Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
×
Chrome AI Google

New Chrome Feature Will Use AI To Describe Unlabelled Images To The Vision-Impaired (techspot.com) 27

An anonytmous reader quotes TechSpot: Google is looking to improve the web-browsing experience for those with vision conditions by introducing a feature into its Chrome browser that uses machine learning to recognize and describe images. The image description will be generated automatically using the same technology that drives Google Lens...

The text descriptions use the phrase "appears to be" to let users know that it is a description of an image. So, for example, Chrome might say, "Appears to be a motorized scooter." This will be a cue to let the person know that it is a description generated by the AI and may not be completely accurate.

The feature is only available for those with screen readers or Braille displays.

"The unfortunate state right now is that there are still millions and millions of unlabeled images across the web," explains Google's senior accessbility program manager.

"When you're navigating with a screen reader or a Braille display, when you get to one of those images, you'll actually just basically hear 'image' or 'unlabeled graphic,' or my favorite, a super long string of numbers which is the file name, which is just totally irrelevant."
This discussion has been archived. No new comments can be posted.

New Chrome Feature Will Use AI To Describe Unlabelled Images To The Vision-Impaired

Comments Filter:
  • Let's see: "Google, show me a picture of Patrick Ewing in the 4th quarter of a basketball game".

    ...hilarity ensues...
  • HTML alt Attribute (Score:4, Informative)

    by Chromal ( 56550 ) on Sunday October 13, 2019 @09:51AM (#59302824)
    "The required alt attribute specifies an alternate text for an image, if the image cannot be displayed. The alt attribute provides alternative information for an image if a user for some reason cannot view it (because of slow connection, an error in the src attribute, or if the user uses a screen reader)." ( https://www.w3schools.com/tags... [w3schools.com] )
    • But reinventing the wheel is the cool thing to do

      • by Chromal ( 56550 )
        I'm mostly annoyed they haven't made the feature accessible to all users. I was sitting here contemplating what I had written (above). It's true HTML isn't always used the way that was envisioned at CERN so long ago, and it isn't realistic to simply condemn non-compliant HTML. I can't fairly say "shame on the publisher for putting it out in a non-accessible manner," since there isn't always a publisher/editor/author process behind all HTML documents on the web, nor are all HTML documents simply articles wit
    • "The unfortunate state right now is that there are still millions and millions of unlabeled images across the web," explains Google's senior accessbility program manager.

      "The required alt attribute specifies an alternate text for an image, if the image cannot be displayed. The alt attribute provides alternative information for an image if a user for some reason cannot view it (because of slow connection, an error in the src attribute, or if the user uses a screen reader)."

      As TFA clearly says, "there are still millions and millions of unlabeled images across the web". Also, it is an overstatement to claim that the alt tag is required. It is not [medium.com]. You wouldn't want to put an alt tag on a shim, or an image that's just a background for some bit of content. Your quotation is therefore both incorrect and irrelevant.

      • by Chromal ( 56550 )
        The IMG ALT tag is very much relevant to the problem and solution this article is about. Most of what you just said was discussed further in the follow up comment posted prior to your reply.
        • by tlhIngan ( 30335 )

          The IMG ALT tag is very much relevant to the problem and solution this article is about. Most of what you just said was discussed further in the follow up comment posted prior to your reply.

          And there are plenty of images on the web that are simply not labelled or labelled with something generic. Forum images, for example, may or may not allow one to specify an alt attribute when you inline an image.

          Some comics are simply ALT'ed with the date of the comic. Others contain the text elements.

          In an ideal world,

      • Your quotation is therefore both incorrect and irrelevant.

        As you KNOW, he quoted the official and only spec. It may be irrelevant, but it sure as fuck isnt, as you so stupidly and ignorantly put it, "incorrect."

        Given what you KNOW, you are being a dishonest fuck right now.

        • As you KNOW, he quoted the official and only spec.

          Despite the name, as you apparently DO NOT KNOW, w3schools is not affiliated with w3c [w3schools.com]. As the linked page says, however, "Use alt="" if the image is only for decoration". But there is literally no web browser which will refuse to parse an img tag without an alt property, and an empty property is logically equivalent to a missing one, so the alt tag is not required.

          Given what you KNOW, you are being a dishonest fuck right now.

          Hush now, child.

          • Okay, here's the official W3C spec (https://www.w3.org/TR/html401/struct/objects.html#adef-alt)

            "The alt attribute must be specified for the IMG and AREA elements. It is optional for the INPUT and APPLET elements."

    • by Merk42 ( 1906718 )
      Except an alt tag may simply say "Pizza" which isn't descriptive enough.
  • I fully plan on trying this feature (with my eyes closed) to view pornigraphy...

    I'm just curious if it would say something like, this is a pornographic image, or actually describe what is happening, and perhaps give some commentay..

    • Around 13 years ago someone posted a page that had a porn image, that would speak the name of whichever body part or object, in a harsh synthetic sounding voice when the given region was clicked: "knee", "nose", "nurse's hat", "shaved whatever", ... This should have been saved as a prior art reference, lest someone try to weasel a patent.
  • It will use universal trainable functions to match patterns in input vector data to names, using weight matrices created via optimization.

    That has as much to do with artificial intelligence, as blue cotton candy flavoring liquid ... extract derivative ... sustitute ... concentrate* has with food.

      (* contains a source of phenylanaline)

    • That has as much to do with artificial intelligence, as blue cotton candy flavoring liquid ... extract derivative ... substitute ... concentrate* has with food.

      These days, everyone puts an "AI Inside!" sticker on everything.

      Heck, the way folks are twisting the definition of AI now . . . my Dad's ancient slide rule would be designated as AI.

  • "The feature is only available for those with screen readers or Braille displays."

    Likely because if this feature was sufficiently advanced then lawsuits over DRM Public Performance BS would ensue.

    This is not just something that is useful for blind people, this is useful for everyone & everything just especially for blind people. I would love to have a high quality "algorithm" read a web page off to me & describe images or video while I do something else, this would be awesome! It could be used for

  • 90% of the time the AI will say: "You don't want to know!"

    • 90% of the time the AI will say: "You don't want to know!"

      Naw.

      "This is a cat napping."
      "This is a cat riding a Roomba."
      "This is a cat napping."
      "This is a cat napping."
      "This is a cat napping."
      "This is a cat chasing a laser pointer."
      "This is a cat napping."

      Not a very difficult algorithm to train, really. Everyone knows the Internet is just cat pictures.

Every nonzero finite dimensional inner product space has an orthonormal basis. It makes sense, when you don't think about it.

Working...