Can AI predict human life expectancy?

Artificial intelligence (AI) systems based on large language models can predict events in a person's life and even estimate their likely life expectancy. This conclusion was reached by an international group of scientists from the USA, Denmark and Switzerland in a study, the results of which were published in the scientific journal Nature Computational Science (NCS).

The scientists used a model called life2vec and fed it with health and employment data from 6 million Danes. After training the neural network at the initial stage, it outperformed other advanced neural networks and began to predict the time of death with high accuracy and even talk about personality traits.

Lead author of the study, professor at the Technical University of Denmark Sune Lehmann, noted that he and his colleagues were interested in the question of how accurate AI forecasts based on historical data could be. What's important to scientists, he says, is not just the predictions themselves, but also the aspects of the data that allow the model to provide accurate answers.

The life2vec model encodes data into a large system of vectors, organizing various information. It takes into account such parameters as time of birth, study, education, salary, housing and health.

The model's response is consistent with existing findings and social science data, for example showing that people with high income or leadership positions are more likely to live longer.

The authors of the article emphasize that there are a number of ethical issues associated with the life2vec model, such as the protection of personal data, confidentiality and bias in evaluation. These issues need to be better understood before the model can be used to assess disease risk or prevent certain life events, the scientists say.