Danish researchers are using machine learning to predict aspects of people's lives, including their death

(CNN) — Danish researchers say they have used powerful machine learning algorithms to accurately predict certain aspects of a human's life, such as the likelihood that a person will die soon.

he Stadypublished this week in the academic journal Computational natural sciencesexplains in detail how a machine learning algorithm model called life2vec predicted a person's life outcomes and actions when presented with very specific data about them.

Using this data, “we can make any kind of prediction,” says Sonny Lehmann, lead author of the study and a professor at the Technical University of Denmark. However, the researchers note that it is a “prototype research model” and that in its current state it cannot perform any “tasks in The real world”.

Lehmann and his colleagues used data from a national registry in Denmark detailing a diverse group of 6 million people. It included information from 2008 to 2016 related to important aspects of life such as education, health, income, and occupation.

The researchers adapted language processing techniques and created vocabulary for life events so that life2vec could interpret sentences based on data, such as “In September 2012, Frances received twenty thousand Danish kroner as a guard at a castle in Elsinore” or “During” In her third year at boarding school, she took Hermione has five elective classes.

Lehman says the algorithm then learned from that data and was able to make predictions about certain aspects of people's lives, such as how they think, feel and act, and even whether a person might die in the next few years.

To predict how quickly someone would die, the team used data from January 1, 2008 to December 31, 2015 on a group of more than 2.3 million people ages 35 to 65. Lehman explained that this group was chosen because predicting deaths in this age group is more difficult.

Life2vec used the data to infer a person's probability of surviving four years after 2016.

“To verify effectiveness [life2vec]“We chose a group of 100,000 individuals, half of whom are alive and the other half are dying,” Lehmann explained. The researchers knew people who died after 2016, but the algorithm did not.

Next, they put it to the test. They had the algorithm make individual predictions about whether or not someone lived after 2016. The results were impressive: the algorithm was correct 78% of the time.

According to the report, Life2vec also outperformed other recent models and baselines by at least 11% in predicting mortality outcomes more accurately.

Males were more likely to die after 2016. Being a skilled worker, such as an engineer, or being diagnosed with a mental health problem, such as depression or anxiety, also led to early death, according to the researchers. At the same time, holding a management position or having a high income often pushes people toward the “survival” column.

However, the research had several limitations. “We noted that the trials were not randomized and that the researchers did not intentionally make allocations during the trials and evaluate the results,” the report stated.

The researchers only examined data from an eight-year period, and there may be sociodemographic biases in the sampling, even though everyone in Denmark is included in the national register.

“If someone does not have a salary, or chooses not to participate in health systems, we cannot access their data,” they noted.

The researchers also noted that the study was conducted in a wealthy country with strong infrastructure and a strong healthcare system. It is unclear whether life2vec's findings can be applied to other countries such as the United States, given economic and social differences.

Lehman says he knows the algorithm sounds “ominous and far-fetched, but it's actually something that a lot of work has been done on, mostly driven by insurance companies.”

Dr. Arthur Kaplan, director of the department of medical ethics at New York University's Grossman School of Medicine, agrees that insurers will be eager to get ahead of consumers as models like life2vec become more commercial.

“This will make it harder to sell insurance,” he says. “You can't sell risk insurance if everyone knows exactly what it is.”

However, Kaplan, who was not involved in the new research, points out that life2vec does not predict what age a person will die or how. For example, an algorithm cannot predict whether a person will die in a car accident.

Kaplan expects more advanced forecasting models to emerge in just five years.

“We will have better models, with larger databases, that will suggest what to do to prolong life,” he says.

Ultimately, according to Kaplan, using AI to predict when we might die eliminates the one aspect of our lives that keeps them interesting: mystery.

“We're worried that robots will take over the world and decide they don't need us,” he says. “What we should be concerned about is that robots manipulate information and are able to predict so many things about our behavior that we end up having a life that is so predictable that it takes away some of the value of life.”