This spooky AI model will tell you the time of your death

“Scientifically, what is exciting for us is not so much the prediction itself, but the aspects of data that enable the model to provide such precise answers," he added.

They used a dataset called life2vec from Denmark, which had data from 6 million Danes, including information about various aspects of their lives, such as health, education, occupation, income, address, and working hours, recorded daily.

Life2vec predicts general questions like "willWilleone die within four years?" 

The researchers created a way to organize these life events into a single "space" or system, showing that it is stable and has a clear structure.

When researchers examined the system's answers, they found that the results match what social science has already discovered. 

For example, all else being equal, people in leadership positions or with high incomes are more likely to survive. On the other hand, being male, skilled in a particular area, or having a mental diagnosis is associated with a higher risk of dying.

Their models can then predict various outcomes, from early mortality to personality traits, and the researchers claim that their models perform better than the best existing models by a significant margin. 

"What's exciting is to consider human life as a long sequence of events, similar to how a sentence in a language consists of a series of words,” added Lehmann.

“This is usually the type of task for which transformer models in AI are used, but in our experiments we use them to analyze what we call life sequences, i.e., events that have happened in human life,” he added.

It’s spooky to know of the time of your death or the circumstances around it. It’s not everyone’s cup of tea. So, naturally, the researchers have raised some ethical and privacy concerns around this study.

The researchers explained in the press release that they think the next step would be to incorporate other types of information like text and images or information about our social connections. 

The study was published in Nature Computational Science.

Study abstract:

Here we represent human lives in a way that shares structural similarity to language, and we exploit this similarity to adapt natural language processing techniques to examine the evolution and predictability of human lives based on detailed event sequences. We do this by drawing on a comprehensive registry dataset, which is available for Denmark across several years, and that includes information about life-events related to health, education, occupation, income, address and working hours, recorded with day-to-day resolution. We create embeddings of life-events in a single vector space, showing that this embedding space is robust and highly structured. Our models allow us to predict diverse outcomes ranging from early mortality to personality nuances, outperforming state-of-the-art models by a wide margin. Using methods for interpreting deep learning models, we probe the algorithm to understand the factors that enable our predictions. Our framework allows researchers to discover potential mechanisms that impact life outcomes as well as the associated possibilities for personalized interventions.