Artificial Intelligence (AI) and Accessibility

Auto-generated alternative text is already common. We see the option in Microsoft 365 documents.

It is also available in various mobile apps, including wearable technology. It requires a camera to capture the image that needs to be described. This often requires access to the Internet to take advantage of the AI engine's database.

Optical character recognition (OCR) software is already common. We see the option in Adobe Acrobat Pro. It helps us make scanned PDF documents readable. The text becomes real text instead of an image of text. It is becoming more common on mobile devices. You can take a picture of text and the device will OCR the text for you.

An AI engine's success at identifying content in an image depends on alternative text already available on images. This is what is recorded in the AI engine's database. Therefore, its accuracy is dependent on our current usage alternative text and images.

If we do not provide correct alt text on images, the AI will provide poor image descriptions. Additionally, AI does not understand human context for using a particular image. The AI would have to scan the entire message for the context to properly describe the image. It is important to make all the content accessible to the AI for scanning.

Text-to-speech has also been available for a while. Screen readers are commonly known text-to-speech software. AI is helping to improve the voices used. They are sounding more human. To improve accuracy, they must understand the language(s) being used in our content. You can set the language of an electronic document or section of a document, such as a Word document or a PDF document.

A future opportunity is in audio descriptions of visuals seen in video.

Auto-generated captions are already common. We see the option in our meeting and collaboration applications, like Microsoft Teams or Zoom. We also see them in video hosting platforms like YouTube. The auto-generated captions in YouTube appear as text displaying one word at a time.

An AI engine's success at identifying text from speech depends on databases. Per the 2024 study by 3Play Media, the accuracy is still not high enough on its own. Recent databases include multiple languages. That expansion appears to have lowered the accuracy level. This may be in part due to the inability to determine which language is being used at a given time.

To improve accuracy, we must identify the language(s) being used in our content. YouTube and other video hosting solutions allow you to set the language of your recordings.

As mentioned before, speech recognition has been around for some time. Instead of using a mouse or a keyboard, commands are given by talking.

This is commonly seen through AI assistants like Siri, Alexa, and Google Assistant. These are often used to control "smart" devices. You can operate these devices or check on their status hands-free.

Dictation is also possible in word processing environments, chat rooms, or messaging areas.

The accuracy of speech recognition software depends on the database the AI uses. If it can learn from your voice, then the accuracy can improve significantly.

The above solutions describe ways to do hands-free operations. However, there are other solutions that assist everyone. These include:

• auto-correct: the AI guesses at the word you are trying to spell and fixes it for you
• auto-completion: the AI guesses at the word or phrase you want to use and provides it for you

The AI learns from your previous communications or usage of application(s). You may be able to "teach" the AI common responses by storing certain common information about yourself or actions you want to make.

Note: Auto-completion is also available for developers. A developer can ask the AI to generate common code to take a particular action. The AI determines the programming language and code to generate.  This is like how generative AI is used to create images based on a prompt. These can be adjusted by the developer after generating.

Many of the recently trending AI solutions can assist people with cognitive disabilities. Many are useful to people without cognitive disabilities, such as the generative AI solutions for auto-completion.

AI has been used for a while to review grammar and word choice in word processors.

You can now prompt AI to review long documents and provide lists or summaries of the content. This is appearing in social media platforms as well. AI is summarizing the comment sections of posts to reduce reading time.

Course developers can prompt AI to review instructional content and display it in a pre-defined format within an application. This is like how generative AI is used to create images based on a prompt. These can be adjusted by the instructor after generating.

Authors can also prompt AI to review their content and rewrite it in a particular tone. They can prompt the AI engine to be more concise or expand on a topic.