What is an AI Bot?
Artificial Intelligence (AI) bots are computer programs that notice keyword patterns in writing and provide rational answers to questions. You’ve probably already interacted with a few of them: Apple’s Siri and Microsoft’s Cortana are simple and widely accessible applications of the incredibly broad field of artificial intelligence research.
AI bots are becoming increasingly skilled at mimicking human speech patterns. So much so that a recent class of online graduate students didn’t realize that their questions were being answered by a chatbot teaching assistant, powered by the same IBM technology that recently bested the reigning champions of Jeopardy.
But there’s far more potential in AI than just watching a computer win a game show, or having your phone locate a restaurant for you. AI bots are also poised to impact healthcare services, and this could further advance remote patient monitoring and improve medical outcomes.
Artificial Intelligence for Healthcare
Healthcare AI bots are not about tricking your patients with computers masquerading as doctors. Rather, healthcare AI will incorporate technology your patients are already familiar with, reinforcing the professional judgment of clinicians and strengthening evidence-based best practices in outpatient care. Research continually finds that incorporating mobile technology into outpatient programs increases patient engagement as well as patients’ adherence to their physicians’ advice—and both of these are consistently associated with improved medical outcomes [1,2].
Using healthcare AI bots, patients can chat with an app about symptoms and self care. Responsive AI bots can quickly understand which follow-up questions are relevant for patients, and present the information that best fits an individual patient.
With AI apps designed for particular outpatient programs, patients can access personalized medical resources without needing to schedule a clinic visit. And with interoperable AI apps, collected data can be securely transmitted to clinicians, providing context for the biometric data collected on wearable biometric devices.
But with the potential for data-driven healthcare, many clinicians worry they will find themselves spending more time with statistical spreadsheets than with patients. This doesn’t have to be the case. The best mobile health programs algorithmically synthesize data to ensure that clinicians are presented with figures that support clinical decision making. And decision support algorithms are a key component of what AI researchers call “machine learning”.
Machine Learning for Mobile Health
Machine learning is a branch of artificial intelligence that enables programs to make predictions based on collected data, continuously recalibrating in response to new information, without the need to be reprogrammed. Researchers are already applying machine learning technology to help predict the onset of seizures for persons with epilepsy.
Machine learning can also be incorporated into congestive heart failure outpatient programs. A simple app can remotely measure and report a patient’s heart rate. An AI bot can ask a patient to provide context on times when heart rate is elevated, asking about particular stressors such as physical activity or emotional well-being.
A healthcare AI bot with machine learning can go further, predicting that a patient will have an elevated heart rate during their weekly tennis match, and prompting a patient to prepare particular medications. Machine learning health apps will synthesize this data to help care teams better understand individual patients, measure medical outcomes, and intervene before chronic conditions become medical emergencies.
- Nakamura N, Koga T, Iseki H. A meta-analysis of remote patient monitoring for chronic heart failure patients. Journal of Telemedicine and Telecare. 2014 Jan;20(1):11-17.
- Hamine S, Gerth-Guyette E, Faulx D, Green BB, Ginsburg AS. Impact of mHealth Chronic Disease Management on Treatment Adherence and Patient Outcomes: A Systematic Review. Journal of Medical Internet Research. 2015 Feb;17(2):e52.