Show HN: Think Like a Bot – Guess labels applied by AI to random images

The "I'm done" and "Guess" buttons are confusing.

"I'm done" can be replaced with "Give up and see answer" (Yep, it's an horrible name, try to find something shorter.)

I don't have a good suggestion for "Guess". I tried writing multiple words in the text box. Perhaps you can use JavaScript to ban spaces in the test box.

Also, perhaps you can change the game, so it says it's a success when you guess 5 labels. People really like to win, and sometimes getting all the label is boring.

I played through 10 or images. I think you should order the images from “AI is extremely confident” to “AI is confused” so we can see how progressively more complex images begin to be too hard for the AI. Or even getting into abstract images. What does the AI see in a Rorschach test?

Hi HN! I made this game where you have to guess what labels an artificial intelligence system (AWS Rekognition) applied to random images. Right now the images come from Unsplash. The idea was inspired by Geoguessr.

Let me know what you think. It's skillful and I think it's fun. There's a way to send challenges to your friends, as well.

It's an interesting idea. The first image I got had seven tags, including 'handrail' 'bannister' and 'railing'. Similar words seem like they'd be hard to guess, as I'd need to spend additional tries finding possible synonyms. Since I also don't know the categories the AI is looking for I could be hitting on something the AI wouldn't consider. It's interesting, but doesn't hold my attention the same way something like geoguesser would. A change that could work might be to include the AI's confident answers in a bank of about 10 choices. The user checks off their guesses and is scored on correctness after so many images.

An alternate mode I think could be fun would be to show two images and guess which was tagged for a given attribute and confidence level. IE you have a picture of a beehive and a picture of a flower and you need to guess which one was tagged for "Bee:85%"

I’m not sure why, but I grew tired after just two images. Most of the labels are duplicates, and while I can see why it’s good to have those, trying to guess all the different permutations of a name isn’t fun.

Ha, I don't know why I wrongly assumed this, but I already got stuck on the first image of a simple car. I assumed the game is about the edge cases of image recognition and that the labels will have nothing in common with the image like (photo of sheep dog -> floor mop), or (photo of pug -> bread)

This reminds me of Google Image Labeler which used a game interface to have people identify what is is images (and I presume train their AI to identify).

Is this the plan here as well? Or is it 100% purely a game?

https://en.m.wikipedia.org/wiki/Google_Image_Labeler

This reminds me of the game show Family Feud, except instead of trying to predict what 100 other humans answer, I'm predicting what the algorithm answers.

I keep clicking on I'm done, is it just me or should it be on the right side instead of on the left side?

I wrote my 10 answers in one line and said I am done, of course both actions were incorrect. I figured out the game but wasted time.

Multi answer in one line should be enabled. I am done button should be changed to make this game more fluid. My 2 cents.

Nice game.

Fun, thanks for making this!