"Danish Life2Vec AI Predicts Lifespan"

Danish researchers have developed an artificial intelligence model that can process sequences of key life events, such as a person's medical history, education, work, income, and marital status, to predict various things from the person's personality to their expected lifespan.

This tool, called Life2Vec, was developed by researchers at the Technical University of Denmark and is based on a transformer model similar to the one used by OpenAI to train ChatGPT. The training dataset covers the entire population of Denmark, and the researchers claim that it can predict a person's time of death with 78% accuracy.

The researchers say that this system is a significant leap in predictive analytics and may have transformative applications in the healthcare industry.

According to Professor Sune Lehmann Jørgensen from DTU, Life2Vec aims to enhance our understanding of key events in human life by analyzing extensive data. It has great potential in medicine, particularly in the early detection and potential life-saving interventions for certain types of diseases.

Life2Vec, dubbed an "AI death calculator," uses similar techniques as those used for LLMs, such as ChatGPT and Google's Gemini. However, it analyzes life events instead of text-based data.

The researchers at DTU initially focused on death prediction because there is a wealth of such data available from insurance companies. They examined data from thousands of individuals who purchased health insurance between 2008 and 2016 and asked the algorithm to predict how many of these people would still be alive in 2020. In this regard, it demonstrated an impressive accuracy of 78%.

The algorithm was fed detailed information about people's life events and assigned numerical values to each event. For example, it may receive data such as "Francisco earned 20,000 kroner per month as a guard at Helsingør's Kronborg Castle in 2012" or "Hermione took five different A-level subjects in high school."

It is said that the model works by transforming these life events into vector representations in an embedding space. With multiple embeddings, the model can then classify and establish connections between these life events to form the basis of its predictions, taking into account the order and details of each event.

Jørgensen states that the preliminary results shown by Life2Vec are just the beginning of a larger effort to use AI to predict various aspects of human society.