A groundbreaking study published in Nature Computational Science reveals that an artificial intelligence (AI) algorithm, akin to fortune-telling, can predict life outcomes, including health, income, and the likelihood of premature death. By analyzing extensive data from millions of individuals, the algorithm, named "life2vec," demonstrated an eerie accuracy in foreseeing various aspects of a person's future. However, ethical concerns surrounding data privacy and potential biases in predictions have surfaced.
The researchers, led by Sune Lehmann from the Technical University of Denmark, employed large language models similar to those powering ChatGPT. These models, known for analyzing vast amounts of text, were trained to recognize patterns in sequences of life events, effectively creating a digital life story for each participant. Using Danish national registers containing work and health records for approximately 6 million citizens, the team translated details such as salary, job titles, and health diagnoses into a synthetic language.
The study trained life2vec on individuals' life stories from 2008 to 2016, seeking patterns that could predict outcomes. The algorithm's predictions were then tested on whether individuals from the Danish national registers had died by 2020, resulting in a 78% accuracy rate. Factors associated with a higher risk of premature death included low income, mental health diagnoses, and male gender.
While the results are intriguing, scientists caution that the patterns observed may not generalize to non-Danish populations. Youyou Wu, a psychologist at University College London, suggests adapting the model with cohort data from other countries to explore universal patterns or cultural nuances.
Concerns about biases in the data and potential ethical implications have been raised. Biases could lead to inaccuracies in predictions, influencing areas such as insurance premiums and hiring decisions. Sandra Matz, a computational social scientist at Columbia Business School, notes that predicting certain aspects of behavior might be more challenging, and skepticism remains regarding the algorithm's ability to forecast a wide range of behaviors.
Lehmann envisions potential applications for the model in identifying disease risks to help individuals take preventive measures. However, the ethical implications and data privacy issues associated with such applications need careful consideration and discussion before implementation. As AI continues to advance in predicting life outcomes, striking a balance between technological innovation and ethical safeguards becomes imperative.
