I've made an NVDA add-on called "Image describer" that takes the current navigator object and then says what it contains by looking at it visually. Kind of like the Facebook auto alt text, but this works anywhere.
The shortcut to interpret an image is NVDA+Control+I.
The add on can be downloaded at http://www.screenreader.ai/
I'm a sighted machine learning researcher that recently have caught an eye on the field of web accessibilit
Websites are primarily designed to be interacted with visually which is hard to do if you have a visual impairment. Here I think there's big potential for AI to help and look at the web page visually and then give correct and helpful information to the visually impaired or blind user.
I'm looking towards making much more than this, please contact me or reply to this thread if you want to get help with something accessibility wise or have any kind of feedback.