MIAMI.- Artificial intelligence (AI) has made great strides in predicting the future. Now a group of scientists from the Technical University of Denmark (DTU) has taken a big step in this context by developing an AI model capable of predicting whether a person will die in the next four years.
Life2vec can predict with 78% accuracy whether a human being is going to die in this period of time, 11% better than the best current models, also based on neural networks.
The researchers analyzed an extensive database on education, employment, medical reports and personality traits covering more than six million people, managed between 2008 and 2016 by the Danish government’s Statistics Denmark company.
They detailed that the results were verified until 2020, however the machine only obtained data from 2008 to 2016, which revealed four years later the high effectiveness of the model. Factors such as each person’s health, job or education were taken into account.
This study focuses on a population of people between 35 and 65 years old, which is statistically more difficult to predict when they will die.. Data that they divided into pairs because the number of deaths is too small and the machine could always say no and have a very high success rate, above 90%,” they specified.
Professor Adolfo Sanz de Diego highlights that this artificial intelligence model reflects how technology constantly improves itself, even reaching the point of “predicting death.”
What technology did they use?
The creators explained that they used a system of “deep learning”which is based on a filter adapted to the life circumstances of each individual, allowing extreme personalization in predictions.
Instead of using the data in numerical terms, they translated every life circumstance of citizens into words, so that this deep learning system, based on a large language model (LLM), could do predictions about these situations like those made by the famous ChatGPT.
“The key is that we think of human lives as sequences of things that happen, turning them into sentences, where one word follows another and we use language models to make predictions,” said Sune Lehmann, lead author of the study.
In the article, published in the magazine Nature Computational Science They highlight that this artificial intelligence model reflects how technology constantly improves itself, “even reaching the point of predicting death.”
“Our framework allows researchers to uncover potential mechanisms that impact life outcomes, as well as associated possibilities for personalized interventions,” they add.
Could be used for health care
On the other hand, the team of experts assures that this mechanism can be used in medicine, health or preventive treatments, to detect common risks and intervene earlier, or by governments to reduce inequality and emphasizes that companies could also use it in a way harmful.
They emphasized that these types of advances open a debate about the limits that should be placed on artificial intelligence, since while thousands of citizens defend the implementation of this model in people’s daily lives, others see it as a threat.
They believe that this machine learning approach can be useful in accurately predicting different aspects of human life and providing a quantitative understanding of human behavior.
They concluded by mentioning that the next step would be to incorporate other types of information, such as text and images or information about social connections.
@Lydr05
Source: With information from the journal Nature Computational Science
