Video Ads that Outwit Spammers
Websites may soon be asking you to watch a short ad rather than enter garbled text.
Sign up for an account on almost any website and you’ll be asked to read and type out a string of distorted characters. This task, known in the business as a Captcha, is meant to be easy for a human but near impossible for a software “bot,” thus preventing spammers from automatically creating accounts in order to send out e-mail or social networking spam.
A startup called NuCaptcha plans to refresh the Captcha concept by asking users to enter text displayed as part of a short video advertisement. The service launches today with advertisers including Disney and the computer games publisher Activision.
“Captcha is the front-line defense on the Web, and use of it has doubled over the last two years,” says Michel Giasson, CEO of NuCaptcha. “But its current form has become less usable and not as secure as it used to be.”
Recent years have seen an arms race between Web companies and the spam industry, with text Captchas becoming increasingly complex and difficult to read. The text has become more distorted in response to better image-recognition software that can get past older Captchas.
Giasson says using videos instead of mangled text is easier on users, more secure, and creates a lucrative new form of advertising.
NuCaptcha replaces jumbled text with video ads. A line of text–usually an advertising slogan–scrolls over the top of a video, with a code a few characters long at the end that a user must type in. A publisher can also choose to require that people type in the whole slogan to complete the Captcha. Video is much harder for software to process, says Giasson, meaning the text can be easier for humans to read than conventional Captchas.
A Stanford University study provided evidence earlier this year that Captchas are becoming more difficult for users. Researchers recruited users through Amazon’s Mechanical Turk service to test the difficulty of static Captchas. They found that people could only solve 75 percent of the Captchas posed by Recaptcha, one of the most common Captcha services on the Web. Some of the text it uses is drawn from Google’s book scanning project; when people solve one they transcribe text that has stumped automatic text-reading software. The comparable figure for eBay’s Captchas was 93 percent. NuCaptcha says that trials of its technology yielded a figure of 97 percent.
“Text-based Captcha is not broken per se, but it is clear that some combinations of features are really hurting the user experience,” says Elie Bursztein, a member of the Stanford team. “Video is certainly an interesting direction to explore–in theory, it contains more information than a fixed image, so if the brain is able to use that, a video Captcha might be easier.” However, Bursztein notes, providing more information could also give attack software more to work with.
NuCaptcha has built added security layers into its system. It tracks individual users and serves up more complex Captchas if any behavior suggests the viewer might be a software bot or someone paid to solve Captchas as part of a spam operation. If the system detects abnormal activity, the text can, for example, be made to move faster across a video, or the string of text can be made longer.
“We try to tell different individuals apart, and based on the pattern of interactions we record, we can adjust the Captcha if we think they are a risk,” says NuCaptcha’s chief technology officer, Christopher Bailey. “What you did yesterday affects what you see today.”
NuCaptcha was first developed as a purely security-focused product, with the advertising element added later. “No one really pays attention to banner [ads] anymore,” says Giasson, who adds that the NuCaptcha approach “ensures publishers and advertisers of the attention of users.” Some studies suggest that people’s recall is improved when they type or write down what they see, he adds.
NuCaptcha’s technology is unlikely to end the arms race with spammers. “Devising a good Captcha comes down to knowing what is hard for computers and easy for humans,” says Bursztein. “As the AI field progresses, the gap between the two becomes narrower and narrower.”