A recent study found that a novel artificial intelligence system that resembles ChatGPT and was trained on the life stories of more than a million people is very good at predicting people’s lives and chances of dying early.
According to researchers from the Technical University of Denmark (DTU), the AI model was trained using the personal data of the Danish population and demonstrated a higher degree of accuracy than any current system in predicting the people’s odds of dying.
Researchers examined 6 million Danes’ health and employment data, which was gathered between 2008 and 2020. This data included details about each person’s schooling, hospital and doctor visits, diagnoses that resulted from those visits, income, and work.
In order to train a sizable language model known as “life2vec,” which is comparable to the technology underlying AI apps like ChatGPT, scientists translated the dataset into words.
According to a study published in the journal Nature Computational Science on Tuesday, the AI model could beat other sophisticated systems and predict outcomes like personality and time of death with high accuracy once it learnt the patterns in the data.
Researchers asked the AI system to forecast who would live and who would die based on data on a group of individuals from the set who were between the ages of 35 and 65, half of whom passed away between 2016 and 2020.
They discovered that its forecasts were 11% more accurate than those of any other AI model in use or the technique life insurance firms employ to determine policy prices.
Considerating human life as a protracted series of events is fascinating, much like how a phrase in a language is made up of a number of words, according to research first author Sune Lehman of DTU.
In their investigations, however, they utilize them to assess what we call life sequences, i.e., events that have transpired in human life. Typically, transformer models in AI are employed for tasks like this, according to Dr. Lehman.
Researchers used the model to find general answers, such as the likelihood that an individual will pass away in four years.
They discovered that the model’s predictions match previous research, showing that when all other variables are taken into account, people in leadership roles or with high incomes have a higher chance of surviving, while being male, skilled, or diagnosed with a mental illness is linked to a higher risk of passing away.
They applied the model to the central question: How much can they forecast future events based on circumstances and past events in our lives? Dr. Lehman stated
According to him, the parts of the data that allow the model to produce such accurate results are more intriguing from a scientific standpoint than the forecast itself.
In addition, the model performed better than current AI systems at correctly predicting personality test results for a subset of the population.
According to the paper, their framework enables researchers to discover novel plausible mechanisms that influence life outcomes and related opportunities for personalized interventions.
Scientists warn that because of ethical issues, life insurance companies shouldn’t utilize this model.
It is obvious that an insurance company should not use this approach because the whole point of insurance is that we can sort of share the burden of not knowing who will be unlucky enough to be involved in an accident, die, or lose your backpack, Dr. Lehman told New Scientist.
The usage of life2vec raises additional ethical concerns, according to researchers, including privacy concerns, securing sensitive data, and the influence of bias in data.
They emphasized that while their work explores the boundaries of what is feasible, it should only be applied in the real world in accordance with laws that safeguard people’s rights.
