- A new study says it’s possible to predict someone’s time of death accurately.
- The study used machine learning to create detailed life sequences for 6 million people in Denmark.
- Beyond morbid curiosity, the model could impact the work of insurance companies.
Death and taxes may be the only certainties in life. And now advances in AI may have brought us one step closer to anticipating the former.
A new study found that it’s possible to make accurate predictions about a person’s death using a subset of AI known as machine learning.
The study’s authors said the “sheer scale” of their dataset was key to their research. They collected a decade’s worth of day-to-day records from more than 6 million people living in Denmark, according to the report.
The dataset includes health records, salary, working hours, residences, and more. Using that data, the researchers created a deep-learning model called “life2vec” to map out detailed sequences of an individual’s life events, according to the study.
To test life2vec, the researchers ran a subset of the data to see if it could predict if someone had survived in the four years after 2016. The researchers knew the answer, but the algorithm did not.
“To test how good [life2vec] is, we pick a group of 100,000 individuals where half survive and half die,” the study’s lead author, Sune Lehmann, a professor at the Technical University of Denmark, told CNN. They focused on using individuals between the ages of 30 and 55 — when mortality is harder to predict.
Life2vec was right about 78% of the time, according to CNN. It found that individuals with high incomes or those who held managerial roles were more likely to survive. It also found that men, skilled workers, or individuals who had been diagnosed with a mental health disorder were more likely to have died.
The model outperformed existing “state-of-the-art methods” by about 11%, the authors said. Not surprisingly, life2vec performs better for groups with a greater number of health events in their records. The researchers found it also performed better with cohorts of younger people and women.
The researchers also used the model to make predictions about more subjective human details, like personality traits, which are typically gathered through questionnaires. Life2vec was able to make predictions about a host of human personality traits from a person’s self-esteem to their sociability.
While life2vec may sound “ominous and crazy,” it’s likely to have an impact on work that’s already going on in the real world, especially work driven by insurance companies, Lehmann told CNN.
Google has also been working on AI technology that can predict a person’s death by analyzing health records.
Life2vec is not ready to perform any “real-world tasks” in its current state but, theoretically, “We can make any kind of prediction,” Lehmann told CNN.
