Hi everyone! π
I wanted to share my thoughts and see if any of you have had similar experiences or feelings. As a blind person, I rely heavily on Alt Text for images to fully experience the digital world. Recently, I've been using Be My Eyes (Be My AI) and I must say, the descriptions I receive are usually rich, detailed, and quite entertaining. π They help me understand and enjoy the content much more deeply.
However, I've noticed that these beautifully crafted AI descriptions might not always be accurate. On the other hand, human-written Alt Text can sometimes be basic, plain, and less inspiring, but often more reliable.
Despite this, I've realized that I prefer taking my chances with the detailed, rich, and entertaining descriptions from AI. They make "looking" at the images so much more enjoyable for me. π
Am I being ungrateful or too critical?? I don't think so. I deeply appreciate every effort made to include Alt Textβmost don't even bother! π
So, I'm curious to know what you all think. Do you prefer the richness and detail of AI-generated descriptions, even if they might be inaccurate? Or do you value the reliability of human-written Alt Text, even if it's more basic?
Please let me know your thoughts! Are my preferences out of step? Let's chat and share our experiences. ππ¬
Looking forward to your insights!
Comments
Put them together
Maybe put AI descriptions with human descriptions and see what happens basically what Iβm trying to say is, maybe get AI to describe it, and get a human to verify whether the descriptions that AI gives is valid. So then you have the best of both worlds
Oh, I do that if
if it is important, you can't trust your safety to Ai. But for other stuff, like a photograph of my local beach - BME told a lovely story - the person who posted it just said it was "a photograph os X beach with Y island on the horizon."
Depends
Yeah, if it's unimportant, fun stuff, definitely AI. if it's more serious things, I'd cross-verify the AI-generated description with somebody. An alternative strategy for serious stuff is to use 2 different AI models to generate descriptions, and if there's factual difference, depend on humans. Having said that, I totally enjoy the AI-generated rich descriptions.
I agree
I agree with what?Gokul Said.
It is surprising
Just how good the AI descriptions are. I am always trying to ge tpeople who issue blanket condemnations to try them.
I also try and verify as many as I can. Some of the things on Mastodon - honestly I wouldn't beleive them if I hadn't had them checked. Memes of humanoid frogs smoking cigars in a space ship! That one was correct.
I even checked with a guy to see if how it described him and his cat - tru to, or at least he said it was.
Be my AI is wonderful
I havenβt really gotten around to exploring other ai image describers, but from what I can tell with be my eyes on iOS and windows, bmai AI description is brilliant
Claude 3. example
They claim it is even better. Here's is what it just said about my logo:
"Alt text: A stylized digital illustration of a cybernetic eye against a teal background. The eye is composed of glowing pink and white lines, with a dark pupil at the center. The iris is filled with intricate circuit board patterns, suggesting advanced technology. Energy-like wisps in pink and teal surround the eye, creating a sense of movement and power. Below the eye, white text reads "THE BLIND AI" with "BLOG" in smaller letters underneath."
That is how I wanted it to look, so I hope it is right!
Passive Alt AI Experience
I'm only aware of coming across AI generated alt text on Facebook, but maybe I'm just not aware. It ranges from substandard to less than useless compared to the text posted here as examples of AI or LLM descriptions. My iPhone generates better descriptions through VO. I don't always trust what people think they see in pictures. I very much remember, when I could see, not always understanding what I saw in a picture because of a weird angle or some kind of lighting issue, like the old Halloween trick of shining a flashlight upward across your face, or even looking at a projector slide of someone that was loaded backward, looking at yourself in a reversing mirror and finding out how everyone else actually sees the part in your hair on the left instead of the right...
A while back, someone posted a video of two AI devices discussing a room. Just to find out what would happen, I would like to see a demonstration of two AI arguing over which description is best or most correct, or closes to what a human would describe.
Fine tuning for Alt Text
It would be possible to post-train a model for alt text descriptions. Of course, this won't happen, but it could and the alt text would get better.
Facebook's auto-generated image descriptions are an abomination!
This is a nuanced question
I love the fact that I can capture an image and have it independently described, and agree that it is often very rich and detailed. However, I prefer human alt text. Firstly, it's accurate, but secondly, that richness you mention from AI can feel fake and overblown after a while because it uses the same kind of sense and concept words in its description. I also don't appreciate that it tells me how to interpret the image. In your logo, for example, it says the circuit boards signify advanced technology. But I might have a different thought. I might, perhaps, think of it as representing the idea of human and technological coming together, or as something to do with computer vision, specifically. This is not to say I want my alt text to be free of emotion or opinion--indeed one of my favorite things about human written alt text is that it can be so very creative and evocative in ways that AI is not, but I do notice.
Related to human alt text being better, too, is that it has an enhanced creativity if people bother. AI will always, like I said before, use the same types of words, the same type of tone, when perhaps the picture might have a different tone, sometimes funny, sometimes poetical or thoughtful. Look at enough AI descriptions and you will start to notice how each reads exactly alike.
So do I think AI alt text is helpful? Yes! Do I use it when I need something described, or someone was very brief, also yes! But do I think it's better than human written--absolutely not, never, no way no how. Between the creativity of humans and the qquestionable accuracy I will continue to fill my feeds with people who write alt text because it makes me smile.
Interesting Faerie
I wonder if in the end, it will come down to the same sort of difference as between couture clothing and off-the-peg. I've been looking at a lot of cat images on #Caturday and some people describe them like you say, some AI sound formulaic, but on the whole, I'm enjoying what I'm seeing of the visual internet after al these years.
