[ad_1]
Consider, for example, a computer vision system that recognizes objects and can describe what is, for example, on a table. Most likely, the algorithm was trained with data collected by capable people, from their point of view, probably standing.
A person in a wheelchair looking to do the same might find that the system is not as effective from that lower angle. Likewise, a blind person will not know to hold the camera in the correct position long enough for the algorithm to do its job, so it must be done by trial and error.
Or consider a facial recognition algorithm that should indicate when you are paying attention to the screen for one metric or another. What is the probability that among the faces used to train that system, a significant number have things like a fan, or a blow and blow controller or a head strap that obscures part of it? These “confounders” can significantly affect accuracy if the system has never seen anything like it.
Apple’s head of global accessibility on the company’s new features for iOS 13 and macOS Catalina
Facial recognition software that fails in dark-skinned people, or is less accurate in women, is a common example of this type of “garbage in, garbage out.” Less often, but no less important, is the visual representation of people with disabilities or their point of view.
Microsoft today announced a handful of efforts co-led by advocacy organizations hoping to do something about this “data desert” that limits the inclusion of AI.
The first is a collaboration with Team Gleason, an organization formed to raise awareness of the degenerative neuromotor disease amyotrophic lateral sclerosis, or ALS (named after former NFL star Steve Gleason, who was diagnosed with the disease a few years ago ).
Your concern is the above regarding facial recognition. People living with ALS have a wide variety of symptoms and assistive technologies, which can interfere with algorithms they have never seen before. That becomes a problem if, for example, a company wants to ship gaze-tracking software that relies on facial recognition, as Microsoft would surely like to do.
“Computer vision and machine learning don’t represent the use cases and the looks of people with ALS and other conditions,” said Blair Casey of the Gleason team. “Everyone’s situation is different and the way they use technology is different. People find the most creative ways to be efficient and comfortable.”
As ADA turns 30, technology is just beginning to help people with disabilities
Project Insight is the name of a new joint effort with Microsoft that will collect face images from volunteer users with ALS as they go about their business. Over time, the facial data will be integrated with Microsoft’s existing cognitive services, but it will also be freely published so that others can improve their own algorithms with it.
Their goal is to have a launch by the end of 2021. If the time frame seems a bit long, Microsoft’s Mary Bellard of the company’s AI effort for accessibility noted that they are basically starting from scratch and that getting it right is important. .
“Research leads to insights, insights lead to models that engineers put into products. But we have to have data to be accurate enough to be in a product in the first place,” he said. “The data will be shared; it sure is not about improving any product, it is about accelerating research around these complex opportunities. And that is work we don’t want to do alone.”
Another opportunity for improvement is to obtain images of users who do not use an application in the same way as most. Like the person with vision or wheelchair problems mentioned above, there is a lack of data from their perspective. There are two efforts aimed at addressing this problem.
Images taken by people who need to identify or locate objects in them.
One with the City University of London is the expansion and eventual public launch of the Object Recognition for Blind Image Training project, which is putting together a data set for every day to identify everyday objects (a soda can, a keychain) using the camera from a smart phone. However, unlike other datasets, these will be sourced entirely from blind users, which means that the algorithm will learn early on how to work with the type of data that will be given to it later anyway.
The other is an expansion of VizWiz to better cover this type of data. The tool is used by people who need immediate help to know, for example, if a cup of yogurt is expired or if there is a car in the driveway. Microsoft worked with the app’s creator, Danna Gurari, to enhance the app’s existing database of tens of thousands of images with associated questions and captions. They are also working to alert a user when their image is too dark or blurry to analyze or send.
Inclusiveness is complex because it involves people and systems that, perhaps without even realizing it, define “normal” and then fail to function outside of those norms. If AI is to be inclusive, “normal” needs to be redefined and that will take a lot of work. Until recently, people didn’t even talk about it. But that is changing.
“This is something the ALS community wanted for years,” Casey said. “This is technology that exists, it’s on a shelf. We’re going to put it to use. When we talk about it, people will do more, and that’s something the community needs as a whole.”
Here’s the curtain on Sight Tech’s global agenda
