Is The Human Life Predictable?

Mahanoy City, PA

Have you ever read the book They Both Die in the End by Adam Silvera? It’s about two strangers looking to find a friend in a world where your death is predicted. To some people, it may be amazing to know when you’ll die, but to others, it’s terrifying. Predicting someone’s life seems like an unrealistic idea to the general public. Asking the question, "Is the human life predictable?" has been going on for years. But as it happens, Danish academics have been working over the past few years to figure out how to apply computational science to forecast when a person will die. A study on this topic, led by Professor Sune Lehmann at Denmark University, was published under Nature Computational Science in 2023.

Life2vec is a neural network transformer model that uses artificial intelligence to predict future health outcomes based on data. A neural network is a computer model based on the brain and nervous system, while a transformer model is a data architecture used to learn about language and other tasks. The data includes the specifics of 6 million people from 2008 to 2016. Researchers added major particulars in life such as birth, education, income, housing, and health. Life2vec’s algorithm learns from that data to make predictions about the person; like how they might think or behave. This program was impressive, but it needed to be tested.

So, the researchers picked a group of 100,000 people, of whom half died after 2016; however, the program didn’t know that. When the researchers provided the data, the algorithm accurately predicted the deaths with a total percentage of 78%. Life2vec also showed improvement from other models and baselines since the forecasts were 11% more accurate. The researchers learned a variety of things from these predictions, including how being diagnosed with a mental health problem can lead to a premature death. Gender and occupation were also important factors as well, since the data showed that males and skilled workers died earlier.

However, it’s important to note that Life2vec can’t implement every single factor of life in its program. It can’t predict if you’ll get into a car crash or if someone will put something in your drink. And, like all experiments, it had limitations. The data was confined to an eight-year experience with people only from Denmark, meaning that there may be a social bias. Researchers have not tested the program with data from other countries, so it is unknown what effect the difference in location can have on the results. Additionally, because the researchers could only have access to people who had an income or were engaged in healthcare, the studies weren't randomly selected. This shows that the economic factors of the location can also have an unknown effect. These factors are things that need to be explored so researchers can gain an understanding of this model before it’s used.

To solve this problem, the researchers say that the next step is to incorporate other types of information, such as using text and images to show social connections. This additional information will lead to doors opening in health fields to science fields, sparking new ideas and inquiries. Because of the general skepticism, the argument about whether AI could have a negative or positive impact continues. Some say it takes away the mystery of life, and others say we can truly learn from Life2vec. However, information about this topic will only grow in the upcoming years, so it’s unspecified what will come out of this project until the study is updated.

Sources

https://www.nature.com/articles/s43588-023-00573-5

https://www.cnn.com/2023/12/22/health/algorithm-life-expectancy/index.html#:~:text=Video-,Danish%20researchers%20use%20machine%20learning%20to%20predict%20aspects%20of%20individual,including%20when%20someone%20might%20die&text=Danish%20researchers%20developed%20machine%2Dlearning,their%20likelihood%20of%20early%20mortality.

https://www.dtu.dk/english/newsarchive/2023/12/artificial-intelligence-can-predict-events-in-peoples-lives#:~:text=The%20research%20project&text=The%20dataset%20spans%20from%202008,age%2Drestricted%20subset%20of%20individuals.