Danish researchers are leveraging artificial intelligence and data from millions of people to help individuals predict their final stages of life, raising awareness of the power of technology and its dangers. I'm thinking of increasing it.
Far from morbid fascination, the creators of life2vec want to explore the patterns and relationships that so-called deep learning programs reveal that can predict a wide range of health and social “life events.”
“This is a very general framework for making predictions about human life. You can make predictions about anything as long as you have the training data,” says David, a professor at the Technical University of Denmark (DTU) and author of a study recently published in the journal Soon Lehman, one of the authors of the journal natural computational sciencehe told AFP.
For Lehman, the possibilities are endless.
“We might be able to predict health outcomes, so we might be able to predict fertility or obesity. Or we might be able to predict who will get cancer and who won't get cancer. .But you might also be able to predict whether you'll make a lot of money,'' he said.
This algorithm uses a similar process to ChatGPT, but instead analyzes variables that affect your life, such as birth, education, social benefits, and even work schedules.
The research team is adapting the innovations that made language processing algorithms possible to “examine the evolution and predictability of human life based on detailed sequences of events.”
“From one perspective, life is just a series of events: you are born, you visit your pediatrician, you start school, you move to a new place, you get married, and so on,” Lehman says.
However, the program's release quickly gave rise to claims that it was a new “mortality calculator,” with some fraudulent sites using AI programs to predict life expectancy, often in exchange for submitting personal data. They deceived people with this offer.
Researchers claim that the software is private and not currently available on the Internet or the broader research community.
6 million data
The basis of the life2vec model is anonymous data on approximately 6 million Danes collected by Denmark's official statistical agency.
By analyzing the sequence of events, it becomes possible to predict the outcome of life until the last breath.
When it comes to predicting death, the algorithm is accurate in 78% of cases. It is accurate in 73% of cases when predicting whether a person will move to another city or country.
“We're looking at early mortality, so we sample a very young cohort of people aged 35 to 65. And then based on the eight years from 2008 to 2016, the following four years “We try to predict whether a person will die or not,” Lehman said. .
“This model does that very well, better than any other algorithm we've found,” he said.
The researchers say that by focusing on this age group, where there are usually few deaths, they can test the algorithm's reliability.
However, this tool is not yet available outside of research environments.
“Right now, this is a research project investigating what is possible and what is not possible,” Lehman said.
He and his colleagues also want to investigate long-term outcomes and how social connections affect life and health.
“Public Counterpoint”
For researchers, the project represents a scientific antidote to big tech companies' heavy investments in AI algorithms.
“They could build a model like this, but they're not publishing it. They're not talking about it,” Lehman said.
“They're just building them, hopefully right now, to sell more ads or sell more ads to sell more products.”
“It's important to have an open and public rebuttal to begin to understand what can happen with data like this,” he said.
Pernille Tranberg, a Danish data ethics expert, told AFP that this is especially true because similar algorithms are already used by companies such as insurance companies.
“Maybe they'll put you in groups and say, 'Okay, you have a chronic disease, your risks are this and this,'” Tranberg says.
“This system can be used to discriminate against us, making us more likely to have to pay higher insurance premiums, lose access to loans from banks, or lose access to public health care if we die. Yes, anyway,” she said.
When it comes to predicting our own doom, some developers are already trying to commercialize such algorithms.
“You can already see predictive clocks on the web that tell us how old we will be,” Tranberg said. “Some of them are completely unreliable.”
© Agence France-Presse