Artificial Intelligence
Can Artificial Intelligence Predict Your Life, and Death?
New research suggests accuracy in AI's fortune-telling.
Posted December 21, 2023 Reviewed by Lybi Ma
Ever since the explosion in popularity of ChatGPT about a year ago, there has been much discussion around the use of artificial intelligence in academia, health care, and in consumer's everyday lives. The potential benefits—and potential risks, ethical concerns, and complications—are numerous, and the answers are far from easy. Now, new research opens yet another door for the ways that AI has the potential to change everyday life for modern humans: it appears that it can predict certain life outcomes and even potentially estimate the time of death, with significant accuracy.
The study, recently published in Nature Computational Science, involved large data sets involving the health characteristics and behaviors of 6 million Danish people. The data included such demographic details as job traits—like salary and industry—and also health records, including diagnoses and doctor's visits. The data sets went back more than a decade, with the idea that by training the AI model to detect patterns in the participants' past life events, it could make useful predictions about what is to come in the future.
Of course, the idea of using demographic and health characteristics to predict life outcomes, including mortality, is nothing new. Insurance actuaries have been doing this for decades, and there is a robust understanding in mental and physical health research that demographic factors have profound effects on health outcomes. Things like gender, job status, education level, and the presence of various diagnoses affect, statistically, the likelihood of early death. And even the most rudimentary of computer programs can make some meaningful predictions based on these sociological factors.
This latest AI model, however, appears to outperform existing predictive models. With questions like age range at death, it showed increased accuracy compared to anything that's been used before and showed a generally strong ability to use past events as a predictor of future events.
Like much news about artificial intelligence, this of course can be seen as both exciting and disconcerting. There are several important ethical issues, also acknowledged by the researchers, who were led by Germans Savcisens and Lars Kai Hansen at the Technical University of Denmark, and Tina Eliassi-Rad at Northeastern University. Questions come to mind: What if the model develops very human-like biases? How can we be careful to not normalize the invasion of privacy that comes with giving up all of our health information? If some life event—or, more morbidly, the date of our death—can be predicted with increasing accuracy, is that something we would want to know? There is a fine line, for instance, between the general awareness that if we exercise more and quit smoking, we'll add years to our lives, versus the specific knowledge that we are likely to die in a given year in the future.
And if AI is perfected to the point where it really can "see the future" with startling reliability, how will that change the course of how we even think about our lives and our autonomy within them?
Like any interesting research, this newest finding brings both questions and answers.
References
Germans Savcisens, Tina Eliassi-Rad, Lars Kai Hansen, Laust Hvas Mortensen, Lau Lilleholt, Anna Rogers, Ingo Zettler, Sune Lehmann. Using sequences of life-events to predict human lives. Nature Computational Science, 2023; DOI: 10.1038/s43588-023-00573-5