Are the Robots Getting Better at Image Recognition?

I was logging into an account today and was presented with a CAPTCHA that struck me as quite odd.  Normally, the CAPTCHA images are as clear as ever. However, look at these images below:

The sample image of the car at the top right is fairly clear, yet the selection choices are so highly pixelated as to make me check my screen resolution and my eyeglass prescription.

It should be noted that the site I was accessing belongs to an “internet giant” that has deep insight into artificial intelligence and machine learning.

I checked another site, which is not run by an internet giant, and although the images are provided by another giant in the internet realm, remarkably the images are very clear:

Is it possible that the internet robots are not only becoming more adept at checking the button that says “I am not a Robot” but that they are also now able to pick out the images on the CAPTCHA that follows?

When you take a moment to think about CAPTCHA images, it is perfectly reasonable to assume that this is possible. A computer is capable of reading the text on the screen, including reading it aloud for you. Is it beyond the realm of reason that if the computer recognizes the word “crosswalk,” the next instruction would be to search the page for all hash marks and select those images? Or perhaps if the word “car” is recognized, the computer can also select all images with wheels and glass?

One of the problems with the CAPTCHAs is that the images must be recognizable to a broad population.  Have you ever noticed that most of the CAPTCHAs are quite repetitive? For example, how often are the images those of Crosswalks, Staircases, Motorcycles or Cars?

Creating these (Read more...)