'We don't want to leave people behind': AI is helping people with disabilities in surprising new ways

AI News

new york

When Matthew Sherwood goes clothes shopping, he needs help to make sure what he picks is the color and style he's looking for.

Mr Sherwood has been blind for more than 15 years but has a family, a successful investment career and a dog, Chris, who helps him get around in the world – but he says everyday tasks like shopping still pose obstacles to independence.

Artificial intelligence may soon be able to help.

Currently, Sherwood says he occasionally uses an app called Be My Eyes, which pairs blind users with sighted volunteers who can help via live video check things like whether a shirt matches the rest of their clothes or if a carton of milk has expired, but advances in AI technology are already making volunteer helpers unnecessary.

Be My Eyes partnered with OpenAI last year to enable an AI model, rather than another human, to see and describe what's in front of a user. OpenAI's latest product demo featured footage of a person using an AI-powered version of Be My Eyes to hail a cab. The app told the user exactly when to raise their arm when a car was coming. Google announced in May that it was adding a similar feature to Lookout, an app designed to help users with visual impairments.

Applications for the visually impaired are just one area where AI is contributing to advances in “assistive technology,” tools designed to help the physically disabled and elderly.

Apple, Google and other tech companies are rolling out a growing number of AI-powered tools aimed at making life easier for people with a variety of disabilities, from eye-tracking tools that allow physically impaired users to control an iPhone with their eyes to detailed voice directions for blind Google Maps users.

Since the explosive release of ChatGPT over a year ago, it's clear that AI will change the world by upending the way we work, communicate, and even what we perceive as reality. But for people with disabilities, AI has the potential to change their lives in an entirely different way.

“It used to be that if you were blind, you had to have a clerk read it to you if you wanted to work in a business,” Sherwood says, “but now you have this new power, and for some it's a fantastic technology. For people who are blind, it's an opportunity to get employment, an opportunity to compete for business, an opportunity to succeed.”

For years, tech companies have been using early forms of AI to make their products more accessible, such as automatic captioning in videos and screen readers.

But experts say the massive datasets and powerful computing systems that underpin modern AI models are accelerating what's possible in the field of assistive technology. For example, to reliably help a blind person hail a cab, an AI tool needs to recognize what the cab looks like with great accuracy, which requires training models on a huge corpus of examples.

In another example, Google's tool that communicates to blind and low vision users what they see on their screen has been upgraded with a “Question and Answer” feature that incorporates the company's generative AI technology.

“The potential of AI has been clear for years, but it needs to reach this level of quality before it's useful enough to be built into products,” Eve Anderson, Google's senior director of product inclusion, equity and accessibility, told CNN.

New generative AI tools are particularly promising for accessibility applications because they are designed to understand and generate information in a variety of formats, including text, audio, photos, and video. This means that when a person needs to consume information in a specific medium, AI can act as an intermediary, converting portions of speech into text for users who are hearing impaired, for example.

“[People's]accessibility needs take many forms, but the majority of disabilities are really about input and output, how a person perceives information,” Anderson said. “There are hearing impairments, visual impairments, motor impairments, speech impairments, cognitive impairments, and all of those can come with the need for different modalities, and one of the things that AI is good at is translating between modalities.”

Continued investment is needed to ensure that AI systems can continue to serve all kinds of users.

Because AI models are trained on human-generated data, experts warn they could reproduce the same biases that exist among humans, and early examples have already emerged, including an AI image generator that appears to struggle to understand the concept of race and an algorithm that allegedly displayed job ads based on gender stereotypes.

In one effort to address this risk, a group of leading tech companies, including Apple, Google, and Microsoft, partnered with researchers at the University of Illinois at Urbana-Champaign to create training datasets for AI speech recognition tools that contain a diverse range of speech patterns. Speech recognition tools, such as translators, voice assistants, and speech-to-text apps, are especially important and useful for users with disabilities.

The effort, called the Speech Accessibility Project, involves collecting recordings from volunteers with Parkinson's disease, Down syndrome, amyotrophic lateral sclerosis, and other disorders that affect speech. With the help of the project's recordings, which now number more than 200,000, the sample speech recognition tool the researchers built misunderstood speech just 12 percent of the time, down from 20 percent before it was trained on the new dataset.

“The more diverse the types of speech and severity of symptoms we can feed into machine learning systems, the better these systems will be at understanding people with 'audiobook narrator'-like speech impairs,” says Clarion Mendez, a speech-language-hearing pathologist and clinical assistant professor who is leading the project.

“Through this project, we've spoken to a lot of people who face huge barriers to participating in life because of communication, people who have great degrees but can't get a job because of communication barriers,” Mendes said. “When something like assistive technology helps people find fulfillment in their hobbies or their work, suddenly they have a lot more independence in activities that once took a lot of time or required them to rely on others.”

Anderson added that investing in AI for accessibility is not only the right thing to do, it also makes good business sense.

“We don't want to leave people behind, and technology in general has the power to level the playing field,” Anderson said, “but there are also economic reasons, like being able to sell to government agencies and educational institutions.”

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *