AI Can Accurately Predict Death About 80% of the Time, New Study Finds

Key Points:

• Machine learning can accurately predict death nearly 80% of the time
• Researchers studied 6 million people in Denmark to create a deep-learning model able to predict mortality

A new study has found that machine learning can accurately predict death nearly 80% of the time. Researchers in Denmark collected a decade’s worth of day-to-day records from over 6 million individuals and used that data to create a deep-learning model called “life2vec” to map out detailed sequences of an individual’s life events. The researchers tested the model and found that it was right about 78% of the time when predicting survival outcomes for individuals between the ages of 30 and 55. The model also outperformed existing methods by 11%. Additionally, the researchers found that the model could make predictions about subjective human traits, such as personality traits, based on an individual’s records.

The study’s authors believe that the “sheer scale” of their dataset was key to the accuracy of their predictions. The dataset included health records, salary, working hours, residences, and more. The researchers believe that these predictions could have an impact on work being done by insurance companies and other organizations. However, they also note that the model is not yet ready for “real-world tasks” in its current state.

While the use of AI to predict death may sound ominous, the researchers believe that their model has the potential to predict a wide range of outcomes if given appropriate data. Google has also been working on similar AI technology to predict a person’s death based on health records.