In almost every industry, artificial intelligence (AI) is on the fast track to outpacing human endeavor. Machine learning technologies are already better than the average person at gaming, creating content and even building AI, and it appears they are only going from strength to strength.
As a result of their developing intelligence, the most common question AI critics have been asking is whether it’s ethical to be putting ourselves out of a job. YouTube video essayist CGP Grey put it best when he said that, by investing in AI development, we are steaming ahead towards a market in which “humans need not apply” without adequately preparing the population for that scenario.
However, there is another ethical question to ask about superhuman AI: do we truly want all our questions answered? Is there some knowledge that, given the option, we’d actually prefer not to have? Perhaps the most profound piece of knowledge any one of us could have would be knowing when we die. The idea that we could predict death with 100% accuracy has been the subject of art and literature from Ancient Greece to modern science fiction and beyond, and it’s no wonder. The preservation of life is an evolutionary instinct and knowing whether and when that life will end is necessarily part of preserving it.
AI And Healthcare
With regard to preserving and prolonging life, AI already has a very good track record. Frances AI in the hands of medical experts is a truly powerful tool to detect and deter disease. Deep learning technology based on retinal scans was shown to be a good indicator of cardiovascular health and a predictor of potential heart attacks, and also supremely accurate at indicating diabetes with the addition of expert assessment.
The greatest advantage of these early warning systems was the ability to anticipate treatment plans, particularly for conditions with potentially precipitous declines. One such disease is Alzheimer’s, the appearance of which can be hard to notice before the effects are irreversible. That’s why a 2017 study attempted to use machine learning to identify incipient Alzheimer’s dementia in patients. The system predicted the progression of dementia within the next 24 months and was accurate 84% of the time.
AI And Death
Considering all of this, it’s not all that surprising that AI is getting very good at predicting death. The most-quoted example of this was the University of Nottingham’s study last year, which developed a deep- and machine-learning algorithm to predict premature death in patients aged 40 to 69.
Based on health data from 2006 to 2010 from over half a million people within the age range, the deep learning program was “significantly more accurate in predicting death than the standard prediction models developed by a human expert.” What this means in numbers is that the two AI algorithms were able to accurately identify 76% and 64% of subjects who died, respectively, while the human-generated prediction model predicted only 44%.
One of the lessons from the University of Nottingham study is that AI can be used to enhance human predictive models. The two systems used in the study arrived at their predictions by looking at different variables than the human model. While the human model leaned heavily on the ethnicity, gender, age, and physical activity of the subjects, one algorithm focussed on factors like body fat percentage and fruit and vegetable intake, while the most accurate algorithm looked mostly at job-related hazards and the consumption of alcohol and medication.
This means, that far from replacing scientists and healthcare professionals, AI can be used to shed new light on old problems, creating a partnership of humans and machines that could lead to new innovations.
The Ethical Question
However, the question still stands, how much do we want to know about our own mortality? Of course, the ability to identify life-risking habits and behaviors is an invaluable way to prevent unnecessary death and ease the burden on the healthcare industry worldwide. As systems become more sophisticated they will likely be able to identify specific actions and individual decisions that lead to a prolonged or foreshortened life. Insofar as prolonging life is the purpose of healthcare, AI certainly has a future as a tool to enhance the vital work of doctors and health scientists.
But at what point do we begin to shape our lives around the algorithm? Progressing to its logical conclusion, AI systems will likely soon have the ability to accurately predict the life expectancy of anyone. If you know you have 40 more years to live, how will that change the way you live those 40 years? What if it was 2 years?
Furthermore, is it possible that we are building a world in which we allow the predictions of machines to interfere with our ethical choices? A pregnant mother could know with near certainty that their baby will be born disabled from the moment of conception. How will that affect the ethical debate on abortion?
I do not have the answers to any of these questions — I don’t think anyone does — but they begging to be asked. As we develop our technological abilities further, we need to assess how they affect our social and ethical lives.
- 5 Things AI Does Better Than Humans (2019), WireDelta
- CGP Grey, Humans Need Not Apply (2014), YouTube
- Poplin et al., Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning (2018), Nature.
- Schlegl et. al., Fully Automated Detection and Quantification of Macular Fluid in OCT Using Deep Learning (2018), PubMed.
- Sulantha Mathotaarachchi et. al., Identifying incipient dementia individuals using machine learning and amyloid imaging (2017), ScienceDirect.
- University of Nottingham, Artificial intelligence can predict premature death, study finds (2019), EurekAlert!
- Ethics guide: Disability in the foetus (2014), BBC.