If you could flip ahead a few pages in the story of your life, would you take a peek? Artificial intelligence (AI) may now offer a version of this choice, say researchers behind a study out today in Nature Computational Science. By crunching data from millions of people’s lives, the authors note, a fortune-telling algorithm can predict a person’s life outcomes with eerie accuracy, such as their lifetime earnings or their likelihood of facing an early death. The finding adds to a recent trend blending machine learning with the social sciences.
If the approach can be shown to work across different societies, it could give social scientists a new tool to explore how traits and events affect a person’s fate, says Matthew Salganik, a sociologist at Princeton University who was not involved in the work. “I think it raises more questions than it answers. And I mean that in a positive way.”
Previously, Salganik and collaborators—along with more than 100 other teams—attempted to develop machine learning models to predict life outcomes using data on health, family relationships, and education from about 5000 children over 15 years. Yet none of their models yielded accurate predictions.
Sune Lehmann, a network and complexity scientist at the Technical University of Denmark, and colleagues wondered whether these models could also find meaning in other sequences—such as those that make up our life stories. “Just like language, the order in which life events happen is really important,” he says. Receiving a cancer diagnosis just after landing a job with good health benefits is likely to have a different impact from if those events were reversed, for instance.
For data to plug into the algorithm, the researchers turned to the Danish national registers, which contain work and health records for each of the country’s approximately 6 million citizens. The team translated details such as salary, social benefits, job title, and hospital visits and diagnoses into a synthetic language in which single life events became sentences. For example, “In August 2010, Agnes earned 30,000 Danish kroner as a midwife at a hospital in Copenhagen.” By placing these events on a timeline, the model recreated each person’s digital life story.
The researchers trained the model, called “life2vec,” on every individual’s life story between 2008 to 2016, and the model sought patterns in these stories. Next, they used the algorithm to predict whether someone on the Danish national registers had died by 2020.
The model’s predictions were accurate 78% of the time. It identified several factors that favored a greater risk of premature death, including having a low income, having a mental health diagnosis, and being male. The model’s misses were typically caused by accidents or heart attacks, which are difficult to predict.
Although the results are intriguing—if a bit grim—some scientists caution that the patterns might not hold true for non-Danish populations. “It would be fascinating to see the model adapted using cohort data from other countries, potentially unveiling universal patterns, or highlighting unique cultural nuances,” says Youyou Wu, a psychologist at University College London.
Biases in the data could also confound its predictions, she adds. (The overdiagnosis of schizophrenia among Black people could cause algorithms to mistakenly label them at a higher risk of premature death, for example.) That could have ramifications for things such as insurance premiums or hiring decisions, Wu adds.
Lehmann and colleagues also found that their model accurately predicted other aspects of people’s lives, such as whether they were more or less likely to be extroverted. That’s not so surprising, says Sandra Matz, a computational social scientist at Columbia Business School. Even simpler algorithms could correlate certain vocations—a hairstylist, say—with extraversion. “Whether it could predict all kinds of behavior, I’d be somewhat skeptical.”
This algorithm could predict your health, income, and chance of premature death | Science | AAAS
Lehmann says he imagines the model might one day be useful for identifying a person’s disease risk, which could help them take steps to stay healthy. Such applications come with a host of questions about data privacy, he notes, which will need to be sorted before his model can help anyone. “The best way I can think of to start this discussion is by forming an image of what’s even possible.”
No hay comentarios:
Publicar un comentario