April 7, 2015
The future of artificial intelligence Part II: Smarter and smarter
In part one of our look at the future of artificial intelligence, expert Charlie Ortiz told us that to envision a future in which machines transcend their programming and become a problem is too negative a view to take, despite such suggestions from high profile figures like Stephen Hawking and Elon Musk.Ortiz also said that machines with intelligence to rival our own were “nowhere near” being a reality. In this follow-up piece, we’ll find out what the more realistic, short-term future of AI will look like, and what sort of exciting new uses we can look forward to.
“In terms of the near future what you’ll see are intelligent assistants becoming smarter and smarter,” said Ortiz, who is Senior Principal Manager of the Artificial Intelligence and Reasoning Group within Nuance’s Natural Language and AI Laboratory. “They will help us in everyday tasks, making our lives less stressful and getting rid of the drudgery of day-to-day life.”
An assistant could efficiently direct our chores around town during a Saturday, leaving more free time for ourselves, and then plan a night out that’s going to go smoothly instead of being wasted through poor organization (so we can get wasted through efficient organization).
One of the major steps will be to improve the way in which AI systems understand language. Up to now they have learned to recognize sounds, but they don’t necessarily understand meaning.
For instance, says Ortiz: “You might want to tell your self-driving car ‘take me to the park near the grocery store’ without having to give it a specific address. That ability to refer to things in the world and what you want a device to do in abstract terms, like we do as humans, is something that a lot of people are working on.”
Testing the common sense of AI
Some of those people might take Nuance’s Winograd Schema Challenge. The test aims to measure the “common sense knowledge” of new technology, which really is essential for true AI.
Ortiz explains: “If I tell my personal assistant ‘I’d like you to book a reservation for dinner after my meeting’ and it goes and makes a reservation for next week, technically it is not incorrect because next week is after my meeting, but it’s not what you meant.” But, he continues, “If I tell my son ‘make sure you do your homework after you get home from school’ and he later says ‘Yeah, I was gonna do it next week’ I’d get very angry!” We want our machines to have that kind of common sense (without that kind of childhood sneakiness).
An example of a question that the challenge might ask is:
Question: The beach chair will not fit in the trunk because it was too big. What was too big?
Answer A: The beach chair
Answer B: The trunk
For humans, this answer is obvious: Answer A must be true because the chair must be too big if it doesn’t fit in the trunk. Most AI still currently fails to answer this, however. According to the developers, a computer that is truly as smart as a human (or even close to it) must be able to answer questions such as these.
The Internet of Things will be a major part the advances. “I should be able to hand you a new toaster or a TV or something else, and there’s a way in which a natural language system can communicate with it and find out what it’s capable of,” Ortiz explains. “Right now the only way we can do that is to build a specialized language interface for each device, but you want your interaction between devices to be seamless. As I move around in my world and interact with different devices, my commitments and my goals should follow around with me; I want some continuity in the interaction.”
“As the barrier comes down,” he adds, “it will be more like Star Trek.”
Now you can’t ask for more than that, can you?