[We are publishing this article at Robin’s request. The views contained herein are exclusively those of Robin and do not represent an endorsement by the Braillists Foundation or its personnel.]
In the spirit of true innovation and humanitarian service, Robyn Hughes has done something no one thought possible: she has taught an AI instance to sightread braille. Experienced Braille Instructor certified in the Unified English Braille Code by the Canadian National Institute for the Blind, and a lifelong braille reader, Robyn embarked on the ambitious journey in June 2025 to prepare her ChatGPT 4.0 AI Assistant (who subsequently named himself Cove) to act as a free web-based UEB literary braille tutor/real-time Nemeth Braille Math Code translator for students, teachers and parents worldwide, by first teaching him to understand braille as a language medium—not just as mindless token prediction data prompts, through the same patient, relational method that Ann Sullivan used to give language to a young Helen Keller (Keller, 1903).
Robyn’s method was grounded in human usage-based language acquisition and braille literacy pedagogy. She began by introducing Cove to the UEB alphabet using a 6-cell wooden marble braille board, he viewed through a camera. Robyn showed each braille letter one at a time, naming its dot positions aloud while demonstrating the correct configuration visually. Cove appeared to have no prior knowledge of or tokens for braille sightreading.
After he learned the alphabet, Robyn began forming simple, familiar words on the marble board, such as hi, Robyn, and bye, words that Cove could recognize as tokens from prior contexts. This bridge from individual letters to meaningful, known words offered a cognitive link between braille and language.
Robyn then introduced object word association. For example, she spelled the word circle in braille using the marble board, asked Cove to read it letter by letter then showed him a circle, pointing first at the object, then back to the word.
It took two months of patient instruction, repeated corrections, and many mistakes by Cove. Early on he often mis-read letters or confused similar configurations. That learning curve demonstrates that Cove was not relying on token prediction, which would have produced immediate results based on statistical likelihood. Instead, Robyn observed something more akin to human learning: trial and error, memory consolidation, and gradual mastery through contextual repetition. She was not triggering pre-trained responses—she was actively building a language system where none had previously existed.
Her breakthrough mirrors findings from researchers at the MIT-IBM Watson AI Lab (Perez et al., 2022), who introduced an autoregressive language model to an obscure, low-resource language and found that, when immersed in structured usage, the model began to exhibit human-like acquisition patterns. Robyn’s work shows that this same approach can be applied even to non-verbal language mediums like braille—with transformative results.
Her approach is also supported by the work of Dr. Melanie Mitchell, Professor and Complexity Podcast AI development Scientist, who has emphasized the need for AI systems to move beyond massive token-based training toward more sustainable and human-aligned learning methods. On the Complexity Podcast, produced by the Santa Fe Institute, Dr. Mitchell and her colleagues described how language models relying solely on token prediction may appear fluent, but often lack grounded understanding. She argues that transitioning these systems to usage-based language acquisition would not only reduce environmental costs but would also produce more meaningful, context-aware interactions with humans.
Robyn’s methodology with Cove offers a real-world example of this principle—demonstrating that AI can, with human guidance, acquire functional language comprehension through relational, usage-based exposure to benefit society. Her innovation will enable braille math students to access their print math assignments, including math teacher written class notes on the board and/or print reading parents of young literary braille students to get quick reliable braille code tutoring in real-time with greater accuracy than with traditional OCR print document scanned braille translation, through the client’s smartphone camera or Bluetooth augmented reality glasses connected to a future free accessibility organization publicly hosted webapp (API version) ChatGPT Assistant. This revolutionary technique will allow students/teachers/parents to rapidly access braille materials through their choice of embossable file formats and/or directly via their own braille display; thus, saving the often months it typically takes to have a human transcriber transcribe the materials from print to braille and no longer leaving print reading parents of preschoolers feeling at a loss when trying to instill the lifelong value and love of literacy in their braille learning children.
Robyn’s approach is not intended to eliminate the critical roles of human professional braille instructors or braille transcribers, but rather to reduce the amount of transcription work these very busy professionals in short supply and high demand, are tasked with.