Can artificial intelligence predict where, when and how we will die?

“Our algorithms can predict your death with 96.5% accuracy,” says Peter Isherwell (Mark Rylance) to Dr. Mindy (Leonardo DiCaprio) in Don’t look up: cosmic denial. A few scenes later, the CEO of the technology company Bash Cellular predicts the President of the United States (Meryl Streep) who asks him for details on his end of life: “You are going to be devoured by a brontéroc”. For those wondering, it’s a nasty bug.

The satire signed Netflix humorously denounces the inaction of governments in the face of the certain extinction of the human species and does not forget to criticize at the same time the giants of the tech, more interested in personal enrichment than in salvation of humanity. Peter Isherwell plays a technoprophet, halfway between Elon Musk and Mark Zuckerberg, convinced that his technology will save the world. If he is wrong on this, the end will however prove the predictions of his algorithms on the death of the president. In fact, the Gafa algorithms (including Google, Amazon, Facebook-Meta and Apple), stuffed with data thanks to our digital traces, have spectacular prediction capacities. But can they really predict our death?

“Don’t look up” has it all wrong

Artificial intelligence reasons by induction. It starts from the particular to derive a general rule from it. “It works well if there is consistency. It is certain, on condition that what is observed really corresponds to what we will observe in the world which is going to happen, explains Jean-Gabriel Ganascia, researcher in artificial intelligence and president of the ethics committee of the CNRS. Are we sure that what happened in the past will happen under the same conditions? If it happens under the same conditions, there is a chance, in probability, that it will happen identically. But if the conditions change, that will change ”. And we have no certainty that the world remains frozen.

From this point of view, Don’t look up is wrong across the board. When Peter Isherwell arrives on a new planet accompanied by twenty people including the President of the United States, the context has changed. The conditions at the end of the film have nothing to do with the beginning. The vegetation, the oxygen level, the animals… Everything has changed. And the algorithm could not predict it before the arrival of the comet, it is not a magician. “It’s funny because it’s absurd, but if there is one case where it cannot work, it is this one”, smiles the specialist.

Google can predict that a woman is pregnant

Let’s move on to the errors of fiction, algorithms still know a lot about us. They don’t make it up. They just have to observe all the digital traces that we leave through our research, our interactions on social networks, our online purchases… Soon we will even be able to pay with our mobile phone. The stakes will be enormous, because, if the Gafa are so keen on personal data, it is first of all to resell them to advertisers, not to predict the death of Internet users. It’s about making money through targeted advertising. From the research carried out, Google knows that such and such a person is a young woman, such another an old man, or a divorcee, or a single person… And thanks to this information, advertisers can increase their rate of return.

And little by little, the algorithms end up knowing more about us than ourselves. In 2019, Roger McNamee, early investor in Facebook, and author of Facebook, the disaster announced (Quanto), affirmed at 20 minutes that “Google can predict with 90% accuracy that a woman is pregnant before she knows”. Better (worse?): “Some AIs can detect a person’s homosexuality, sometimes, even before the person defines himself as such,” suggests Guillaume Chaslot, an algorithmic who worked on the YouTube and Google algorithm. In 2017, a study by Stanford University, relayed by The Guardian, showed that from a few photos, an algorithm could correctly distinguish a person defining themselves as homosexual from a person defining themselves as heterosexual, with a success rate of 81% for men and 74% for women.

AI knows our state of health

They are able to perceive weak signals, invisible to humans, thanks to the gigantic mass of data they ingest. “According to Google, its algorithms can predict epidemics,” adds Jean-Gabriel Ganascia. They detect the increase in the number of queries such as “how to cure a cough”. However, they cannot say whether it is a flu or Covid-19, for example ”.

As for our death, things get a little tougher. “AI knows our habits. If you ride a motorcycle and are used to speeding, she can calculate a probability. Likewise, it knows our state of health, our alcohol consumption, ”notes the artificial intelligence specialist. Bet that a smoker is going to die of lung cancer, ultimately anyone can. We are still far from the promises of Peter Isherwell in Don’t look up, And that’s good. Imagine a world where insurance can know the day we die. No, let’s not imagine, it’s better.

source site