Microsoft and other companies continue to confront challenges as they work towards achieving the “key aspirations” of artificial intelligence (AI) and using the technology to augment and extend human capabilities, according to Eric Horvitz, Microsoft Research managing director and technical fellow.
“There’s so much upside to basically getting this right,” he told the Microsoft Research Faculty Summit: The Edge of AI in Redmond, Wash., July 17.
Horvitz identified three key aspirations for AI development as the ability to attain more general intelligence, mastering human-AI collaboration, and leading with insights on AI, people and society.
Among the many sectors in which AI technology has already started being used is the car industry, in automated vehicles, he noted. The designers of automated cars face multiple challenges that include “all the edge conditions” that could conceivably come up quickly in “real world” conditions, he said.
For example, while testing out the Autopilot feature on a Tesla car that he purchased, he saw that “the system knows when it’s not confident enough to continue,” he said. But he added that after having his family join him in the test: “It was looking very reliable — until the car turned into oncoming traffic without any sign or any signaling at all.” His wife made him promise to “never put this thing in Autopilot ever again,” he said.
Another incident using the car made him guess that the Tesla designers “weren’t thinking about the rear ends of deer,” he said, adding that’s an example of the challenges presented by “unknown unknowns.”
There are, in fact, technical and policy challenges when it comes to AI, he said, telling the summit that ways to address challenges presented by automated cars include the expansion of real-world testing.
He went on to say: “We’re in a very data-rich world right now … but we still often face challenges with data scarcity” when trying to personalize data for individual users. That creates “opportunities” for AI, he said. But he said there are potential concerns raised by machine biases that could make such data flawed.
In announcing the two-day summit, Microsoft said its AI researchers were “striving to create intelligent machines that complement human reasoning, and amplify human ingenuity with intelligent technology.” It added: “At the core, is the ability to harness the explosion of digital data and computational power with advanced algorithms that extend the ability for machines to learn, reason, sense and understand, enabling collaborative and natural interactions between machines and humans.”