That could lead to a curious predicament. In the future, as machine learning advances and more AI is used to help with decisions, Tochinni argues that people might respond by purposefully turning off those features – in the car and on their phones – and rely on intuition. As many AI experts suggest, the best scenario is to combine machine learning with human insight. An AI system might make mundane decisions (which coffee to buy) but not complex decisions.
Virtual digital assistants
In case you haven’t clicked on the small circle to the lower-left of the taskbar in Windows 10, that’s a new virtual assistant named Cortana. (She comes from the Halo video game franchise.) There are some powerful AI routines at work that can help you order an Uber car, set a reminder for later in the day, track a FedEx package, and provide details about meeting participants culled from LinkedIn. From an AI standpoint, Cortana learns which features you use most often. For example, if you usually click on meeting reminders and open your calendar, you’ll see more of them. When Cortana sees an upcoming trip, she will download maps of the area and find exchange rates.
Facebook recently introduced a new AI-controlled assistant as part of the Facebook Messenger app. It works for only a limited group of people in Silicon Valley, and the robot is assisted by humans. You can have M complete complex tasks such as book a flight. As Mike Schroepfer, the CTO of Facebook, noted in a post recently, Facebook is slowing expanding what M can do. The company recently automated the process of ordering flowers. The AI now asks for a budget for the order and where it should be sent before a human operator takes over.
Another new virtual assistant is called Amy (or Andrew, depending on the robot you choose). It was developed by the company X.ai. The idea is simple. When you want to schedule a meeting with someone, you skip all of the back and forth required to discuss times, location, and topics. Instead, Amy or Andrew take over by asking meeting participants about the meeting using natural language that mimics an actual assistant. Often, the participants don’t know it’s a bot. (When you receive the email, Amy or Andrew use the last name Ingram.)
As Dennis Mortensen, the founder and CEO of X.ai, explains, this kind of robotic assistance reduces the number of emails between participants from about eight down to just one. “All I want to do is to talk to John, and that eight-email long ping-pong, which comes along with the job of negotiating date, time and place, is a great example of information overload,” says Mortensen. “Those emails add nothing to the meeting itself; if anything they detract from it.”
Of course, digital assistants can’t always help with every task – they won’t be able to type up a business document or make phone calls. At least, not yet.
“By the end of 2016 there will be a significant increase in the launch of “smart apps” – applications that combine contextual awareness and machine learning to deliver elevated insights. For example, virtual digital assistants will become increasingly prevalent within the enterprise. These digital assistants will be able to use contextual clues and past interactions to predict what users want and offer more intelligent recommendations, without them needing to ask,” says Johan den Haan, the Chief Technology Officer at Mendix, a cloud provider.
“The next generation assistants will likely be able to keep track of the user's goals and intentions and develop and execute multi-step plans,” adds Subbarao Kambhampati, an AI expert and professor at Arizona State. “While learning will still be important, the systems will be exploiting the learned knowledge to support richer inference needed in sequential tasks.”
Where do we go from here? Kambhampati says that we are in a new digital frontier where having an email helper is not too far-fetched, someone who learns about our preferences and behaviors in the same way a real assistant might. Yet, there is a lot of research, he says, into making sure that these bots can build trust. It only takes one errant email to destroy it.