AI is getting smarter
That's the word from artificial intelligence researchers and industry analysts attending the AAAI-15 conference last week in Austin.
"I think big leaps have been made in the last few years," said Geoffrey Hinton, a distinguished researcher at Google and professor at the University of Toronto. "AI is undergoing a growth spurt. We're beginning to solve problems that a few years ago we couldn't solve, like recognizing images."
Artificial intelligence, also known as AI, will be significantly more advanced in another five years, said Hinton, who is known for his work in machine learning and artificial neural networks, which are learning algorithms inspired by animals' central nervous systems, particularly the brain.
That can be a tough idea to sell in an industry that has seen scientists and observers waver drastically over the past 30 or 40 years between times of great optimism and equal amounts of pessimism about the potential of AI.
Back in the 1980s, for instance, there was heavy attendance at AI-focused conferences, but there was also little science being done to support all the buzz.
"People just didn't realize how hard this was, so even in the '60s we thought we'd soon have human-like robots that have all these human-like skills," said Lynne Parker, a professor at the University of Tennessee and a division director in Information and Intelligent Systems with the National Science Foundation. "Then we thought we were on a false trail."
Some of the old predictions haven't come to pass. We don't have robotic servants folding our laundry and taking care of our kids or elderly parents. We don't have robotic airplanes flying us to business meetings, or mobile phones that connect with our offices, homes and cars.
"I think repeatedly we've not met the estimates that we keep making about where we'd be in the future," said Sonia Chernova, an assistant professor of computer science and director of the Robot Autonomy and Interactive Learning lab at Worcester Polytechnic Institute. "Reasoning is just really hard and dealing with the real world is very hard... But we've made amazing gains."
But AI research is catching up to the hype that's surrounded it for so long.
Today, AI is on an upswing fueled by academic research labs at institutions like Carnegie Mellon University, WPI and MIT, as well as in industry, where tech companies including Google and Microsoft are throwing their financial and intellectual might behind AI efforts.
"Right now, we're in more optimistic times," said Parker. "There have been a lot of advances in robotics, with [IBM's] Watson and natural language processing and speech recognition with technologies like Apple's Siri."
So what advances are just ahead of us
Major gains are being made, or are about to be made, in natural language processing, speech recognition, object recognition, computer vision, machine translation and neural networks.
Many of those technologies will be used to build robots that move more fluidly, like humans. They also will help scientists integrate multiple capabilities into one robotic system.
"We've made a lot of deep advances in many focused areas, but we need one big system to pull a lot of these systems together into one machine," Parker said. "To have a household robot that can obey your commands, we're still pretty far from that. I would say 10 to 20 years. It's not about the glue. When you build one subsystem, it affects how another subsystem should be designed. You can't build them in isolation and just glue them together. It has to be holistic."
Different areas of AI research also will come together to support the creation of the smart home or the Internet of Things.
"People expect a lot from this futuristic system," Chernova said. "They want their system to predict what they want. They get frustrated if they have to tell it something more than once. AI falls into that with modeling and predicting the behaviors of people."
She added that artificial intelligence is ingrained into IoT and should be able to take the industry within about 10 years to a level where people are interested in using it.
Google's Hinton said he's most excited about gains in neural networks that would enable computers to understand the content of sentences and documents.
"That is close to the core of Google because it involves understanding sentences, and if you can understand what a document is saying, you can do a much better search," Hinton said. "That's a core AI problem. Can you read a document and know what it's saying It could work in the legal field where you're looking for precedents. You can't read all the cases there ever were, but computers can."
AI has probably received the most attention in the past few years because of Google's work with autonomous cars.
Most agree that self-driving cars will advance significantly in the next several years, and Google expects to have them on the road by 2020. What will likely slow that progress, according to Parker, will be the legal issues that surround autonomous cars.
"Who will be to blame if the car makes a mistake" Parker asked. "From a technical perspective, Google has been able to have cars drive thousands of miles in a restricted area. You take that car and put it in the middle of Maine in the middle of a blizzard, it probably won't work. Maybe we'll see them working as a lift service on a large campus, that's much more restrained because you're on a single, known campus. That sort of thing is much more likely soon."
Hinton said there is little standing in the way of great advances in AI, but he wishes there were more people working on new ideas to push it even further along.
"A robot that can walk around your house and open doors and go upstairs Well, that might be unlikely, but I don't think it's out of the question," he said. "Good ideas. Good data and fast computers. If you have those, you can do almost anything."