An overwhelming majority of US citizens are currently skeptical of artificial intelligence (AI) being used in the healthcare sector, with significantly fewer people willing to accept treatment when they know the service is not provided by a human. What many people don’t know, however, is the accuracy levels that such AI systems are capable of achieving.

Pharmacy

Suppose that you’re in need of a medication refill, so you head down to your local pharmacy. The pharmacist there makes an error which leads to you becoming ill. Would you prefer this over an automated machine that has assembled 1.5 million prescriptions per year without one single error? This automated pharmacy has been a reality since its development by the University of California San Francisco (UCSF) in 2010.

For human pharmacists, as workload rises, the error rate increases to almost 5 errors per 100,000 prescriptions, amounting to many millions per year. Medical errors, including incorrect diagnoses and prescriptions made up 10% of US deaths in 2015. Is it, therefore, unethical for us to continue allowing humans to dispense medicines when errors are so prevalent, and in the case of pharmacies, there is clearly a safer alternative?

Automated pharmacies are, of course, worlds away from doctors and nurses who require a much larger skill set than being able to find and prepare a dosage of a predetermined medicine correctly. Even so, with the fast-paced world of AI improving day by day, it won’t be too long before we see virtual systems taking over the roles of many general healthcare workers. Over time, 80% of a doctor’s role is estimated to be replaced by AI. And in many instances, the levels of accuracy will lead to improved safety, while the virtual aspect would free up specialist resources, and keep costs down for both patients and providers.

Diagnosis

The main area that AI could help to elevate within healthcare is primary diagnosis. Patients are already benefiting from AI in this sense without even realizing it, as huge success has been demonstrated in interpreting diagnostic images showing retinal disease and lung cancer, among other illnesses. In such studies, deep learning technology was able to draw attention to issues with scans that doctors may have missed, and as with other machine learning techniques, its accuracy will only improve with the more scans it analyses.

Diagnostic imagery AI is currently providing the much needed stepping stone between patients and their adverse thoughts surrounding virtual healthcare. By laying the groundwork behind the scenes now, hopefully, our trust in AI will gradually grow, and through incremental advancements, we could get to a point where people are comfortable interacting with simulated healthcare professionals.

Such simulations would be based on AI capability to detect diseases based on a range of symptoms presented to it. IBM’s Watson, famously known for winning at Jeopardy! in 2011, reportedly performs at 90% accuracy in lung cancer diagnosis, whereas human doctors only diagnosed correctly 50% of the time.

Watson has recently been specifically adapted for use in cancer treatment recommendations. The program, named Watson For Oncology, is currently being used to analyze symptoms alongside doctors’ expertise in China, as a way to improve accuracy when providing treatment plans to patients. IBM has not yet published any official data demonstrating how effective the technology actually is for patients or providers, however, we are on the right track to technologies such as this becoming commonplace.

There are already mobile apps available to pharmaceutical companies, insurers and employers, which use AI as an interface service to provide symptom assessments, wellness information, and chronic care advice in over 32 different languages. By developing this idea further into a universal system incorporating deep learning, we could relieve huge amounts of pressure within our current healthcare systems.

“And just like in the home, where we’re using Siri and Alexa, the future will bring virtual assistants to the bedside for clinicians to use with embedded intelligence for order entry.” Adam Landman, MD, Vice President and CIO at Brigham Health

Virtual Healthcare Assistants

Imagine that you’ve been feeling under the weather for a little longer than usual. Instead of booking an appointment with your GP, you open an app on your phone and spend as long as you like talking with a virtual healthcare assistant, explaining your symptoms, and whatever else that’s going on in your life that you feel relevant. If there are any physical symptoms, you use the camera to show these. The app spends a few moments processing all the information you provided, as well as every single piece of medical data that has ever existed, and gives you a predicted diagnosis and some treatment advice.

