There seems to be no shortage of applications for artificial intelligence, with scientists using these systems to create new tools in a variety of fields. A team of Danish researchers has now revealed their work developing a tool capable of predicting an individual’s potential health problems and even a probable date of death, barring unforeseen accidents.
Called “life2vec”, this new AI tool is capable of predicting events in people’s lives, including the date of their death. The eye-opening project is being carried out at Technical University of Denmark, the University of Copenhagen, the IT University of Copenhagen, and Northeastern University in the United States.
To create this tool, the researchers developed a language model dedicated to analysing data related to events in six million Danes’ lives. To do this, they compiled large amounts of data related to their health – ranging from visits to the doctor or emergency room to various diagnoses – and their working conditions, including sectors of activity, income and social benefits. Most of this data concerned the period of 2008-2016.
After training their model, the scientists were able to create a tool that could predict an individual’s life expectancy, as well as create assessments about their risk of contracting a disease. This model currently only applies to the Danish population, as it was created and trained exclusively on data from residents of this country.
Unsurprisingly, individuals in leadership positions or with high incomes are the most likely to live a long life, while having a mental-illness diagnosis was associated with a greater risk of death within a given period, such as four years.
Beyond assessments regarding just how reliable such a model is, this research clearly raises important ethical questions concerning the use of sensitive medical data and its confidentiality, as the researchers themselves point out.
Nevertheless, this type of investigation opens up new and unexpected perspectives, noted the experts, whose research was published in the journal Nature Computational Science.