Fortune-Telling AI? Researchers Present Algorithm That Can Predict Your Future

Beyond mortality predictions, the study demonstrates the model's ability to accurately predict other aspects of individuals' lives, such as personality traits.

While the results are intriguing, scientists caution that the patterns observed in the Danish population may not necessarily apply universally.

Source : Getty

In a groundbreaking study published in Nature Computational Science, researchers reveal that artificial intelligence (AI) may present a unique perspective on predicting life outcomes. The study explores the application of a fortune-telling algorithm, fueled by data from millions of individuals' life experiences, to forecast various aspects of a person's future, including lifetime earnings and the likelihood of early mortality. This innovative approach intersects machine learning with the social sciences, marking a notable trend in contemporary research.

Matthew Salganik, a sociologist at Princeton University, who was not part of the study, suggests that if the method proves effective across diverse societies, it could provide social scientists with a valuable tool to investigate the influence of traits and events on an individual's destiny. He emphasises that the findings generate more questions than answers, presenting an opportunity for further exploration.

The study diverges from previous attempts, where machine learning models failed to accurately predict life outcomes based on health, family relationships, and education data from a cohort of 5,000 children over 15 years. In this new research, scientists employed large language models, akin to those powering ChatGPT, which analyse extensive textual data to discern patterns in language. These models, such as the one developed in the study known as "life2vec," examine the sequence of life events, acknowledging the significance of their order.

To train the model, researchers utilised data from Danish national registers, containing work and health records for approximately 6 million citizens. These details, including salary, job title, and health information, were translated into a synthetic language, creating sentences that represented individual life events. By arranging these events chronologically, the model reconstructed each person's digital life story.

The "life2vec" model, trained on life stories from 2008 to 2016, successfully predicted whether individuals in the Danish national registers had died by 2020 with an accuracy rate of 78 per cent. The model identified factors associated with a higher risk of premature death, such as low income, mental health diagnoses, and male gender. However, researchers acknowledge challenges in predicting accidents and heart attacks.

While the results are intriguing, scientists caution that the patterns observed in the Danish population may not necessarily apply universally. Youyou Wu, a psychologist at University College London, suggests adapting the model with cohort data from different countries to explore potential universal patterns or cultural nuances.

Concerns about biases in the data are also raised, particularly in the context of potential implications for insurance premiums or hiring decisions. For instance, biases in the data could lead to algorithmic errors, such as mistakenly labelling certain populations at a higher risk of premature death.

Beyond mortality predictions, the study demonstrates the model's ability to accurately predict other aspects of individuals' lives, such as personality traits. Sandra Matz, a computational social scientist at Columbia Business School, acknowledges the predictability of certain behaviours but expresses scepticism about the model's capacity to forecast all types of behaviour.

Sune Lehmann, a network and complexity scientist at the Technical University of Denmark and lead author of the study, envisions future applications for the model, such as identifying disease risks to help individuals proactively manage their health. However, he emphasises the need to address data privacy concerns before implementing such applications. As discussions around the potential uses of the model unfold, questions about ethical considerations and privacy safeguards remain at the forefront.