Artificial intelligence has been hyped as the next big advancement in healthcare for years. Proponents across healthcare and tech would love for adoption and implementation of AI to pick up the pace, but it’s time that we reset our industry’s expectations.
Healthcare is going to be slower to adopt AI than other industries, and that’s actually a good thing. The healthcare industry needs to be deliberate and methodical with application, development, product goals, and business strategies that go along with engineering and implementing AI across healthcare and medical technology.
AI in healthcare has the potential to encompass every sector within it: payers, pharmaceuticals, devices, health systems, clinical research, and more. Its definition is currently too broad to celebrate an AI advancement in one sector as an achievement across the entire industry.
If a smart algorithm that helps me decide what movie to watch on Amazon Prime suggests the wrong movie, it's not a medical emergency and there aren’t any serious, real-life consequences. But if my AI algorithm that’s automating suturing a patient pulls the suture thread too hard due to an error in the algorithm, that’s going to hurt the patient, or worse.
Over the next few years, our industry is going to see a push and pull where innovation that exists in machine learning, deep learning and artificial intelligence is happening at a much higher level in academia and other non-medical device applications. The application of that technology into medical devices is going to be behind the curve, simply because we need to approach AI implementation carefully. There are significantly higher stakes when human lives are involved.
As innovators, engineers in our field need to take advantage of this lag behind other industries and learn from their successes and failures while we consider how we apply this technology across healthcare.
AI Adoption in Healthcare Remains Limited
A Brookings Institute report released earlier this year explored AI adoption patterns in healthcare and other industries based on job advertisement descriptions. The job postings offer a glimpse into technology adoption patterns across industries. The findings? Out of 40,000 job postings, only approximately 1 in 1,250 job postings in healthcare required AI skills.
The report says this number is lower than other skilled industries like scientific, technical services, and finance and insurance. Healthcare ranked only above the construction industry when it comes to AI implementation. It’s evident that there are barriers when it comes to introducing AI to healthcare, including a cautious regulatory environment surrounding patient data privacy and security, device approvals, and liability concerns.
High Quality Data Needed for AI is Lacking
There’s also the issue of high costs associated with widespread adoption, combined with the need to access high quality data sets. To generate high quality data, our industry needs to develop a more robust variety of sensor devices that collect data in a centralized way. Data sharing and transparency will likely continue to be an issue.
Our industry has been collecting troves of data that isn’t yet being utilized. That data also isn’t necessarily usable or “clean data” -- meaning it needs to be processed in order to extract useful information. That so-called “processing” is unsustainable because it involves humans manually culling the data for valuable insights and extracting that information for use elsewhere. The bottom line is that we need high quality, shareable data to feed algorithms that will enable healthcare to utilize the true potential of AI.
In our development of AI technology in healthcare, it is extremely important that software engineers get firsthand experience inside operating rooms to understand how the technology they are building comes into effect. Engineers who are working to develop AI solutions in the medical field require far more personal and empathetic work than in other fields, like telecommunications or media. It’s critical that they understand the complex, nuanced environment of the operating room in a way that allows them to develop solutions, like AI-integrated electronic medical records, that allow us to bring an AI-supported healthcare environment to reality.
There’s a common misnomer that AI technology is going to become so efficient and powerful, that it will replace humans doing their jobs, an idea that is repeated in response to AI’s entry to healthcare. We are so far away from this being a possibility in our lifetimes, let’s put it to rest once and for all.
Utilizing AI technology in the medical field is going to be a careful, intentional process. It won't look anything like what we’re seeing in other industries. That’s exactly what we want – technology that is tailored to the high stakes healthcare environment. While there are barriers to overcome, this doesn’t mean that innovation will cease in the meantime, as it will present exciting new opportunities for complementary technology development.
In the meantime, we need to proceed carefully and prioritize human wellbeing in a way that ensures transparency, explainability, and intelligibility throughout the entire process. We need to promote data sharing and AI solutions that are responsive and sustainable. Our industry needs to prioritize the right partnerships, development, and integrations in our approach to AI that will positively impact more human lives and advance medical technology in a way that’s more beneficial than racing to the AI implementation finish line.