For the interested reader, her full research along with transcripts of her conversations with GPT for this experiment for full transparency is available here.
Hema Thakur
In the conferences she conducts as a trainer for academic editors and researchers at centers like Ashoka University among others, Hema Thakur comes across a common query: “Will AI replace us?” Initially, she’d try explaining how she foresaw AI as an assistant rather than a replacement. But challenging apprehensions requires proof more than theories. So, Hema began to dedicate herself to some experiments with well-established AI models to prove the worth of human intervention.
Given that Title II of the Americans with Disabilities Act is expected to go into effect on April 24, 2026, she thought it pertinent to examine AI’s capabilities to address an important gap: simplifying technical feedback (the kind that authors of theses or journal articles receive from mentors or peer reviewers) for people with cognitive disabilities such as dyslexia and ADHD.
What she did next was simple: She picked 10 technical comments about a finance paper and asked ChatGPT to simplify the feedback via two different prompts: one specified that the person requiring this assistance has cognitive disabilities and the other just requested for a simplification in plain writing without mentioning anything about disabilities. She tried each combination for each technical comment twice, to also check for consistency.
For the interested reader, her full research along with transcripts of her conversations with GPT for this experiment - for full transparency - is available here. It has also been featured on various other platforms like Sage’s Social Science Space for the research community in general, the European Association of Science Editors’ blog because it carried relevance to editorial intervention as well, and Disability Horizons, for the community it catered to.
Turns out it consistently did a bad job - it often distorted or oversimplified the meaning. For example, it changed terms like “bounded rationality” to “limited thinking ability” - a condescending and incorrect definition.
But why should this matter to us? There are in fact several takeaways for the business community. In India alone, statistics indicate there are 30 to 40 million people with disabilities (likely underreported), making user-friendly products for this group not only “nice to have” but also core to retention and reach. The noteworthy performance of many accessibility-forward companies adopting ESG frameworks backs this.
Also, remember when we introduced closed captions for people with hearing disabilities? It benefited not only the community but also people who wanted to binge watch their favorite series on the way to work or simply multitask, or who weren’t native speakers of the language of certain series. Identify edge cases, and you make the product more valuable for ALL users - case in point, if we have AI products that do a better job at simplifying feedback, we are not only assisting people who have been diagnosed with disabilities but also those who may be on the spectrum, those who tend to draw a blank when faced with submission deadlines and could use some help, or those who simply want assistance in understanding technical feedback.
Also, when we cater to a more diverse market, our dataset for training models needs to be just as diverse, which means we have scope for making these more robust, and robustness is critical for tech companies, big and small.
About the concern of being replaced expressed by Hema’s conference attendees, she says, “For now, we can see clearly why our intervention is required, and for the future, as AI models refine themselves and become even more widespread, whom do you think will need to be the first line of defence against misapplications?” She adds, “remember, even a refined model isn’t a perfect one; besides, why waste time doing things an assistant can help you with when you can instead spend it on subtler aspects that require your skillset - curating research narratives with empathy being the first but not the only example of this?”
Subscribe today by clicking the link and stay updated with the latest news!" Click here!
