Artificial intelligence in healthcare has had significant hype over the last several months. Open your email inbox and you will find multiple medical journals emailing about “AI in healthcare.” However, AI is hardly a new presence in medicine, or the world; it is simply that we are reaching a day-and-age in which previously only imagined possibilities are becoming realities. Over the last decade, we have seen significant advancements in AI-driven technologies, such as digital assistants (Alexa, Siri, etc) and self-driving vehicles. In November of 2022, the popular chatbot ChatGPT was released to the public and has been credited with harkening the current AI boom. Undoubtedly, we are in an AI revolution; the questions are not “if” or even “when,” but “how much,” and “how do we responsibly manage” this new age of development?
What is Artificial Intelligence (AI)?
IBM defines artificial intelligence as “technology that enables computers and machines to simulate human intelligence and problem-solving capabilities.” In short, it is the digitization of human probabilistic thinking, reasoning, and logic through very complex arithmetic and algorithms, which has been powered by increasingly stronger computers and supporting hardware. While a detailed explanation of the algorithms behind AI are beyond the scope of this author and column, a cursory understanding of machine learning is necessary.
The concept of artificial intelligence has been around throughout human history, fantasized about for centuries by dreamers and philosophers. It conjures far-out ideas of automated living like The Jetsons, as well as less innocent stories of apocalyptic robotic takeover (e.g., I, Robot). But the transition from fantasy to reality began in the 1950s when Alan Turing began the conversation about machine learning. Early history of artificial intelligence included many decades of building and refining gaming computers like Deep Blue, the infamous opponent of chess champion Gary Kasparov. These basic concepts have evolved to the current day in which speech, language, and image recognition, and computer data processing are more sophisticated than ever.
Figure 1: Artificial intelligence is an umbrella term for multiple fields in which technology is used to performs tasks similarly to humans. Machine learning and its sub-fields (eg, deep learning) are a sub-set of AI. (HPE Community.)
Machine learning, a sub-field of AI, involves programming in which algorithms instruct a machine to employ one of three types of “learning” toward solving a problem: supervised, unsupervised, and reinforcement. Supervised learning involves a “training dataset” in which human-labeled data are provided to the computer, and a “testing” dataset assesses accuracy of performance. An example would be object identification within an image, such as a malignant skin lesion. Unsupervised learning occurs when a machine is instructed to identify patterns within data that it receives. Its potential application might be to analyze a human genome sequence to identify patients who may benefit from a particular drug treatment. Reinforcement learning is an algorithm in which a machine is instructed to perform a task and then “learns” from its mistakes and successes to improve on its response or intervention.
Neural networks and deep learning are sub-disciplines of machine learning that allow multiple layers of complex data processing. Deep learning is unsupervised, reinforced learning on a massive scale, allowing computers to function, learn, and refine without continuous human input. Deep learning is behind the complex processes in use with digital assistants, self-driving vehicles, and many applications within healthcare.
Current Use in Healthcare
At the end of 2023, there were nearly 700 AI-driven medical devices with FDA approval for use in the medical field. Some uses involve simple administrative tasks, such as assisting with or automating scheduling to improve office efficiency. More complex applications have included fracture detection on x-rays (a task at which AI performs on par with humans), as well as assistance in detection of adenomatous polyps during colonoscopies (AI proved to be helpful at times when providers might be fatigued).
Figure 2: Digital mediums have placed enormous amounts of data at the fingertips of healthcare workers, sometimes overwhelmingly so. One potential promise of AI is the ability to process, organize, and analyze that data faster than any human might be able to in order to provide coordination of care and unforeseen insights. (World Health Organization.)
Though there are currently significant limitations, symptom trackers help streamline emergency room triage and registration. In the prehospital setting, AI has been shown to improve emergency medical dispatching algorithms to hasten ambulance response, reduce call wait times in mass casualty situations, and one particular application combines caller speech recognition with an out-of-hospital cardiac arrest detection model to aid dispatchers in detecting major events such as heart attack and stroke.
With vast amounts of information in medicine and electronic health records, AI has the potential to offer significant assistance by making sense of the data, automating processes, and freeing up time for providers. This does not come without some significant concerns, though.
Concerns and Pitfalls
In 2021, the World Health Organization released a guide, Ethics & Governance for Artificial Intelligence in Health, outlining ideas for an approach to regulation of AI in medicine. The United States Senate held a hearing in February of this year to begin the conversation between policymakers, physicians and tech company representatives.
There are numerous and varied concerns regarding the use of artificial intelligence in healthcare. Moral and ethical concerns include consent for use of personal medical data in AI systems, how to protect against breaches of sensitive data, and how to assign liability to companies developing AI in the event of adverse outcomes that occur from use of these technologies. While the promise of AI could be a more equitable distribution of healthcare in underserved and socioeconomically challenged communities, there is concern that high technology may only worsen disparities between classes.
One problem already identified with AI algorithms is the perpetuation of implicit and explicit human biases, particularly racial biases. Because AI relies heavily on human input, these biases are built into the data, making findings and potential medical recommendations unsafe for some patient populations. Thus, the plea is for tech companies to provide transparency about algorithms, and to work closely alongside medical professionals to build systems that are appropriate and trustworthy. This may require intermittent human oversight to certain algorithms, particularly deep learning models that are built for unsupervised functions.
Moreover, there is considerable unease about how the use of artificial intelligence will affect the knowledge and skills of healthcare providers currently in training. If machines are relied upon to analyze image findings, electronic health data, etc., will our skills atrophy? What AI competency requirements will become necessary for current practitioners and trainees? And what implications will AI have for job security?
Future Implementation – AI’s Role in Wilderness Medicine?
One advantage wilderness medicine providers may have is a need to rely on soft and hard skills more so than urban or front country providers owing to the limited tech presence in the back country and austere environments. Though future development of small, portable technologies and even the possibility of worldwide internet access could significantly extend tech into the wilderness, there will likely remain a need for human creative thinking and involvement in outdoor scenarios.
With that being said, AI has the potential to significantly assist wilderness medicine providers. Image recognition software on cell phones can help identify poisonous plants or unlabeled pills. Language processing tools can help translate hundreds of spoken and written languages. And portable medical devices such as smart watches, electrocardiogram software, and ultrasound could be augmented with AI to provide a more detailed picture of a backcountry clinical scenario, potentially offering specialty-level guidance when it would otherwise be absent.
Another potential application could be in backcountry search and rescue scenarios. Though drones are already being used to assist in SAR missions, AI could further augment these efforts with their GPS capabilities, or heat-sensing technologies, for example in avalanche burials. Perhaps even in the distant future, AI-driven devices or robots could be deployed into remote environments to perform rescues more quickly and safely than humans. Even now, virtual reality and AI-augmented simulation environments can enhance training scenarios for rescuers by closely replicating dangerous wilderness environments.
AI-enhanced possibilities truly seem endless. There are certainly negatives to be aware of, and ethical and regulatory concerns to consider. But the AI revolution will be an amazing thing to witness and be involved in as development continues.