This is getting frightening.
Danish researchers have harnessed potent machine-learning algorithms to forecast various aspects of human lives, including estimating how early someone might pass away. In a recent publication in the journal Nature Computational Science, Sune Lehmann, lead author and professor at the Technical University of Denmark, unveiled their life2vec model’s capabilities, detailing its predictions based on detailed individual data. While life2vec exhibited remarkable predictive potential, the researchers underscored its current status as a “research prototype,” stating its limitations in performing real-world tasks.
The study leveraged a comprehensive dataset encompassing 6 million individuals from a Danish national register, encompassing diverse facets such as education, health, income, and occupation between 2008 and 2016. Employing language processing techniques, the researchers trained life2vec to interpret life events described in detailed sentences, enabling it to generate predictions encompassing individuals’ thoughts, emotions, behavior, and potential mortality.
To gauge mortality forecasts, the team focused on a cohort of over 2.3 million individuals aged 35 to 65 between 2008 and 2015. Assessing life2vec’s accuracy, they found it correctly predicted survival past 2016 in 78% of cases, outperforming other contemporary models by at least 11%. Notably, male gender, certain occupations like engineering, and diagnoses of mental health issues correlated with earlier mortality, while managerial positions or higher incomes were linked to increased survival rates.
Despite the promising outcomes, the study faced limitations. Its scope spanned eight years, potentially introducing sociodemographic biases despite its comprehensive coverage of Denmark’s population. Moreover, the study’s reliance on national registry data restricted insights into individuals without salary records or limited healthcare engagement.
Furthermore, the research was conducted within a well-developed, affluent country, raising doubts about the applicability of life2vec’s findings in economically and socially diverse settings like the United States.
While acknowledging the algorithm’s potential implications, including future insurance applications, Dr. Arthur Caplan from New York University’s Grossman School of Medicine highlighted its limitations. He cautioned against overestimating life2vec’s capabilities, emphasizing its inability to predict precise causes or ages of death and anticipating more sophisticated models emerging in the near future.
Caplan expressed concerns about the impact of such predictions, fearing a loss of life’s unpredictability and uniqueness due to increasing data predictability. He emphasized the potential societal consequences of highly predictable lives, cautioning against technological advancement removing the enigma that makes life interesting.
Discussion about this post