You can’t have an #AccessibilityFailFriday series without talking about AI-generated captions. They are not a substitute for accurate transcription. One glaring example? A video where “because she had a cochlear implant” was captioned as “because she had a cocaine implant.” 
This is why AI-based captioning is sometimes referred to as autocraptions. This type of error isn’t just embarrassing. It’s harmful, confusing, and ultimately undermines the goal of accessibility. Why do AI captions fall short?
This is why AI-based captioning is sometimes referred to as autocraptions. This type of error isn’t just embarrassing. It’s harmful, confusing, and ultimately undermines the goal of accessibility. Why do AI captions fall short?1. AI struggles with specialized language, context, dysfluent speech, and non-native accents. AI-based captions can only work within the data that has been provided. Biased data guarantees inaccurate results. AI-based captions often substitute completely unrelated terms when dealing with medical, technical, or conversational phrases. Effectively, it’s guessing. Mistakes like “cocaine implant” instead of “cochlear implant” are unfortunately common.
2. AI-based captions rarely get reviewed by a human, so errors go uncorrected. This lack of oversight means these tools can turn sensitive topics into jokes or completely misinform viewers, causing major communication breakdowns.
3. Consistently poor captions erode trust. People who rely on captions, such as those who are d/Deaf or have hearing loss, often get frustrated and may abandon entire platforms or content creators due to these frequent inaccuracies.
Don’t publish AI-based captions without human review. A hybrid approach with skilled human captioners can ensure accuracy and context, correcting errors before they become misinformation. Investing in quality and context-specific training data is crucial for future AI captioning improvements.
Until these issues are fixed, relying solely on AI for captioning does not meet the minimum required to call yourself inclusive and accessible. It’s performative inclusion at best.
Alt: Screenshot from video of person lying in bed with second person, pulling back their hair to show a cochlear implant. Alt text for this scene says “because she has a cocaine implant.”