Messaging has become a popular form of communication in the connected world. We are in an era of increasing automation, and chatbots are becoming the new way that patients are getting answers and solutions for their problems. Other industries have benefited from the administrative time and cost-saving benefits of using chatbots to handle simple interactions. Through increased use, chatbots are constantly applying artificial intelligence learning to improve their interactions with humans. The more conversations that the chatbots have, the more they learn to communicate better. But are they a good fit for healthcare? One of the first prototype chatbots, created in 1966, called "Eliza" was actually a psychotherapist that could simulate a conversation. It wasn't a genuinely intelligent conversation, but we have come a long way since then.

Most of us, who have interacted with a chatbot, can reference a customer support interaction. Generally, these are more pleasant than sitting on an automated phone call waiting to get to talk to a human. The chatbot can provide immediate basic information, which solves the majority of requests, and forward more complex requests to the right person who can help. As the population in the United States ages and demands for care increase, the physician shortage we are experiencing now will be more problematic. Automation can help physicians to make patient interactions more efficient and reduce administrative time.

For example, a chatbot can help guide patients through the appointment-making process. As at-home wearable health monitors become more widely used, they can access personal health monitor data, take into account the patient's current condition, and suggest a referral. Besides finding the right time, the bot can help the patient figure out how much the visit will cost and even assist in getting them a ride to the appointment.

The administrative process when visiting a healthcare provider can take as long as the actual direct interaction with clinicians. All of this paperwork can be handled by the chatbot when scheduling an appointment on the provider's website, saving patients time and repetition as well as office administration costs. The chatbot can collect first-level triage information about a patient's symptoms and have it ready in the EHR for the clinician. There is an added benefit here, depending on the clinician's workflow. If they review the information before consulting with the patient, they have that much more time to digest information and come up with the best next steps. Another example of help chatbots can provide is the system that helps pregnant mothers decide when to make the trip to the hospital by asking about contraction frequency and duration. This chatbot remembers conversations and keeps track of the information so it asks relevant questions as the interactions continue.

Automated text reminders for appointments have been around for a while, but now similar and more sophisticated bots are being used for prescription reminders and physical activity coaches. Bots can remind patients or caregivers to change a dressing if someone has had surgery or is injured. They can remind people not to eat the night before surgery (an important one!).

Bots can act as the first line of communication for a medical concern at any time of the day. They can help with decision making if a patient feels like they are sick. Through a series of questions asked by the chatbot, a patient can determine what steps they should take. Is it something they could ride out with some rest? Should they make an appointment with the PCP? Or is it something that needs immediate attention in the emergency room? Although chatbots aren't making a diagnosis, they can help the patient determine next steps for care.

There are some problems and shortcomings. Chatbots lack the capability for emotional support that patients may need. Unlike banking or retail, where chatbots can help customers with relatively little risk of doing any harm to them, medical applications have a different set of problems to be solved. A recent study showed physicians see extensive administrative benefits, but don't believe that bots can provide accurate diagnoses and emotional support. They also pointed out that bots could be harmful if a patient overuses them to self diagnose (Semi-sentient Dr. Google? Pretty scary thought). Although they can help with triage, bots can't replace the judgment of a doctor and understand all of a patient's complex personal factors. We also need to be aware of encoding our existing biases into chatbots. Care must be taken to test chatbot interactions for diverse people and unique concerns. The recent stories of sexist credit card algorithms are bad enough, let alone applied to medical care.

Chatbots will improve as time goes on and provide extensive benefits to patients and physicians alike. Their current state of intelligence could lead to some infuriating, and in some cases, harmful experiences. Initially, human supervision is the best way to be sure that chatbots can handle their tasks. There should always be a "human bailout" until functionality and reliability become more consistent.

Have you interacted with a chatbot today? What dangers or benefits do you see for chatbots in healthcare?