A new AI system, akin to ChatGPT but developed using the personal stories of over a million people, has demonstrated remarkable precision in forecasting individual life paths and the likelihood of early death, as revealed by a recent study.
Denmark's population data fuels AI insights
This AI was programmed using the personal records of Denmark's entire population. Researchers from the Technical University of Denmark (DTU) report that it surpasses any current system in predicting death risks.
The team examined health and job market data of 6 million Danish citizens from 2008 to 2020, which included details like educational background, medical visits, diagnoses, income, and job type.
life2vec: Training AI on life stories
They transformed this information into text to train a sophisticated language model named "life2vec", utilizing technology similar to that in AI applications like ChatGPT.
After learning the data patterns, the AI was able to outdo other high-tech systems in predicting aspects like personality traits and death timing with notable accuracy, as stated in the Nature Computational Science journal this Tuesday.
The AI was tested on a subset of individuals aged 35-65, half of whom passed away between 2016 and 2020. It successfully predicted who would live or die with 11% greater accuracy than any other AI or method used by life insurance companies.
Sune Lehman from DTU, the study's lead author, described the novelty of viewing human life as a sequence of events, akin to words forming sentences. He explained that they applied transformer models, typically used in AI, to analyze these 'life sequences'.
The researchers used the model to estimate the likelihood of a person dying within the next four years.
Their findings aligned with existing research, indicating that those in leadership roles or with higher incomes have better survival chances. Conversely, being male, skilled, or having a mental health diagnosis increases death risk.
Sune Lehman's novel approach to human life analysis
Dr. Lehman said, “Our primary interest isn't the prediction itself but understanding the data elements that enable such precise forecasts.”
The model also more accurately predicted personality test results in a segment of the population than existing AI systems.
Researchers believe their framework could help uncover new mechanisms affecting life outcomes and tailor interventions.
However, they caution against using this model for life insurance purposes due to ethical implications.
Dr. Lehman emphasized that the model's use by insurance companies would undermine the principle of shared risk, which is fundamental to insurance.
The team also highlighted ethical concerns regarding the use of life2vec, such as data protection, privacy, and bias.
They concluded that while their research showcases the potential of such models, their real-world application should be regulated to safeguard individual rights.
