. On its curriculum, MogIA also has spot-on predictions for the past three US presidents and the winners of the corresponding primaries.

In the end there were enough of Trump's supporters on the net who shared enough relevant information in a large enough quantity for the Artificial Intelligence algorithms to extract the relevant information and no one even realised it. On the local scale, you could look back on your Twitter timelines, Facebook walls and YouTube channels. No one was in favour of Trump in the same way that no one was in favour of Brexit or against the peace referendum in Colombia or intending to vote for the PP in the last Spanish election.

No one saw what was happening because no one knew where to look, aside from algorithms like MogIA. Instead of asking about voting intention in polls —we are born liars— it is better not to ask and to deduce intention according to the public positions we take. It is very easy to lie to a binary question in a telephone poll and very difficult to do the same in a sustained and coherent way over a year with everything we post online. And here the algorithms have a clear advantage over humans.

Netflix can help us understand what happened. When we log in to the popular video service for the first time, it asks us what type of films we like and, as we do when expressing our preferences in public, we tell it what we would like to like: auteur, European, social, documentary, when what we end up watching are action films. And Netflix know this. Through its Artificial Intelligence algorithms the system gets to know us and recommends content it is sure we will like (the business works) closing us in a comfort zone it is difficult to escape. The more action films we watch, the more it recommends to us and the more it recommends, the more likelihood we will end up watching action films.

So, this is what has happened, a double Netflix effect. On the one hand, those polled were not clear about their preferences, and on the other, we users have stayed in the comfort zone of our timelines, Facebook walls and favourite media, reading, watching and listening to what we like in a sort of determinist vicious circle à la Netflix. In the end the system, which knows what we actually look at, has ended up recommending Torrente.
Avui et destaquem
El més llegit