A newly introduced AI death calculator is purportedly capable of predicting our time of death with an unprecedented level of accuracy. The algorithm at the heart of this innovation, known as Life2vec, relies on just four key pieces of data to generate predictions, according to findings from a recent study. The Life2vec algorithm boasts an impressive 75% accuracy rate, positioning it alongside comparable models designed to forecast various life outcomes.
“We use the technology behind ChatGPT (something called transformer models) to analyse human lives by representing each person as the sequence of events that happens in their life,” Sune Lehmann, lead author of the December 2023 study ‘Using sequence of life-events to predict human lives', told the NY Post.
Developed by researchers in Denmark and the United States, the AI death calculator was trained using a dataset comprising six million Danish citizens' information from 2008 to 2020. This AI incorporates various details such as income, occupation, location, injuries, and pregnancy history for its predictions. The research team conducted tests on a sample of people aged 35 to 65, half of whom had died between 2016 and 2020.
Scientists provided the AI with specific information about each person in the study, using simple language like "In September 2012, Francisco got 20,000 Danish kroner working as a guard" or "Hermione took five elective classes in her third year of secondary school."
They gave different codes (like S52 for a forearm fracture or IND4726 for working in a tobacco shop) to each piece of information. Using this data, Life2vec accurately predicted who had died by 2020 more than 75% of the time. The study found that factors like being male, having a mental health diagnosis, or having a skilled job could contribute to earlier death. On the other hand, earning more money or having a leadership role was linked to a longer life.
The research revealed that Life2vec's forecasts surpassed the accuracy of other AI models by 11%. “We use the fact that in a certain sense, human lives share a similarity with language,” explained Ms Lehmann. “Just like words follow each other in sentences, events follow each other in human lives.”
“This model can predict almost anything,” she said, adding, "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. This is usually the type of task for which transformer models in AI are used, but in our experiments, we use them to analyse what we call life sequences, i.e., events that have happened in human life."
She continued, “Clearly, our model should not be used by an insurance company, because the whole idea of insurance is that, by sharing the lack of knowledge of who is going to be the unlucky person struck by some incident, or death, or losing your backpack, we can kind of share this burden.”
The AI, Life2vec, is currently not accessible to the general public or corporations. Even if it were to be widely available in the future, Lehmann emphasises that the AI tool probably won't be used for purposes like creating insurance policies or making hiring decisions. The primary goal of Life2vec is to explore the boundaries of what can and cannot be predicted rather than to inform real-world decisions.
