Riya and Privacy

Riya doesn't look for anyone in your photos unless you have trained Riya on that person. So you have to have known them already. Jennifer Grannick wrote this article about privacy. While she is touching on a bigger shift in the concept of privacy in the digital world, I thought I'd take this opportunity restate how Riya works.

Riya doesn't look for you or tag you in another's photo unless they train us and have you in their friends list. Even if they do, it is not public unless they make it public.

So if you are walking with your mistress (you'd be surprised how many people bring up this example with me..;-) and a stranger has that photo in his collection unless he can already identify you, neither will Riya.

The Riya algorithm starts performing very poorly after about 300 unique people so even if someone wanted to use Riya they couldn't it will be complete garbage.

Think about the fact that even human beings can't do it this. If I give you a list of 1000 people to recognize I'm positive you will have trouble after a few hundred. There are a lot of studies that show after 100-200 people in an office most people loose the ability to idenitfy most of the people.

All of that being said, our alpha is closed on purpose (Jennifer's article indicates that we are in open Beta but we are not). Our approach is not "tart". We wanted to learn more about the privacy implications of our system from the closed alpha. As we gather that we will figure out ways to help safeguard this. For example - maybe you can make any photo private in which you where recognized. We are very vigilant of this but we first wanted to learn and get feedback from our existing alpha testers.

Riya is not alone in this issue. Today I can upload a photo to Flickr - tag it with the name of the person in the photo and make it publicly searchable. Now that is integrated with Yahoo you can see the photo of the person when searching.

Comments

[Peter Rothman]

With all due respect, if Riya's face recognition system could recognize 5000 people would this make your system less secure? If it could only recognize 50 people would it be more secure?

You are assuming a very unsophisticated approach to attempting to use your system to violate someone's privacy. The entire security of your system hinges on an outsider's inability to get access to the image database. If I can get into your database, or scarf it up gradually over the Internet, then I can potentially point my own face recognition software at a copy of the database and search for whomever I want.

Even in the scenario you mention, searching for someone in a photo with their mistress, the recognition system's performance doesn't impact Riya's security. Imagine I am a hi-tech private detective trying to find out if a rich executive is cheating on his wife. Potentially millions in alimony are at stake.

In this scenario I am only looking for exactly one person in a large database of images. If I can get into your database and I have my own FR software (remember I said I was a hi-tech PI) then I can look for and probably find my target if he's in there.

This is not only true of Riya, but also of any publicly searchable image database such as Flickr, 23hq, etc. Any of these sources could be used to the extent that the images are published openly or can be accessed covertly through some sort of security breach. Its the database of images you need to protect not access to the face recognizer.

Now of course, if you hadn't taken the very sensible steps which you've already mentioned to prevent easy use of the recognition software for improper purposes, this would be much easier to do I admit. You've essentially made it hard or impossible for the average person to casually snoop on their spouse or whatever. That's important of course. But you should also realize that your system can still be used in ways you might not expect by someone that is more sophisticated and determined.

[Andrew]

I am afraid I agree with Peter; in the wrong hands, this potentially could be very dangerous.

As incredibly powerful an amazing your software is, it doubles as terrifying. Privacy is a great concern of a great many people. Tomorrow a case goes to the Supreme Court regarding the requirement to show identification before boarding an airplane. Potentially, the need to ever show identification at any place could be removed entirely if a camera hooked to the internet and your database captures the face of a person before they board the plane. This is a hypothetical situation, but a serious concern. Privacy in general could be at stake if further development of this software continues without fully considering the infinite amount of possible security infringements, and I am afraid such is and impossibility.

I am warranting my concern, and warning you to be careful. You have a powerful invention in your hands, do not misplace it.

[David Telleen-Lawton]

Actually, the real culprit is that magical box that can capture the spirit of the person and put it into pixels. Kodak, Canon, and others are just pawns of the state...

[Dan]

Although I already wrote about this yesterday and voiced my opinions to Munjal here: http://techandother.blogspot.com/2005/12/riya-and-false-hysteria-over-privacy.html, I want to expand on my own posting here, since this seems all a little too out there.

I can't say I agree with the comments so far. The privacy concerns being expressed about Riya aren't so much a function of the Riya service, but of being captured on film.

Being captured on film isn't anything new, and it definitely is used to invade people's privacy, but I fail to see how Riya makes this situation any worse.

As stated in the post, the training you do in your own Riya account can't be used by others, and you have the option of making your pictures private. Yes, there is the possibility of some super-hacker breaking in, as in any system, but I'd be more nervous about said hacker using their skills to get into my bank account than a Riya account.

Could a super-powerful private detective build a facial recognition system in the hopes that someone somewhere took a picture of his/her target and put it in an online service? Sure. But I think they would probably just go take the picture themselves.

As far as having a camera in a public place like an airport that is linked to a database that links your photo to your information, it's already out there, and has been for quite a while. If you've ever walked into a casino, you're probably in their database, which it then shares with every other casino possible. If you think the government doesn't already do this, you're kidding yourself.

The only security concern here is the protection of data from theft. In Riya's case, they need to protect the "trained eye" of each Riya account from being stolen, combined with a million other "trained eyes" and used in some other context. (The most likely candidate would be something like the advertising in 'Minority Report') But again, as stated, the Riya technology isn't sophisticated enough to be that accurate. There are, however, plenty of systems out there that are much more capable. Nothing is stopping any one of us from buying the software, and writing a script that pulls down a hundred pictures at a time from Flickr and processes them.

Basically, I find the argument of Riya being a security threat ridiculous. Who, exactly, are you worried about knowing who you are? The people who would want to invade your privacy already have access to your background check and credit report. From this they know where you live and can take as many pictures as they want of you, your family, your car, when you go to work, etc. There are a lot more ways to get to your information than Riya, and Riya is one of the least efficient ways of doing so.

Riya isn't making this situation worse, it is the already-legal methods that can be used that are a threat to your privacy. The issue here is that Riya is new, and everyone likes to take the new, unknown technology and try to find things wrong with it.

Riya is absolutely no different than Flickr, except that its tags are automatically generated for the first hundred users in your account, and you have to manually make the tags in Flickr. Unfortunaely, whenever any new technology comes out, everyone wants to blow the doors off some supposed huge security risk and become the next internet superstar. This just isn't the case with Riya.

[Dan]

Sorry, link picked up the punctuation:

http://techandother.blogspot.com/2005/12/riya-and-false-hysteria-over-privacy.html

[Peter Rothman]

I think we are mostly in agreement. My point was that facial recognition, your "trained eye", isn't where the privacy threat is here. It doesn't matter how good my FR is if I don't have access to a database of images. Even if I had 100% perfect FR it would be useless so long as the pictures remained in millions of cameras which I can't access. However, once these images start to be present in large online databases, and Riya is only a tiny one at present, the possibility of accessing and searching them exists. If I have access to the database I can get my own FR software and search it, limited only by my own financial and intellectual resources. Its the aggregation and publication of image databases that poses the privacy threat here, not the use of FR instead of hand coded tags to sort the pictures. This technology is still somewhat out of reach for the average person, but its certainly not for more sophisticated and enterprising individuals and organizations especially when there is profit to be had. I've already mentioned the case of finding incriminating pictures related to a divorce, but there are other ways to make money with this too.

[Antonio Carlos Mansur]

Very god idea!

Congratulations!

acgmansur@yahoo.com