I'm also not sure I can agree with your objections over AI offering examples or interpretations, it seems like part of communicating the idea of an image...as you say, it is subjective.
I am so very glad we are able to have a sensible discussion.
Human descriptions
Human descriptions can also be subject to bad intentions, trolling and whatnot, not to mention willingness to put alt text there in the first place. Human descriptions, assuming the human wants to be helpful and honest, can be very accurate but bland in some cases, and let's not forget that even the most honest humans can make mistakes. Indeed, AI is programmed by mistake makers, so there is that LOL. There always seem to be some drawback one way or another. When in doubt, verify, then verify the verification.
Thoughts
There was a discussion on Double Tap about this exact thing. Personally I think alt text on images is helpful, then I go to AI if I want more. I see alt text as replacing the iOS image description. It's not great but it is enough to help me know whether to bother or not.
Having said that there are places where getting AI descriptions is tricky. For example, Facebook seems to have no option to share to a different app. So I can save the image then load it again or I can screenshot or whatever but frankly I'm too lazy and not desperate enough to put every single Facebook image into AI on the off chance it's interesting.
One disadvantage of alt text is that it is only available at source. So if I save a photo locally then I lose the alt text.
One thing that I've noticed recently. I'm sure in the Photos app on the iPhone it used to give me a simple description of the photo as I was browsing through. But now I don't. Image descriptions is still enabled. Did I imagine this feature worked here or has it just broken?
@mr grieves
iOS does give a description of each photo in the Photos app for me. What has probably happened on your phone is after some update or another, the global settings buried deep within Settings>Accessibility>VoiceOver>VoiceOver Recognition have all been switched to off. You need to turn some of the switches back on to get it to work again, whether or not you have the settings in your rotor set to describe images.. At least that is what happened to me a while back.
* Comments about Facebook's "May be an image of text," alt text edited out... for my own sanity.
@OldBear
Thanks - I had double-checked all those settings and turned VoiceOver on and off, and used the quadruple 2 finger menu to ensure image descriptions were on, but still nothing.
However I am slightly ashamed to say that rebooting my phone has fixed it. I've watched enough episodes of the IT Crowd to have known to try that one already.
The upshot is that we have more tools!
And this can only be a good thing! When humans fail or don't give enough detail, we have another place to go, and this generally improves accessibility. I, too, am glad for this discussion--some interesting thoughts to ponder for sure!
Also, humans are most definitely subjective sometimes--and they can be some of my favorite kinds of alt text because you can really feel what the person saw when they described it. However even in those cases it often doesn't directly say what something is representing as an absolute fact the way AI tends to. This is what I was talking about earlier regarding objectivity.
Am I ready for Instagram
I do actually have an IG account now, I got it just last year - in comparrison I got a Facebook account in 2007.
I can use BeMyEyes to consume content on Instagram, but the sharing, with the flicking and tapping it takes, doesn't seem quite worth it yet. Maybe once Aple Intelligence is a thing and if it gives me the high-quality or at least longer, more involved, descriptions automatically, I wil go and try again. Who knows, I do love the cat photographs...
@mr grieves
Oh, sorry. I left that part out. It does require a restart of the phone to make those changes. It probably also downloaded a little code when you turned those options on, for anyone else going through this process.
The native, onboard image describer and text detector for iOS aren't too bad.
A question for Faerie
I was just Tooting this out on an image for Silent Sunday. I think it will go against what you said was acceptable, but I wondered what you think? It really paints the picture. Or at least it does for me:
"Here's a detailed alt text description for the image:
A stunning beach sunset scene with "Charlotte" written in elegant script inside a heart shape drawn in the wet sand. The heart is perfectly formed with decorative swirls. The sand has a soft pink hue from the sunset light. The ocean water is calm and reflective, mirroring the dramatic sky filled with puffy clouds in shades of pink, purple, and orange. The sun is low on the horizon, casting a warm golden glow across the scene. The image has a dreamy, romantic quality with its soft pastel colors and serene atmosphere."
That's a lovely image!
I can definitely tell it's AI, but it's welldone!
Alt Description
I've not posted this about three or four times because I'm not an expert, short-description writer, or anything else, for that matter. A couple of things seem... odd to me. ""The heart is perfectly formed with decorative swirls."" I don't quite understand the need to say the heart is "perfectly formed, perfectly," being the word that sticks out somehow.
Next, "the dramatic sky filled with puffy clouds" is somewhat visually mind jarring to me for some reason. It seems to be from using the word "dramatic" with "puffy." Streaks of clouds in those colors might be dramatic, but puffy is dreamy and sleepy to me.
I wouldn't mind coming across alt text like that, and it gets the point of the picture across. There were just a few small things that seemed off about it, and I tried to figure out what they were.
* Reading it again, it's the word "dramatic" that doesn't fit, because it does later say it is dreamy.
I imagine
I imagine this is what one would consider "an elaborate description". There is nothing wrong with it. It is not a good thing or a bad thing, think of it as a description from someone with a sense of visual art.
Regarding the description of the heart, I take it to mean that the lines and contours are symetrical.
Hope this helps. π
OB, you just can't feel it because it's me!
LOL! I can be dreamy and romantic! If I have to.
That's fine
Ya, the beach in your head isn't the beach in my head. Mine has palm trees and screechy seagulls.
Like I said, the word "dramatic" seemed to stick out from things like dreamy and serene. But at this point, I've read the description so many times it just is what it is.