[AI] Kittens Could Solve Spam

Vikas Kapoor dl.vikas at gmail.com
Sat Aug 4 10:20:24 EDT 2007

Kittens Could Solve Spam
As spammers get better at defeating text-based Human Interactive Proofs (HIPs), the next step in anti-spam technology may rely on photos of kittens and
other images that are easy for humans to identify, but difficult for machines.
Nancy Gohring, IDG News Service

Friday, August 03, 2007 3:00 PM PDT

ting spammers. Powerful software tools and supercomputers aren't involved, but kittens are. 

Or rather, photos of kittens. Kevin Larson, a researcher at Microsoft's advanced reading technologies group, has found that asking a user to identify the
subject of a photo, like a kitten, could help block spam programs. 

Currently, services like Microsoft's free e-mail service Hotmail require new users to type in a string of distorted letters as proof that it's a human signing
up for the account and not a computer. Called Human Interactive Proofs (HIPs), Microsoft, Ticketmaster and a host of other companies have been using the
system for around five years, Larson said. He spoke in Seattle on Friday at TypeCon 2007, an annual conference put on by the Society of Typographic Aficionados
for type enthusiasts and designers. 

When Hotmail first started using HIPs, the number of e-mail accounts generated on the first day dropped by 20 percent without an increase in support queries,
Larson said. That was a sign that the HIPs were fooling the computer programs that spammers use to automate signing up for new Hotmail accounts from which
spam is sent. However, spammers learned how to tweak their programs to better recognize the HIPs, he said. 

Now, it's a race for Microsoft to continue to alter its HIP system to fool the computers, which ultimately seem to catch on. Larson's group at Microsoft
experiments with different ways to distort the text used in HIPs in a way that is easy for humans to read but difficult for computers. 

One twist on the HIP idea that they've worked on is to display 16 or more photos and ask for identification of the photos. In an example, he suggested using
pictures of cats and dogs. The problem with the concept, however, is that Microsoft would have to create a massive catalog of photos, otherwise the programmers
could match the correct response with each photo in the catalog and begin to spoof the system, he said. 

Audience members had a variety of ideas for ways to expand on the idea in order to try to beat the spam programs. One suggested that Microsoft continually
take videos of a kitten jumping around a room, as a way to generate a nearly endless string of photos for identification. 

"It's possible that kittens are the wave of the future," Larson joked. 

Microsoft might also be able to use short video clips instead of photos, one audience member suggested. The cost to support that method might be a concern
but it could probably work, Larson said.  

His group is also working on ways to improve the current letter-based HIPs for human users. "We need to figure out how to make HIPs that are more pleasant
to read," Larson said. Many computer users may be familiar with the "ugly distorted texts" that HIPs use, he said. "We let the computer science people
generate this text, but this is a design problem. It seems we ought to bring what we know about legibility to make things more pleasing to identify yet
still stop computers," he said. 

His team has thought about using beautiful calligraphy characters set against ornate backgrounds, but such letters haven't been good at fooling the computers
because a program can identify the form of the letter by the thickness of the font compared to the lines in the background design and because a program
can notice color differences of the font compared to the background, he said. 

With 90 billion pieces of e-mail spam sent every day, according to Larson, companies like Yahoo Inc., Google Inc. and Microsoft that offer free online mail
services have an incentive to try to block spam. Otherwise they pay for the resources that help send the spam. 


Vikas Kapoor,
MSN Id:dl_vikas at hotmail.com, Yahoo+Skype Id: dl_vikas,
Mobile: (+91) 9891098137.

More information about the AccessIndia mailing list