Robots have already replaced humans in many jobs, from manufacturing cars to filling warehouse orders and flipping burgers. But as artificial intelligence programs learn to better communicate with humans, they’ll soon encroach on careers once considered untouchable, like law and accounting.
These chatbots may one day even replace your doctor.
This January, the United Kingdom’s National Health Service launched a trial with Babylon Health, a startup developing an AI chatbot. For six months, Babylon will offer a triage service on its app as an alternative to dialing the U.K.'s 24 hour non-emergency telephone number.
The bot’s goal is the same as the helpline, only without humans: to avoid unnecessary doctor appointments and help patients with over-the-counter remedies. Using the system, patients chat with the bot about their symptoms, and the app determines whether they should see a doctor, go to a pharmacy, or stay home. It’s now available to about 1.2 million Londoners.
But the upcoming version of Babylon's chatbot can do even more: In tests, it's now dianosing patients faster human doctors can, says Dr. Ali Parsa, the company’s CEO. The technology can accurately diagnose about 80 percent of illnesses commonly seen by primary care doctors.
To find a diagnosis, the bot chats with a patient about his or her history and symptoms, creates a probabilistic model, and then asks more specific questions to whittle down the likely culprits — much the same way a doctor would. The result is a diagnosis in consultation with a medical professional — at least for now. There will be some regulatory and legal issues to overcome before the AI can work independently of its human counterparts.
The AI backbone of Babylon Health's chatbot is based on about 300 million strings of information — including established medical facts to aid in diagnosis, and regular updates, like input from all 11,000 journal articles on dermatology published last year. It’s the largest collection of knowledge in primary care medicine, Parsa says.
The bot then uses machine learning to improve its diagnoses based on outcomes.
The reason these chatbots are increasingly important is cost: two-thirds of money moving through the U.K.’s health system goes to salaries. “Human beings are very expensive,” Parsa says. “If we want to make healthcare affordable and accessible for everyone, we’ll need to attack the root causes.”
Get the mach newsletter.
Humans will always have a role in caring for patients, Parsa says, because they’re much better at cognition and empathy. “A machine can’t put hand on you and say, 'Look I’m going to take care of you.'”
But machines can assess the probability a patient will do well on certain medications, and examine relevant data sets to extract information. “No doctor loves doing diagnosis; doctors like taking care of people,” Parsa says. “So let’s allow machines to look into the data and tell you probabilities.”
Globally, there are 5 million fewer doctors today than needed, so anything that lets a doctor do their jobs faster and more easily will be welcome, Parsa says. Half the world’s population has little access to health care — but they have smartphones. Chatbots could get them the help they need.
It might sound like AI puts all human jobs on the chopping block, but the story of technology is one of augmentation, not replacement, says Ross Goodwin, a data scientist who specializes in AI.
“Most skepticism comes from the fact that we’ve read a lot of fiction about robots,” he says. “AI taking everyone’s jobs makes a good story; but less of a good story is AI helping people in mundane, simple ways.”
Even just asking how someone is doing today. That’s what a Facebook Messenger bot named Joy does to help track people's mental health. The bot’s creators say they want Joy to seem like a friend instead of a chart — even though Joy can offer weekly summaries. The aim is for these bots to be a gateway, encouraging a conversation that eventually culminates with users seeking qualified help if needed.
“I don’t think there’s any places that AI can’t contribute,” Goodwin says.
AI chatbots aren't all about life and death. When Joshua Browder started driving in 2015, he contested so many parking tickets that he was inspired to create DoNotPay — a chatbot that uses AI and algorithms to help people fight the system.
The app asks a series of questions about the violation, from why you believe you shouldn’t have been ticketed to specific details about the offense, then uses the responses to produce a 500-word letter with exact legal language to contest the ticket. DoNotPay’s success rate is about 60 percent, Browder says.
Just a year into the project, Browder’s app has helped drivers in London, New York, and Seattle overturn 200,000 parking tickets totaling more than $6 million in fines. The system is expanding this year to cities including San Francisco, Chicago, Denver, and Los Angeles.
That’s just the beginning. Browder, now a student at Stanford University, has also created a project to assist refugees as they complete immigration applications for the United States and Canada. Through a conversation on Facebook Messenger, the chatbot first determines if a person is eligible for asylum, then takes down relevant details and fills out his or her forms — many of which are written in complex language.
“There are so many lawyers charging hundreds of dollars for filing a few documents,” Browder says. “People are terrible at describing their issues; chatbots are really good at translating human input into legalese.”
Browder sees bots soon moving into other areas of law. “Anything you see on late night TV advertising will be fair game: pensions and benefits, workplace harassment, injuries, and so on.”