In a groundbreaking leap into the realm of artificial intelligence, researchers at the Technical University of Denmark (DTU) have unveiled Life2vec, an AI-based death predictor with claims of an impressive 78 percent accuracy rate. Modeled after the well-known ChatGPT, this system takes a departure from crafting creative text or navigating professional challenges. Instead, it delves deep into the personal histories of individuals, aiming to predict not just career success or fashion choices, but the very outcome of a person’s life.
The inspiration behind Life2vec stems from the notion that human lives, in a peculiar sense, share a similarity with language. Sune Lehmann, the lead author of the December 2023 study titled “Using sequence of life events to predict human lives,” explained, “Just like words follow each other in sentences, events follow each other in human lives.” Drawing on the transformer models that power ChatGPT, Life2vec represents each person as a sequence of life events, utilizing personal information such as health, education, occupation, and income for its predictions.
The study, published in the prestigious journal Nature Computational Science, sheds light on the extensive analysis conducted by the researchers. Data collected from 6 million Danes spanning the years 2008 to 2020 was meticulously scrutinized. This wealth of information included details ranging from education levels and doctor’s appointments to hospital visits, resulting diagnoses, income, and occupation. The model, specifically tailored for individuals between the ages of 35 and 65, exhibited an impressive 11 percent improvement in accuracy compared to existing systems or methods employed by life insurance companies.
DTU professor Sune Lehmann emphasized the model’s ability to address a fundamental question: to what extent can AI predict events in your future based on conditions and events in your past? He stated, “Scientifically, what is exciting for us is not so much the prediction itself, but the aspects of data that enable the model to provide such precise answers.”
The study also revealed that Life2vec could predict general questions related to the outcomes of a personality test and even the chances of a person dying within the next four years. However, the researchers issued a cautionary note, stressing that the system should not be used by insurance companies due to ethical concerns.
Lehmann explained, “Clearly, our model should not be used by an insurance company because the whole idea of insurance is that, by sharing the lack of knowledge of who is going to be the unlucky person struck by some incident, or death, or losing your backpack, we can kind of share this burden.”
As the scientific community grapples with the ethical implications of such a predictive system, the study opens up intriguing possibilities for the future. Life2vec prompts us to consider the unfolding narrative of our lives as a series of events, akin to the structure of a sentence in language. The utilization of transformer models in this context, while unconventional, showcases the versatility and potential of AI technologies.
In essence, Life2vec offers a glimpse into a future where the progression of individual health sequences and labor history could be harnessed to forecast life outcomes. As we navigate the intersection of technology and personal privacy, the development of such predictive models challenges us to ponder the depths to which AI can unravel the mysteries of our existence.