A Computer Could Become Your Doctor

Artificial Intelligence in Healthcare: Its Utility and Ethical Limitations

While these results could be surprising and even fear-inducing for those of us who believe that professionals such as radiologists will soon be replaced by AI, it is worth noting that these automation technologies are far from attaining the multidimensional expertise a medical doctor is required to have when treating his patients. The previously mentioned computer program has only demonstrated superiority in one isolated step of establishing a life-saving diagnosis.

In light of this innovation,

if healthcare services were made more efficient thanks to automation;

if physicians could offer more accurate and adapted diagnoses thanks to computer-assisted predictions;

And if AI alone, accessible through freely available mobile apps and software, has one day the power of providing quality healthcare to populations in remote locations where it was originally impossible;

We could assist at the dawn of healthcare democratization.

Before immediately integrating AI in our healthcare systems, we should start by investigating the place it could take in the strong relationship of trust that patient and physician share. More importantly, it is imperative to look at the practical risks and ethical dimensions that surround the use of such revolutionizing technology.

In favor of using AI in our hospitals and clinics, we can start by acknowledging that it can help us democratize access to healthcare around the world by saving us time and resources. Thanks to automation, medical procedures could be made cheaper and thus more accessible to the global population. This capacity for large-scale efficiency is explained by the numerous types of paid human labor AI could undertake or assist if it were implemented. Repetitive administrative tasks such as data entry and medical record management could be made faster if done by an algorithm fine-tuned to accomplish such tasks.

Easing the installation of such technologies in healthcare, the large amounts of patient information hospitals collect compose vast and considerably representative data sets that can be given to machine learning algorithms in order to produce accurate predictive models. Another reason to believe that we will likely be seeing widespread use of this technology in the near future. (Freethink media)

Indeed, if people were to have access to reliable healthcare services capable of establishing accurate diagnosis and recommending appropriate treatment, many preventable deaths could be avoided. In fact, prevention of life-threatening diseases is best done with early detection of disease symptoms. AI could greatly help in doing so, notably, in the prevention of the most important cause of death globally: cardiovascular disease (World Health Organization).

As an example, when analyzing the performances of a popular algorithm used in the U.S. healthcare system to evaluate the need for care of patients presenting health risks, Ziad Obermeyer and his colleagues found that “Black patients assigned the same level of risk by the algorithm are sicker than White patients”; in fact, the authors determined that this manifestation of racial bias due to the design of the algorithm “reduces the number of Black patients identified for extra care by more than half”.

Some would argue that this issue is inevitable and that such algorithms should be trusted as long as they are proven to be accurate, much like the use of a drug that would be validated after successful clinical trials (Gerke et al.).

These adversarial attacks seem unlikely to happen in practical settings considering the large number of pixels one has to change in order to successfully conduct one. However, Su and her colleagues determined in their study that such attacks can be successful by changing only one pixel (Su et al; Fehér). Indeed, if the right pixel is chosen, a deer can be mistaken for an airplane with 85.3% certainty by the image recognition algorithm (see fig. 2) (Su et al). Knowing that these one-pixel attacks are possible, it is worth questioning the relative safety of using such complex algorithms in healthcare settings like radiology where one might benefit from the image recognition capabilities of AI.

Notwithstanding, when faced with this type of usually accurate model which’s functioning can sometimes be hardly interpretable, how could a physician fully trust the model’s judgment knowing its susceptibility to such minimal perturbations?

Davenport, Thomas, and Ravi Kalakota. “The potential for artificial intelligence in healthcare.” Future healthcare journal vol. 6,2 (2019): 94–98. doi:10.7861/futurehosp.6–2–94

“Ethics Guidelines for Trustworthy AI.” European Commission, 8 Apr. 2019, ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai.

Gerke, Sara et al. “Ethical and legal challenges of artificial intelligence-driven healthcare.” Artificial Intelligence in Healthcare (2020): 295–336. doi:10.1016/B978–0–12–818438–7.00012–5

Obermeyer, Ziad, et al. Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations. 25 Oct. 2019, science.sciencemag.org/content/366/6464/447.full.

Su, Jiawei, et al. “One Pixel Attack for Fooling Deep Neural Networks.” ArXiv.org, 17 Oct. 2019, arxiv.org/abs/1710.08864.

Szegedy, Christian, et al. “Intriguing Properties of Neural Networks.” ArXiv.org, 19 Feb. 2014, arxiv.org/abs/1312.6199.

The Medical Futurist. Top A.I. Algorithms In Healthcare. 7 Feb. 2019, medicalfuturist.com/top-ai-algorithms-healthcare/.

Related posts:

Ethereum has announced its plans for a big shift in its consensus mechanism which will drastically change the way its blockchain operates in the future. This change will have a direct effect on the…

The number of smart contracts has demonstrated exponential growth since the launch of the Ethereum Virtual Machine. The Ethereum Requests for Comments or ERC provided a number of standards that have…

You would not think making a book trailer would be that hard, but I’m here to tell you that for someone like me — at times technically challenged, a novice as to how to manage microphones, lighting…