AI Technology can be highly effective for the proper representation of the outliers in society. However, the data sets must be refined for the AI-systems, which will require active manual effort in the direction.
FREMONT, CA: Artificial Intelligence (AI)-based systems are swiftly replacing the traditional systems as widespread operational transformations accompany them. They also streamline the existing processes and provide deep insights through their ability to access and process massive data sets. However, the companies are also becoming aware of the challenges that come along with the deployment of AI. Many AI developers are well-versed with the need to treat underrepresented groups fairly, especially for gender and race. However, disability is an essential aspect of diversity that has been neglected. According to the World Health Organization, 15 percent of people across the globe have some form of impairment that can result in disability.
Humans experience physical, sensory, or cognitive disability at some stage of their lives. Whether temporary or permanent, it is a challenging part of our experience that technology can address. Here are the significant challenges that the current governments and industrial data teams are dealing with:
Disability is a diverse concept with umpteen numbers of possible variations. As per the United Nations Convention, disability refers to the individuals with impairments, attitudinal, and environmental barriers that limit their participation in society. The challenges can range from physical to sensory to communication barriers. As it comprises a wide range of possibilities, an individual can have a completely different set of challenges as another. For instance, the issue faced by a visually impaired person may require a completely different arrangement than a blind person navigating a city. Disability can lead an individual to do things in a different way, which further alienates them from the masses. It means that data representing a person with a disability can look unique. Thus maintaining fairness by developing balanced training data sets gets difficult. Disabled people may not fit as per the patterns generated by machine learning (ML).
Another common challenge is the privacy concerns over sharing disability information. Disable candidates are judged purely based on their skill sets and the ability to do the job. The trend is supported by the Americans with Disabilities Act that prohibits employers from delving into their candidate’s disability status. Such policies of fairness through unawareness can hurt the representation of the outliers. Disables consider it risky to expose their disabilities as they are not sure how they might be affected henceforth. Though they have different opinions when it comes to providing information regarding their disabilities. Some consider it a hindrance which might jeopardize their job prospect while others may be willing to share his disability status if he feels that it might enhance his chances for a job role. The problem arises when an agency or a department tries to access data for the disables. The conventional systems are unable to access the correct data set, as several cases are yet to be reported. A major risk that arises here is that the outliers may not be appropriately represented, even by the latest AI-based systems, as the data sets used by the technology also rely on learned statistical norms.
It is essential to find ways to include the disability factor in the AI-based systems accurately. Developers must spend time considering who the outliners can be and who might be affected by the solutions they are developing. For instance, a voice-controlled system might consider people whose voices are not as speech impaired or an online test might not be the best judge of disability. The best way is to search for paths to seek out the affected people and work toward a fair system that presents them with adequate opportunity. Here are some of the ways in which the above issue can be addressed:
• Exploring critical questions and cases
It is essential to bring together individuals with disabilities. Once the outliers and their caretakers are convinced over the advantages of data sharing, a higher level of accuracy can be attained.
• Incorporating Measures for bias detection
As mentioned earlier, various people may respond differently when it comes to sharing their disability information for official purposes. The key is to identify the reasons and include the factors that consider the outliers based on patterns and the probable causes to do so. Such proactive measures can benefit not only the disabled but also help society as a whole.