This is a matter of life and death — no doubt a-bot-it.
Although most folks aren’t in a massive rush to learn when they’ll bite the big one, a newly developed AI death calculator can now forecast when a person will die with eerily exact accuracy.
“We use the technology behind ChatGPT (something called transformer models) to analyze human lives by representing each person as the sequence of events that happens in their life,” Sune Lehmann, lead author of the December 2023 study “Using sequence of life-events to predict human lives,” told The Post.
In the report, the professor of network and complex systems from the Technical University of Denmark, and co-authors introduce an algorithm known as “life2vec,” which uses select details of an individual’s life — including income, profession, residence and health history — to determine life expectancy with 78% correctness.
“We use the fact that in a certain sense, human lives share a similarity with language,” explained Lehmann. “Just like words follow each other in sentences, events follow each other in human lives,”
Slightly different than ChatGPT — the ever-buzzy bot that tech wizards have employed to help land their dream jobs or even curate the perfect outfit — life2vec can compute the life outcomes of a man or woman by closely examining their pasts.
“This model can predict almost anything,” Lehmann told The Post, who noted that his research team also used the specialized program to foretell people’s personalities and decisions to make international moves.
“We predicted death because it’s something people have worked on for many years (for example, insurance companies),” he added, “so we had a good sense of what was possible.”
Lehmann’s troop examined a heterogeneous subject population of 6 million Danish people, who varied in sex and age, between 2008 and 2020. The analysts used life2vec to discover which of the subjects would likely live for at least four years beyond Jan. 1, 2016.
“The scale of our dataset allows us to construct sequence-level representations of individual human life trajectories, which detail how each person moves through time,” reads the report. “We can observe how individual lives evolve in a space of diverse event types (information about a heart attack is mixed with salary increases or information about moving from an urban to a rural area).”
Researchers fed the AI-specific information on each study participant, using plain language such as: “In September 2012, Francisco received 20,000 Danish kroner as a guard at a castle in Elsinore” or “During her third year at secondary boarding school, Hermione followed five elective classes.”
They then assigned different digital tokens to each piece of data, which were all quite specifically categorized. For instance, a forearm fracture is represented as S52; working in a tobacco shop is coded as IND4726, income is represented by 100 different digital tokens; and “postpartum hemorrhage” is O72.
Using the information provided, life2vec almost perfectly predicted who had died by 2020 more than three-quarters of the time.
Per the study, some of the factors that can contribute to earlier death include being male, having a mental health diagnosis or being in a skilled profession. Earning a higher income or being in a leadership role were both linked to a longer life.
However, Lehmann emphasized to The Post that no study participants were given their death predictions.
“That would be very irresponsible,” he said, noting that he and his team hope to eventually share more details of their results in a way that protects the privacy of those involved in the research.
“But we can still learn from [life2vec] what the factors are that might help you live longer,” said Lehmann. “We haven’t gone deep with this, but that’s another important application of the model.”
The bot is not currently available to the general public or corporations. And upon its mass rollout — should it ever become open for mainstream use — the prober says the AI likely won’t be used to apprise specific individuals in instances like writing insurance policies or making hiring decisions.
‘”The predictions are not used for anything,” Lehmann insisted. “The point of the life2vec is to understand what’s possible — and not possible — to predict.”
Source
