Health IS Technology Blog

Chatbots: From Shoe Sales to Healthcare


Robot customer service operator with headset and speech bubbles

There are many different types of chatbots, and each one has a different level of ability. Some are able to give a preliminary diagnosis of a medical condition. Some are able to give advice to moms new to breastfeeding. There’s even a bot capable of ordering Domino’s pizza for you!

What is a Chatbot?

A chatbot provides a service and is often powered by artificial intelligence (AI) and interacted with through a chat interface. A chatbot can provide assistance in a number of services. Facebook messenger hosts multiple chatbots, including the weather bot Pancho and a news bot through CNN. Another common way bots are utilized is within retail and product purchasing. An example Chatbots Magazine gives involves someone searching for shoes on the Nordstrom website.

“If Nordstrom makes a bot, which I’m sure they will, you would simply be able to message Nordstrom on Facebook. It would ask you what you’re looking for and you would simply… tell it. Instead of browsing a website, you will have a conversation with the Nordstrom bot, mirroring the type of experience you would get when you go into the retail store,” said Matt Schlicht, Founder of Chatbots Magazine

Just like other forms of AI, chatbots utilize machine learning to improve their efficiency. Bots require a lot of practice in order to better understand the different facets of their specialty.

For example, Public Health England has launched a bot through Facebook Messenger that specializes in giving advice to mothers new to breastfeeding. But, because the bot is new, it has a great deal of bugs to be worked out. Rachel Thompson from Mashable tested it out, and reported that very simple, common questions went unanswered with the bot. When she asked, “How often should I nurse in a day?” the bot responded, “I didn’t quite get that.” This inability to answer such an imperative question is likely due to a lack of machine learning. As the bot interacts with more users, it will gradually develop new responses to new inquiries. Still, even in its infancy, this bot showcases how the technology is pushing different fields toward innovation.

The Turing Test

In the Turing Test a human interacts with a computer and if they don’t know they’re interacting with a computer, then the test is passed. A passed Turing Test means that the computer has achieved artificial intelligence. If a computer program fools human judges more than 30% of the time during a series of five minute typed conversations, then that computer is considered successful in terms of the Turing Test.

Phone customer service chatbots

Customer service chatbots

This test has served as the baseline for how a machine should perform for a long time. But a lot of developers have acknowledged that AI can be designed in specific ways that deceive the Turing Test. For example, in 2014 chatbot Eugene Goostman passed the test with a rate of 33% deception. Eugene was based on a thirteen-year-old Ukrainian boy who spoke English as a second language. Many argue that the bot’s syntax was designed to have certain convolutions that a child learning a new language might have, and because of this, it easily tricked judges. To oversimplify, some think this characterization of the character was cheating. Others think the Turing Test is outdated and should no longer be considered the standard.

“Bots do not need to pass the ‘turing test’, they need to pass the ‘beer test.’ The turing test is ‘Would you think this is a human on the other side?’ The beer test is something that I
invented which is ‘Would you take this bot out for a beer after receiving the service? Is this a delightful bot?’ Bots need to be delightful and need to provide you with the service, and you should feel good after the service,” explained Amir Shevat Head Developer of Relations for Slack in a panel discussion seen on YouTube.

Chatbots in Healthcare

Medic chatbot

The use of chatbots in healthcare is on the rise.

The use of chatbots in healthcare has been on the rise lately. Companies like Your.MD, Babylon Health, and Baidu all have worked to develop healthcare bots. Patients can interact with these bots through messenger interfaces. Someone might send one of these bots a message like, “My head has been hurting all morning. What can I do to make it stop?” Using its machine learning, a healthcare bot will analyze the question it is presented with and generate an intelligent response. 

It’s clear that designers of these bots do not aim to replace human healthcare providers with bots. Instead, the chatbots are designed to provide preliminary diagnoses. So, to the patient with the headache, the bot may suggest an over-the-counter pain killer. Then, the bot might say, “If that does not help, you should consider going in and seeing your doctor. If it does help, your headache isn’t something to worry over.” In this sense, interacting with a chatbot is a lot like reaching out to a friend or family member for advice, only the bot has a breadth of knowledge programmed inside of it.

Can Chatbots Replace Human Interaction?

robot typing on a computer

Designers do not aim to replace human healthcare providers with chatbots.

The simple answer to this question is no. Chatbots cannot replace human interaction, at least not anytime soon. But they do have the ability to replicate it. Sentimental intelligence is a term for when a bot “knows” what a user is feeling. Sentimental intelligence is also seen where bots are designed to understand empathy. Of course, technology can only be taught to imitate a human emotion. Bots do this by can analyzing a user’s facial expression, and replicating it on its own virtual face. This should make a user feel that they are talking to a human. Microsoft has even developed an API that can analyze the emotions on someone’s face. On their website, you can upload your own photos to test their technology yourself.

In some ways, being able to interact with an intelligent being that isn’t human is beneficial to a cause, like a patient being more open about taboo behavior. But, on the other hand, interacting with something that isn’t human can easily discourage users. This is so likely that many bots are designed to recognize emotions on users’ faces that may suggest they are irritated by speaking with a bot. If a bot detects this, it will suggest the user be connected with a human employee, and do so.

Where Do We Go from Here?

A lot of developers predict that chatbots could one day replace mobile apps altogether. Of course, the apps that host the bots will still need to operate, but if buying a pair of shoes were as easy as texting your friend, wouldn’t you do it?

In terms of healthcare, as already mentioned, it’s unlikely that chatbots could replace humans, but they could help to support us. Regardless, it’s clear that the future will only bring more chatbots and with them, greater progress.

 

Happy Computing!