Slashdot Mirror
Google Preparing 'Invisible ReCAPTCHA' System For No User Interaction (bleepingcomputer.com)
An anonymous reader quotes a report from BleepingComputer: Google engineers are working on an improved version of the reCAPTCHA system that uses a computer algorithm to distinguish between automated bots and real humans, and requires no user interaction at all. Called "Invisible reCAPTCHA," and spotted by Windows IT Pro, the service is still under development, but the service is open for sign-ups, and any webmaster can help Google test its upcoming technology. Invisible reCAPTCHA comes two years after Google has revolutionized CAPTCHA technologies by releasing the No CAPTCHA reCAPTCHA service that requires users to click on one checkbox instead of solving complex visual puzzles made up of words and numbers. The service helped reduce the time needed to fill in forms, and maintained the same high-level of spam detection we've become accustomed from the reCAPTCHA service. The introduction of the new Invisible reCAPTCHA technology is unlikely to make the situation better for Tor users since CloudFlare will likely force them to solve the same puzzle if they come from IPs seen in the past performing suspicious actions. Nevertheless, CloudFlare started working on an alternative.
Guess they got tired of people saying store fronts were rivers in their image classification.
Part of me is happy, those things can get annoying, hopefully the algorithms used work well, and doesn't make too many (if any) false positives.
Here's how they usually for me right now.
"Please click the things that are storefronts."
"Uhh... ok. That's a section of a brick wall that might be a house or might be a store, and I can't tell because I can't see enough of it. I'll guess... oh, and that one might be a considered a store, or it might be a government building that isn't a store, and it's hard to tell which from the picture...."
By the time I click done, my probability on each guess gets multiplied together into a number that is apparently about 0.1, and it asks me a new one all over again, "Click the images that contain a car". "Uh..., well that one contains a light truck, do you consider that a car, or not?" And off we go all over again.
I'd be happier if they asked something concrete that wasn't subject to so much fuzzy interpretation. "Click the image that is the derivative with respect to X of the following image of a polynomial". That would have a specific answer not open to shades of interpretation.
Some bots are actually browsers driven by scripts as opposed to a web driver provided by the browser. How will it differentiate me clicking a link from a bot clicking the mmouse on a link?
The "puzzle" (not really a puzzle), solved once by a human, can be tweaked by AI to bypass any useful measures. What is the input they are measuring - mouse directions and velocities? Pretty sure that can be faked out to look like it comes from a human. This is war of the AIs!
Their last system made me uncomfortable.
How can I believe you when you tell me what I don't want to hear?
Have Madonna Do it. She owes everyone who voted for Hillary a blowjob.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
That's what I get for voting for Trump :-(
This can't come out sooner
i didnt vote and didnt care.
fucking idiots trigger captchas just for using advanced search functionality like "site:". i've switched to Bing out of spite.
"According to this specification, CloudFlare is working on a Tor Browser extension that generates one-time authentication tokens, called nonces."
Is the author of TFA sure that they weren't referring to Tor users themselves?
I welcome this.
I've been curious about those "I'm not a robot" checkboxes. How does it differentiate between a humsn and script checking that box?
I've guessed that it had to do with timing of page load vs checkbox, or with tracking mouse movements for human-like movements.
Anyone have the scoop?
How will this improve their data mining efforts? Have all their street view images been successfully tagged and categorized?
Captchas are the AIDS of the internet.
I've been doing something similar for years. It's probably not as complex as what google is doing, but with little more than a few tricks with a hashed timestamp, and making sure that javascript works, it stops most bots.
Yeah...but Hillary got more votes.
See, that means more people actually want her as President.
The only reason Trump is President is because of the outdated, bassackwards, loophole flukes of the Electoral College.
But hey.....keep making yourself feel like your candidate actually deserves it!
If it's like their previous attempts at CAPTCHA-less CAPTCHAs this will involve taking information like mouse movements and your browser fingerprint, and then analyzing that to decide if you're human. Deeply concerning. Google already knows enough about my browsing habits by having their scripts embedded everywhere.
Before Google bought reCAPTCHA, (and shortly thereafter), the goal was citizen science of sorts. It'd give you two words, one is a word from a book that needs to be OCR'd, and another that the machine generated (as usual).
Over time, people filling out CAPTCHAs would help OCR old books.
Now the challenges are along the lines of "help Google learn how to OCR street numbers!", which is less useful.
Yeah, anecdotally I think I get the full recaptcha test when I happen to be in an incognito window. Pretty rarely in the regular browser windows.
why is the internet crippled with this stuff anyway - who cares if i download or a bot does it for me? a huge leap backwards to the dark ages
Yeah...but Hillary got more votes.
That's because Trump didn't invest too much in CA. If the system was different, campaigning would have been different, and result the same.