Modelling for disability: How does artificial intelligence affect unemployment among people with disability? An empirical analysis of linear and nonlinear effects.
AI adoption cuts unemployment for adults with disabilities only after it passes a popularity threshold, with weaker gains for women and non-rich countries.
01Research in Context
What this study did
Abid et al. (2024) asked a big question: does more AI use help adults with disabilities find jobs?
They looked at countries around the world and tracked both early and heavy AI use.
The team ran two kinds of stats to see if the link was straight or curved.
What they found
Jobless rates dropped only after AI interest passed a tipping point.
Women and people in lower-income countries saw smaller gains or none at all.
How this fits with other research
Hedley et al. (2017) and Vazquez et al. (2019) show most autism hiring programs fix the person, not the workplace. Mehdi’s work says AI can fix the labor market instead.
Nitzan et al. (2026) extend the finding to autistic adults. They warn that biased AI tools could undo the gains unless behavior analysts help design them.
Agiovlasitis et al. (2025) sketch an AI app that bundles job-matching chatbots with health care in Saudi Arabia. Their plan lines up with Mehdi’s threshold idea: small AI use is not enough; you need a full bundle to move the needle.
Why it matters
You can stop pushing only social skills. Lobby employers to adopt inclusive AI once interest is high enough in your region. Track the data by gender and country income so you know when the tipping point is near.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Add a line in your next employer meeting: “What’s your AI adoption rate, and can we test it past the threshold?”
02At a glance
03Original abstract
There is a growing debate among scholars regarding the impact of artificial intelligence (AI) on the employment opportunities and professional development of people with disability. Although there has been an increasing body of empirical research on the topic, it has generally yielded conflicting findings. This study contributes to the ongoing debate by examining the linear and nonlinear effects of AI on the unemployment of people with disability in 40 countries between 2007 and 2021. Using the system Generalized Methods of Moments and panel smooth transition regression, the main conclusions are as follows. First, AI reduces the unemployment of people with disability in the full sample. Second, upon disaggregating the sample based on income level (high income/non-high income) and gender (men/women), the linear model only detects an inverse correlation between AI and unemployment among people with disability in high-income countries and among men, whereas it does not influence unemployment in non-high-income countries and women. Third, the panel smooth transition regression model suggests that the effects of AI on the unemployment of people with disability and among women are only observed once artificial intelligence interest search exceeds a specific threshold level. The effects of AI in non-high-income economies and among women are not significant in the lower regime, which confirms the nonlinear association between AI and the unemployment rate of people with disability. These findings have important policy implications for facilitating the integration of people with disability into the labor market.
Research in developmental disabilities, 2024 · doi:10.1016/j.ridd.2024.104732