In a groundbreaking study conducted by scientists from the Technical University of Denmark (DTU), a newly developed AI system, akin to ChatGPT has demonstrated an unprecedented ability to predict individual lifespans and the risk of early mortality. The AI model, named “life2vec,” was trained on the personal data of over one million individuals in Denmark, and it outperformed existing systems in predicting outcomes, including personality traits and the time of death, with remarkable accuracy. The findings of this research were published in the journal Nature Computational Science.
Training the AI with personal data
To develop the life2vec AI model, researchers utilized an extensive dataset containing health and labour market information for six million Danes collected between 2008 and 2020. This dataset included details on individuals’ education, medical history, income, and occupation, among other factors. The data was transformed into text-based information to train the AI model, similar to the underlying technology of popular AI applications like ChatGPT.
Unprecedented predictive accuracy
Once trained, the life2vec AI model exhibited exceptional predictive capabilities, surpassing the performance of other advanced systems. Researchers specifically focused on predicting the time of death. They evaluated its accuracy by comparing it to other existing AI models and methods used by life insurance companies to price policies. The results revealed that the life2vec AI model’s predictions were 11% more accurate than any other model, marking a significant advancement in mortality prediction.
Understanding life sequences
Sune Lehman, the study’s first author, explained that the research aimed to view human life as a sequence of events, much like a sentence in a language consisting of words. While this is typically a task for transformer models in AI, the study adapted these models to analyze “life sequences” or events that transpire over a person’s lifetime. By doing so, the model could make precise predictions about future events based on past conditions and experiences.
Factors affecting mortality
The researchers also examined broader questions, such as the likelihood of a person’s death within a specific time frame. Their findings aligned with existing research, revealing that factors like leadership roles, high income, and gender significantly predicted survival rates. For instance, individuals in leadership positions or with higher income were more likely to have a longer lifespan, while being male, skilled, or having a mental health diagnosis increased the risk of premature mortality.
Implications and ethical considerations
While the study demonstrates the immense potential of the life2vec AI model, the researchers emphasize that ethical concerns and privacy issues must be carefully considered. They caution against using the model in life insurance, as the essence of insurance relies on shared risk among individuals. Moreover, safeguarding sensitive data and addressing biases in data are critical ethical considerations when deploying such advanced AI systems.
Enhancing personalized interventions
The researchers believe that the life2vec AI model can provide valuable insights into mechanisms that influence life outcomes, offering opportunities for personalized interventions. By identifying these mechanisms, researchers and policymakers can explore ways to improve individuals’ well-being and quality of life.
The study conducted by scientists from the Technical University of Denmark has unveiled an AI model, life2vec, with the remarkable ability to predict individual lifespans and the risk of early mortality. While the model’s predictive accuracy is notable, ethical concerns and data privacy issues must be carefully addressed when considering its real-world applications. This groundbreaking research opens the door to a new era of personalized interventions and a deeper understanding of the factors that shape our lives.
