Artificial intelligence taught to predict human death: it is right in 78% of cases

Scientists have created an artificial intelligence model that proved to be quite capable of predicting how much a person will earn in the future, what will happen to their health, and whether they will die prematurely. To do this, AI needs to access a database that contains information about a particular person's past salary, medical history, etc.

The results of a large-scale study were published in Nature Computational Science. So far, relatively local work has been done, using data from almost 6 million Danish residents.

This was not the first attempt to combine sociological data and artificial intelligence to find certain sequences that could predict the future.

Earlier, Princeton University sociologist Matthew Salganik and more than 100 other teams tried to make AI predict the future health, family relationships, and education of about 5,000 children over 15 years. However, none of the models developed gave accurate predictions.

The authors of the new study used a large language model (LLM) similar to the one used in ChatGPT. The algorithm of such LLMs is able to analyze huge amounts of data, finding certain patterns in it. In particular, if you "feed" such an algorithm with large volumes of various texts, it will start guessing what the next word in a sentence will be even before it is written.

Therefore, Sune Lehmann, a researcher at the Technical University of Denmark, and his colleagues decided to test whether such a language model would be able to find meaning in other sequences. In particular, whether it could predict the future if it knew the life story of a particular person.

For their experiment, the researchers received permission to use the Danish national registers from 2008 to 2016. They contain data on the work and health status of each of the country's approximately 6 million citizens.

So the algorithm received initial data such as salary, social benefits, job title, hospital visits, and diagnoses and started working on them.

The AI created a timeline on which all events were placed according to when they occurred.

The researchers named the model life2vec. After it finished processing the data, it was given a new task: to determine whether any of the people whose data was entered into the Danish national registers might die before 2020.

The model's predictions were accurate in 78% of cases. It managed to identify several factors that contributed to the increased risk of premature death. These include low income, mental health diagnosis, and male gender.

The weakness of the model's predictions was accidents or heart attacks, which are difficult to predict.

The results of this study are quite intriguing, but scientists are not yet sure whether the same algorithm will be able to produce as accurate data if it works with information on people from another country.

Nevertheless, Lehmann believes that the technology could be useful, as it could warn people in advance that they need to take more care of their health to avoid the risk of premature death.

Earlier, OBOZ.UA reported that scientists managed to create an AI capable of reading people's thoughts.

Subscribe to OBOZ.UA channels on Telegram and Viber to keep up with the latest developments.