AI trained on extensive life stories has the ability to forecast the likelihood of early mortality

Artificial intelligence trained on personal data covering Denmark’s entire population can predict people’s likelihood of dying more accurately than existing models used in the insurance industry. Researchers behind the technology say it has the potential to have a positive impact on early prediction of social and health problems, but must be kept out of the hands of large corporations. There is.

Sune Lehmann Jorgensen The researchers used a rich Danish dataset covering the education, doctor and hospital visits, resulting diagnoses, income, and occupation of 6 million people from 2008 to 2020.

They converted this dataset into words that can be used to train large-scale language models, the same technology that powers AI apps like ChatGPT. These models work by looking at a set of words and statistically determining which word is most likely to come next based on a large number of examples. In a similar way, the researcher’s Life2vec model can look at the sequence of life events that form an individual’s history and determine what is most likely to happen next.

In the experiment, Life2vec was trained on all data except for the last four years of data, which was kept for testing. Researchers took data on a group of people aged 35 to 65, half of whom died between 2016 and 2020, and asked Life2vec to predict who lived and who died. This was 11% more accurate than existing AI models and life actuarial tables used in the financial industry to price life insurance policies.

The model was also able to predict personality test results for a portion of the population more accurately than AI models trained specifically to do the job.

Jorgensen believes the model has consumed enough data that it has a good chance of shedding light on a wide range of topics in health and society. This means it can be used to predict and detect health problems early, or by governments to reduce inequalities. But he stresses that it can also be used by companies in harmful ways.

“Obviously, our model should not be used by insurance companies, because the whole idea of ​​insurance is that if some unlucky person suffers some kind of incident, dies, loses their backpack, etc. ‘Because we share the lack of knowledge about what to do, we can share this burden to some extent,’ says Jorgensen.

But such technology already exists, he says. “Big tech companies that have large amounts of data about us are likely already using this information against us, and they are using it to make predictions about us. It is.”

Matthew Edwards Researchers from UK professional institutes the Institute of Actuaries and the Faculty of Actuaries say that while insurers are certainly interested in new forecasting techniques, the bulk of decision-making is based on a type of model called a generalized linear model. The research is done using AI, which he says is rudimentary compared to this research. .

“If you look at what insurance companies have been doing for years, decades, centuries, they’ve taken the data they have and tried to predict life expectancy from that,” Edwards said. “But we are deliberately conservative in adopting new methodologies, because when we are creating policies that are likely to be in place for the next 20 or 30 years, the last thing we want is to make any significant mistakes. . Everything can change, but slowly because no one wants to make mistakes.”