Danish researchers have developed an artificial intelligence model that crunches sequences of key life events, such as someone’s medical history, education, job, income and marriage status, to predict anything from that person’s personality to their expected lifespan.

The Life2Vec tool was developed by researchers at Danmarks Tekniske Universitet, and is based on a transformer model similar to the one that was used by OpenAI to train ChatGPT. It was trained on a dataset that covers the entire population of Denmark, and the researchers claim it can predict someone’s time of death with an accuracy of 78%.

The researchers say the system is a significant leap in the realm of predictive analytics and could have transformative applications within the healthcare industry.

According to DTU Professor Sune Lehmann Jørgensen, Life2Vec aims to enhance our understanding of key events in human lives by analyzing a broad range of data. Its potential applications in medicine are very promising, especially in terms of its ability to detect certain kinds of disease to enable earlier and potentially life-saving intervention.

Life2Vec, which has been tagged as an “AI death calculator”, works using similar techniques to those used in LLMs such as ChatGPT and Google LLC’s Gemini. But rather than processing text-based data, it analyzes life events.

DTU’s researchers said they originally focused on mortality predictions due to the abundance of data in this area that comes from insurance firms. They looked at data on thousands of people who had taken out health insurance between 2008 and 2016, and tasked the algorithm with predicting how many of those people would still be alive in 2020. It demonstrated an impressive 78% accuracy in that regard.

The algorithm was fed with detailed information on people’s life events, with digital values assigned to each one. For instance, it might receive data such as “Francisco earned 20,000 kroner per month in 2012 as a guard at Kronborg Castle in Helsingor” or “Hermione had five different A-level subjects in high school.”

The model is said to work by transforming these life events into vector representations in embedding spaces. Armed with multiple embeddings, the model can then categorize and establish connections between these life events to form the basis of its predictions, taking into account the sequence and minutiae of each one.

Jørgensen said the initial results displayed by Life2Vec are just the beginning of a much larger effort to use AI to predict various aspects of human society.

Image: Freepik