Researchers in medicine have unlocked another potential feature of artificial intelligence (AI): predicting the time of human death.

An AI system has recently been trained by scientists to collect information on the general health of more than half a million people in the UK, over the past decade.

They then asked the AI ​​to predict if an individual was in danger of dying early - in other words, dying earlier than the average human age - was caused by chronic diseases.

The results of predicting AI death were judged to be "significantly more accurate" than diagnostic systems that do not use machine learning - that collect information at a large scale.

This assessment of AI capabilities is given by Professor Stephen Weng, currently studying epidemiology and data science at the University of Nottingham, UK.

How does AI foresee death?

To evaluate the probability of death of an object, the researchers tested on two types of artificial intelligence. The first is "deep learning", in which networks process information in a layered fashion that helps the system learn information from real-world examples. In addition, the simpler "random forest" AI is also used to combine multiple models, according to a tree-like mind map, to consider possible outcomes.

Finally, the scientists compared the AI ​​conclusions with those from a manual algorithm, known as the Cox model.

The basic factors to evaluate include: age, sex, smoking history, cancer screening results. While the traditional Cox model focuses on race and physical activity, the AI ​​engine is concerned with body fat percentage, the amount of green vegetables consumed, as well as environmental hazards such as umbrellas. air pollution, alcohol use and drug abuse.

Surprising accuracy

Using these systems, scientists evaluated data in the British Biological Bank (UK BioBank) - a database of genetics, physics and health - made up of more than 500,000 medical records. economy from 2006 to 2016. During those 10 years, nearly 14,500 participants died mainly from cancer, heart disease and respiratory disease.

As a result, the "deep learning" engine produced the most accurate predictions, correctly identified 76% of the subjects who died during the study. While "random forest" gives equally impressive results with 64% accuracy. The traditional Cox model only holds 44% of the results.

The direction for AI in medicine

This is not the first time that experts have harnessed the power of AI to take care of human health. In 2017, AI was used to diagnose early signs of Alzheimer's with 84% accuracy.

Besides, the human ability to have autism, diabetes, heart attack and stroke can also be predicted by AI.

While it may sound odd, using this AI death determination process will "be able to help verify scientific methods and develop the future of this interesting field," said Professor Joe Kai. , one person working for the United Nations.