The whole process puts your mind at ease that all you need is some bed rest and lots of fluids, while also saving you the time and effort of leaving the house. Human doctors are able to dedicate their time to treating more than the common cold, and all patients get the level of attentiveness they deserve from the relevant provider, rather than practitioners having to rush through appointments. On top of all this, you also fulfill your good deed of the day by not infecting other healthy people that you would have encountered by visiting a doctor’s office.

In theory, a system like this could revolutionize how we interact with our healthcare systems. In countries such as the US with no universal healthcare, it could be an affordable subscription-based service. You pay a small fee each month or each year and are free to visit your virtual doctor as many times as you like. Today in the US, there is a growing trend of patients avoiding care due to concerns regarding insurance.

With AI, instead of worrying how much each appointment will cost and whether your premium will increase, you log on and have a check-up for each and every concern. This would undoubtedly lead to a healthier culture of getting symptoms checked sooner, and more diagnoses of previously ignored illnesses. As well as reducing costs for patients, AI nursing assistants are anticipated to generate annual revenue of $20 Billion by 2026, a positive outcome all round.

In contrast, the UK’s National Health Service (NHS) is currently being abused, with people choosing to visit emergency departments for issues that could have swiftly been dealt with by a pharmacist, or eventually, a virtual GP. Services like this would benefit the NHS massively, allowing specialists to treat patients more effectively. This is something that the UK is already seriously considering — as of August 2019, the NHS has been granted £250 million in funding towards a new AI laboratory.

On a lighter note, even though a virtual GP could be so advanced that it would seem you were talking to a genuine human doctor, in the back of your mind you’d always know it wasn’t real. Therefore, if your symptoms happen to be a tad embarrassing, you can have them checked out without even having to see a real person or take time off work! Even though this is obviously less of a problem in society today than overworked providers, AI in this situation would also lead to more people seeking help.

“We don’t have enough labor to manage everyone’s health all the time with a doctor and a nurse… So we need this boost of artificial brains to be able to support people.” Dan Housman, chief technology officer at ConvergeHEALTH by Deloitte

Data

Doctors are prone to mistakes, they can not be correct 100% of the time. After all, they’re only human! We as a species, are incapable of being perfect healthcare providers. An AI program, on the other hand, has the potential to hold in its core more data than a human could ever even dream of learning within one lifetime. Additionally, a new medical study is published approximately every 20 seconds and to keep up with this information, a human would need to read 5000 papers a day. With AI there is no such need, updates would be automatic so nothing would be ‘behind’, and the machine would only grow more accurate with each use as all encounters would improve its understanding and increase the volume of patient data sets.

Major obstacles to this technology currently lie in the amount of data we possess and how skewed it is. Future AI technology would be able to analyze your whole genome and use this to inform its diagnosis, however, 96% of all human genomes ever sequenced so far have been from white Europeans. Obviously an AI system based on this data would only be useful to a small handful of people, therefore the challenge is how can we accumulate enough data to make an accurate AI system possible?

Some countries are already toying with the idea of offering genome sequencing to all children born or even making this compulsory. Even though there are ethical drawbacks to this that need to be further explored, this would greatly improve data sets needed to create virtual healthcare workers.

“ …If you have an AI algorithm and lots and lots of data from many patients, it’s easier to match up what you’re seeing to long term patterns and maybe detect subtle improvements that would impact your decisions around care.” Brandon Westover, MD, PhD, Director of the MGH Clinical Data Animation Center

The potential to improve our healthcare systems by freeing resources, reducing cost, and upgrading accuracy levels will likely result in AI being the most ethical path follow. Although we are far from an AI system such as the one described here, we are making progress day by day. No doubt, at some point in the future we will reach a point where the levels of data and capabilities of machine learning technology will be so advanced that a virtual healthcare assistant will become reality.

For a more in-depth look at where we are currently with similar technologies, and the ethical issues being faced today, Kim Thomas’s article around our trust in AI healthcare is a great resource.

This article was originally published by Ellie harris on medium.