
An algorithm trained on the data of millions of people can read the future with an accuracy that fortune tellers would envy.
The AI system uses the ability to recognize recurring patterns in large amounts of data to predict the evolution of a person's life, from salary increases, illnesses and the possibility of early death. The data of the algorithm was described in the scientific journal "Nature Computational Science".
Sune Lehmann Jørgensen, a network scientist at the Technical University of Denmark, started from data held in a Danish database, the Danish National Registers, which includes work and health information on about 6 million citizens, and converted them into strings. of text, which then fed Life2vec, a large language model (Large Language Model, same category as chatbot ChatGPT).
These systems are able to identify, within a large amount of text, repeated patterns in the succession of words and sentences, and use them to process coherent texts. Therefore, scientists wondered if the same skills could be used to trace a possible trajectory in other sequences, those of important events in our lives.
The first step was to translate details such as salary, job title or medical diagnoses into concise text strings, such as: "In August 2010, Agnes earned 30,000 Danish kroner as a midwife at a hospital in Copenhagen."
The model was trained on the lives of every individual who entered the database between 2008 and 2016, and this information was used to ask the algorithm to "predict" how many of the people studied would die by 2020 (" future", compared to the era in which the AI was trained). The AI's predictions turned out to be correct 78% of the time. The system was able to identify various factors that favored a greater risk of premature death, such as low pay, a diagnosis of mental illness, being male, although it "missed" some less predictable causes of death early such as heart attacks or accidents.
The algorithm also correctly predicted other aspects of people's lives, such as the propensity for certain character traits (easily associated with different types of jobs). The model can be used to imagine the future risk of developing certain diseases and act accordingly, but it can also pose major risks to citizens' privacy, security and rights.