How cognitive diversity in AI can help close the disability inclusion gap

Unemployment among people with disabilities is as high as 80% in some countries, making them among the most marginalized groups when it comes to work.

Those who are employed often face unequal hiring and promotion standards, unequal pay for equal work, and occupational segregation, which makes it difficult to close the disability inclusion gap.

Indeed, the cost of excluding people with disabilities represents up to 7% of the gross domestic product in some countries. A disability-inclusive business strategy, by contrast, could lead to 28% higher revenue and 30% higher profit margins.

Accessibility policy and solutions have received more public attention recently, through legislation such as the EU’s Strategy for the Rights of Persons with Disabilities 2021-2030 and the Resolution on the EU Artificial Intelligence Act for the inclusion of persons with disabilities; and special directives and frameworks like Unicef’s Accessible and Inclusive Digital Solutions for Girls with Disabilities.

However, cognitive diversity – also frequently referred as neurodiversity – mental impairments and disabilities are still characterized by both social exclusion and the lack of public frameworks.

This is important to address as the unemployment rate among those with autism may reach 85%, dependent on the country; while among people with severe mental health disorders, it can be between 68%-83%, and for those with Down’s syndrome, 43%.

These disorders represent some of the most socially affected and one of the biggest populations. In the UK, for example, at least one in six people live with one or more neurological conditions, and one in seven are neurodivergent. At the same time, in the US, one in four adults has a diagnosable mental disorder in a given year.

Such individuals are often falsely stopped by the police, discriminated against during job interviews and/or left socially isolated within academia and education systems.

Technological advances such as emotional and conversational artificial intelligence (AI), assistive robotics and social companions, and specialized hiring and learning platforms and tools can address some of these challenges by creating more accessible workplaces, hiring and learning experiences, and accommodation practices.

Companies like Robokind and LuxAI use social robotics for emotional training for pupils with autism, while Brainpower is a wearable that helps neurodiverse individuals with social-emotional learning. Beme.AI empowers people with autism to thrive through well-being and development tracking and analytics.

Assistive technology like that provided by Eyejustread can support those with dyslexia and attention deficit and hyperactivity disorder. Meanwhile, Ultranauts provides onshore quality engineering and assurance services through a fully-remote workplace, where 75% of staff are neurodivergent.

Other use cases include using conversational AI to support mental health and anxiety, solutions enhancing or augmenting sensory and/or visual impairment, and technologies making urban and city planning more accessible.

However, existing social bias, historical exclusion and insufficient research and data sets pose a challenge for the successful development and implementation of cognitive diversity policies.

Cognitive diversity and disabilities are not monolithic but present a complex spectrum of characteristics, making them comprehensive subjects in research, data, interfaces and involved stakeholders.

Research and development in this area, therefore, requires an understanding of aspects such as comorbidity, underlying physical and mental conditions, as well as influences of intersectionality, gender and socioeconomic criteria.

For instance, between 25% and 40% of people with learning disabilities also experience mental health problems. They are also 1.6 times more likely to have allergies and other conditions.

In addition, girls are diagnosed at a substantially lower rate or misdiagnosed due to the different manifesting criteria and a historical lack of data. Very few urban data sets include data on gender, so it is hard to develop infrastructure programmes that factor in women’s needs.

Particular ethnic and social groups have been historically excluded from research. A Georgia State University study reported that Caucasian parents of autistic children were 2.61 times more likely to report any social concerns to their child’s paediatrician than African-American parents.

This complex spectrum of criterion comes along with other challenges associated with research, development and adoption of assistive technologies and policies. Issues include:

● Existing policy frameworks lack specifications and specific cases related to sensory and neurodiversity.

● Existing audit frameworks for facial recognition and similar systems do not sufficiently target disability-related biases such as facial impairment and different gestures, gesticulation and communication styles.

● Researchers and policymakers lack access to the community’s data.

● There is still limited access to assistive technology, currently at about a 10% uptake, making it more difficult for further research.

● The ecosystem of assistive technology is still fragmented and not sufficiently connected.

● Existing unconscious and conscious social bias, lack of representation and accessible vocabulary.

These challenges present a long-term call to action for technologists, researchers, and policymakers to gradually work together to address the following aspects:

Assistive solutions for neurodiverse individuals or people with cognitive disabilities may target such criteria as attention, memory, communication, learning, executive functions and performing, visual and tactile experiences, emotional status and empathy.

These aspects are accompanied by age, gender and underlying conditions, enabling the identification of an appropriate research framework, as well as requirements for the technology’s interface, data input and labelling.

To better address particular impairments, solutions designed to better support cognitive diversity tend to become more specialized and interconnected, when a few apps, tools or devices are used simultaneously as a part of a supporting ecosystem.

Multiple stakeholders may also be involved in data input, including the individuals themselves, families, caregivers, counsellors and educators.

For example, Beme.AI builds solutions for autistic children that enable tracking of general wellbeing, mood, nutrition and other factors. It also enables the connection of external tracking devices, as well as data input by both child and parent.

About one in eight (79%) of neurodivergent individuals feel socially isolated and assistive solutions can empower social integration and communication, but not replace it.

Technologies such as social robotics or adaptive learning platforms are typically developed along with a curriculum that identifies aspects of interaction and learning for the child, but also includes the involvement of caregivers and educators.

These curriculums may become an even more critical part than the technology itself in order to avoid further exclusion.

Just as AI systems can discriminate against people of a particular race, systems such as computer vision or facial recognition can discriminate against individuals with disabilities, facial differences such as neuropathy and different communication styles.

This leads to the necessity of “disability-centred” auditing approaches at all levels of the problem definition, data sets, algorithms and systems.

The audit should involve the representation of divergent individuals in the research or resource groups. These groups aim to evaluate criteria such as elimination of discriminatory influences, feedback between users and researchers, aspects of transparency and explainability, accountability, and actions and non-actions performed by the system during certain scenarios and interactions.

The World Health Organization and national health service providers regularly work on developing frameworks that address the convergence of technology, medical and social skills.

For the area of neurodiversity and cognitive disability, these frameworks may need to be accompanied by neuro-accessible design, area-specific human capacity and social studies, as well as technical skills addressing specific devices and technologies.

Other areas that could contribute to more accessible algorithms and ecosystems include open diversity data platforms, accessibility statistics provided by employers, and policies and guidelines addressing aspects of intersectionality, gender and age groups.

It’s important to remember that developing policies to support cognitive diversity can impact, and indeed benefit, both an individuals’ general well-being and the wider economy.

These should include technologies for age-associated cognitive impairment, areas of mental health and neurological disorders, smart cities, adaptive learning and workplace technologies.

Because by improving inclusion for people with disabilities through assistive and AI-based technologies to live active and fulfilling lives, we can help build a better society for everyone.