En Accessibility and Artificial Intelligence comentas algunas aplicaciones de la inteligencia artificial para la accesibilidad.
Básicamente, las dos situaciones que comenta son de sobra conocida: la generación automática de subtítulos en los vídeos y la generación automática de texto alternativo en las imágenes.
En Five accessibility bugs GitHub Copilot offers to create explican los peligros de confiar en la inteligencia artificial en cuanto a la accesibilidad de una página web:
I like GitHub Copilot, at least for the narrow range of purposes I mentioned earlier. At the time I'm writing this up, there have been announcements about the next version of Copilot. Perhaps the next version will be fix some of these issues.
The danger here is that developers accept code suggestions, assuming that they're good. The 'wisdom of the crowd' could suggest that code based on millions of lines of code won't contain bugs. As demonstrated, this is not true.
Filtering the output, as it currently does to remove "offensive output", is possible. That probably wouldn't work better than current automatic accessibility testing and linting. So it would be an improvement, but not an absolute fix.
In my opinion, the responsibility always rests with the developer using the tool.
You shouldn't accept code suggestions from GitHub Copilot if you don't understand them. If you're expecting a certain type of suggestion, and you get one with extra attributes, you need to look them up. Don't use the code until you understand what every part of it does.
This could have a positive side. Maybe. It's possible that Copilot suggests accessibility considerations that people would otherwise have missed. Making people consider how to incorporate accessibility into their work normalizes it.
Adrian Roselli tampoco confía mucho en la inteligencia artificial para resolver los problemas de accesibilidad. En No, ‘AI’ Will Not Fix Accessibility explica:
Large language models are habitual liars. Meanwhile, automated image descriptions aren’t much better. To give them the benefit of the doubt, perhaps these tools simply lack context.
As image tools get better at describing every detail of a picture, as language models do a better job of conveying an emoji-laden tweet in actual words, they are still not the authors of that content. They have no sense of why it was created. They cannot tell you that a series of vertical lines is meant to signify a wall in a meme.
Según Eric Eggert, los accessibility overlay basados en inteligencia artificial no cumplen lo que prometen. Lo cuenta en su vídeo 5 False Claims 1-Line “AI” Accessibility Script Vendors Make:
Y el inicio de la transcripción:
They claim to be the be-all and end-all of all accessibility worries: “AI” accessibility tools. They say they provide “100% WCAG 2.1 compliance”. They say it only takes one line of JavaScript to “fix your website for accessibility”.
Who are “they”? VC-backed companies that promise affordable, effortless compliance for your website. If you think “that sounds too good to be true”, you’re right. It is indeed quite the opposite of true.
Here are five claims that “AI” accessibility script vendors make that are misleading or completely untrue.
(Intro.)
I’m Eric Eggert, a Web Accessibility Expert. You can follow me as @yatil — Y-A-T-I-L — on social media. If you have questions about this or other videos, or want to propose a topic, ask in the comments, or use the hashtag #askYatil on Twitter.
Before we get into it, whenever I say AI, please imagine the most expressive air-quotes gesture possible.
Okay!
Automated accessibility testing tools, powered by artificial intelligence (AI) and code-scanning technology, can be an important part of an accessibility testing plan — but, they must remain just a part. Automated testing on its own should not replace manual testing by an accessibility expert and should not be used as the only way to show compliance.Y avisa del peligro del texto alternativo generado de forma automática mediante inteligencia artificial:
As technology advances, there are even image-recognition programs that can with some accuracy guess what an image depicts and generate alt text. On one hand, it's incredible that this technology exists. On the other, it presents an additional layer of review required by human beings. The real purpose of an image in the context of a web page, the real feeling somebody is supposed to be exposed to, the real information conveyed by an image can't be decided by a machine — at least not yet.
El grupo de investigación SoftLab de la Universidad Carlos III de Madrid (UC3M) y el Grupo Lavinia estrechan su colaboración para conseguir una televisión más accesible a todas las personas que presentan algún tipo de discapacidad visual y/o auditiva.Se han desarrollado tres sistemas:
El Reconocimiento Automático del Habla (RAH) o Reconocimiento Automático de voz es una parte de la Inteligencia Artificial que tiene como objetivo permitir la comunicación hablada entre seres humanos y computadoras electrónicas. El problema que se plantea en un sistema de RAH es el de hacer cooperar un conjunto de informaciones que provienen de diversas fuentes de conocimiento (acústica, fonética, fonológica, léxica, sintáctica, semántica y pragmática), en presencia de ambigüedades, incertidumbres y errores inevitables para llegar a obtener una interpretación aceptable del mensaje acústico recibido.
