Human Insights & Intelligence Speed Path to Better AI in Aira

Aira.IO, based in San Diego and launched in 2014, focuses on developing technology and services that provide blind and low-vision individuals greater mobility and independence. Using wearable smart glass technology, Aira connects the visually impaired in real time (through AWS components) with remote, trained Aira Agents. Agents assist users to perform a myriad of daily activities — ranging from navigating city streets and catching a plane to ordering meals from a restaurant menu.

By Suman Kanuganti

Although still largely in the developmental stage, artificial intelligence, or AI, is no longer science fiction. Researchers have placed this emerging frontier on a fast track to produce technology that is destined to dramatically change the way we interact with our world and solve problems. In the realm of personal assistant technology — such as Apple Siri, Cortana and Google — a fact is evident: the human element is important when building artificially intelligent machines that behave like people!

At Aira, this human touch is critical for a personalized AI-driven system we are developing to assist blind and visually impaired users in a wide range of daily tasks and activities. Currently, assistive services at Aira are provided through our network of carefully selected, well-trained and certified remote human Agents.

To accomplish the ambitious goal of converting our current high-tech system to a highly functional AI system, we are focusing on the use of human contextual cues to overcome key project hurdles. For example, how do you train a machine to: Instruct a blind person to get safely on an elevator? Or to locate an empty seat at a restaurant or on a bus? Or to gauge how many feet to walk to the next block? Select a desired item from a grocery store shelf? Recognize a face in the crowd, or read a menu for a sightless individual?

We are finding that the answers to such questions lie in using Agents´ human skills to train Machine Learning algorithms how to aid the visually impaired more efficiently. Thus, the addition of more contextual human cues to our dashboard platform tends to enable the blind to be better connected and immersed in the reality of their assistance from the Aira Agents. Aira leverages existing Machine Learning systems, such as Clarifai, and tags interesting cues for metadata.

Aira’s approach to training its AI system found recent validation with the release of Facebook M, the personal digital assistant service inside of Messenger. Facebook M completes tasks and finds information on the behalf of the user in a wide range of areas — from booking a restaurant reservation to finding great deals on the Internet. It is also powered by artificial intelligence that’s trained and supervised by humans. This distinguishes M from other digital assistant services, which are focused primarily on providing information, but may not be accurate if the system is stumped or stymied in responding to tasks.

As Aira´s AI system awaits full commercial implementation, our approach will increase our existing user-to-agent ratio while enhancing economies of scale. By leveraging Agents in the initial stages, Aira will go to market faster (which is targeted for mid 2016), as it uses human rudimentary skills to develop an adaptive AI platform.

Aira’s AI vision is backed and supported by some notable industry visionaries, including the legendary Ray Kurzweil (the innovator who developed ¨The Law of Accelerating Returns”), and venture capital firms Lux Capital and ARCH Venture Partners, with their expertise in AI and machine learning. As we move forward, our platform will expand beyond the blind market to other assistive services in health care, including dementia and autism.

Aira´s automated AI system is expected to be implemented in increments over the next several years, initially augmenting and later replacing human Agents for standard assistive tasks. Our Agents in turn will be trained and reserved for more complex duties. Aira’s ultimate vision is automated machine assistance that works so well that human Agents will not be required, while providing opportunities for certified Agents who can empower customers that still desire a human touch.