Is an echo-chamber effect a threat to AI-driven healthcare?

What is an “echo chamber”?

Imagine living in a closed system, where information, beliefs, and opinions that comply with personal preferences are augmented and reinforced.

Importantly, any opposing or controversial views, are marginalized or blatantly neglected from any sort of thought process or aspect of life.

Here’s how you create echo chambers on Facebook_Take it from the president of the United States: Human beings are creatures of comfort. We’re not particularly inclined…_mashable.com

Why is this dangerous?

An exposure to biased information strengthens our personal beliefs, while other perspectives become underrepresented, can decrease our understanding of different subject matter immensely, and even undermine our willingness and necessity to try to understand the ‘other side’.

A true problem arises when purposely false information is introduced to an echo chamber. Instead of critically assessing the facts, and questioning the trustworthiness of what is presented, the information is easily accepted as long as it fits the chamber’s preferred form and narrative.

Why Facebook's Changes Are Good News For Fake News And The Echo Chamber_The ascendency of the Facebook News Feed as the primary news source for many Americans was a disaster, and to their…_www.forbes.com

But how does this relate to healthcare?

Let’s run a small, thought experiment.

Imagine a national healthcare information system, in which most of the medical recommendations are data-driven and near-completely autonomous.

Let’s envision an idealistic scenario, in which patients follow strictly prescribed treatments and progress of their therapy is closely monitored and logged.

Finally, let’s assume the system is recursive and continuously learning, so that the outcome of the treatment of one cohort is used to refine the prognosis and therapy scenarios for the next group of patients.

Now, let’s extend our dystopian thought experiment to millions of patients and run this hypothetical programme for e.g. 10 years.

Without a comprehensive inclusion of sample diversity; e.g. precision medicine data; measured at different (and likely random intervals) and certain level of stochasticity (introduced by human practitioners), proposed system is doomed to misdiagnose, and likely claim many human lives.

Why?

A closed-information system, such as the dystopian, AI-driven healthcare model presented above, will suffer from a prohibitive data granularity and autocorrelation.

After a defined number of diagnosis — treatment rounds, our automated and hypothetically efficient AI-driven healthcare will resemble a social media “echo chamber”, in which phenotypic diversity is replaced by granular and “diffused” definition of medical disorders.

Such system, will be at great conflict with (likely) the most important and disruptive medical trend of the beginning of the 21st century — precision medicine. A movement which works towards embracing our genetic diversity and building healthcare solutions around it!

Can we build a sustainable AI-driven healthcare?

The last 50 years have seen tremendous progress in almost every aspect of modern medicine. Thus, the spectrum of curable diseases, is a dynamic variable, which evolves with time and our general understanding of human physiology.

Ultimately, lifestyle adjustments and semi-correlated (let’s envision a future, when CRISPR can fix human genes, heavily influencing natural selection process) population genetics, will yield a society that faces completely different medical issues than early generations. This notion has been very elegantly covered by Niamh McKenna, in her short and thought-provoking article.

AI: Real-time Aging Citizen Healthcare Transformation | Accenture_How to manage healthcare when millions are over 100 years old? AI will teach us - on the fly. Read more._www.accenture.com

Thus, I strongly believe the key to complete and inclusive AI-driven healthcare is in the data and the way they will need to be collected.

We will need to embrace genetic diversity of human population, and sample from it according to a standardized protocol (to ensure data compatibility and reproducibility of our procedures).

Since, population genetics is spatio-temporal (it evolves with time and hypothetically our medical interventions), we will need to learn to sample from it at (likely) random intervals at a pace which outmatches expected population-wide evolutionary changes.

Finally, a coherent and self-tuning healthcare system will make a use of human practitioners, as a source of stochasticity and orthogonal diagnosis.

Summing up

I can envision a future in which we build an inclusive and comprehensive AI-driven healthcare system, in which human practitioners complement but also challenge data-driven treatment of medical disorders.