How Old R U?????
A new smartphone app aims to help kids identify what's behind a screen name
Steven Cherry: Hi, this is Steven Cherry for IEEE Spectrum's "This Week in Technology." By some estimates, as many as one in four kids using the Internet has been approached by an adult pretending to be a kid. It's a serious problem, because some of those adults are pedophiles, and the online deception can turn into real-world harm. It's hard for kids—or anyone for that matter—to figure out who or what's behind a screen name. In an experiment conducted by researchers at Lancaster University in the UK, only 18 percent of the kids being tested could tell when they were talking to adults.
But now there's an app for that. Isis Forensics, a spin-off company of Lancaster University, has developed smartphone software that can tell whether the person you're chatting with online is likely to really be an adult pretending to be a kid. The app was released last week at Apple's app store. My guest today is James Walkerdine, the managing director of Isis Forensics, and one of the developers of the ChildDefense app. James, welcome to the podcast.
James Walkerdine: Hello
Steven Cherry: Maybe we could just start with how the app works. It's looking for certain characteristics in what people type?
James Walkerdine: So what we've done is that we've basically built language profiles for children and also for adults masquerading as children. And then when you get a new piece of text from a chat log or whatever, a conversation, it then gets compared against these two reference profiles to see which one it closely matches. So the actual analysis itself, I mean, it's fairly complicated and it isn't just looking for keywords. It's basically doing statistical analysis and building up a model of the language that's being used in a chat log and then using this various statistics that come out of that model and comparing it to the reference profiles.
Steven Cherry: Can you give us an example of how an adult might accidentally reveal himself?
James Walkerdine: Basically there's little things that will catch them out, things like they're using too many exclamation marks than kids would tend to do. Or maybe they're using commas more often than kids would tend to do. Often it's something that you can't see by the human eye, but a computer when it's doing an analysis of the text and producing all these various stats, that's when it can spot the differences. So like I say, often if I'm looking at, you know, a piece of text that looks like it's bold that is from an adult who's pretending to be a child, you know, I can't normally tell the difference. It looks like a kid's text to me. But it's only when you look at the actual stats that are produced by analyzing all the words together, and the punctuation, the sentence length, and all this sort of stuff, that's how the software can sort of spot the difference.
Steven Cherry: And so how accurate is it?
James Walkerdine: Accuracy is a difficult thing to say because obviously it's very much depending on context and also on how it's being trained. So I'm wary about giving, you know, a specific accuracy, because it does very much depend on, you know, the context and the trials. But certainly, you know, the trials that we've done, we did a trial of basically 50 chat logs that were produced by kids, and then there were 50 chat logs that were produced by adults who were masquerading as kids. In that trial it managed to correctly identify all the adults, but again that's just that one trial. In a different trial it won't be as good.
Steven Cherry: Very good, so aren't there software programs out there already that can analyze language and create a profile of the person typing? What's new in this app?
James Walkerdine: Well, I mean, I guess the key thing is that it's written on the phone itself. I mean you are right in saying that the field of natural language analysis, you know, it's been around for decades and there's obviously lots of work in that area and lots of powerful software that's being produced. But all this software runs on servers. It runs on powerful servers where you send off your text to be analyzed and there's lots of number crunching going on. What we wanted to do was to try and build something that could actually run on the phone itself. So in this case all the analysis is being done on the phone. It's not being sent off anywhere. It's all been analyzed on the phone, which basically means that obviously you're tied by the resources that the phone has and the limited amount of power it has. So the challenge that we faced therefore was trying to build an analysis engine that on the one hand was efficient and could work on the phone, but on the other hand also had a reasonably high level of accuracy.
Steven Cherry: Yeah, there were two interesting things I thought about the software. One is that it runs on phones and not computers and the other thing is that the way it's designed, kids themselves that are doing the detecting and I guess you decided that that wasn't something that parents wanted to be doing. Is that right?
James Walkerdine: Yeah. We had a parent focus group, and we had various parents from the local community attending, and one of the things that came out of the very first group was that parents at the moment don't have much choice in the type of technology they can, sort of, use to help protect their children. At the moment it's very much based around the idea of basically monitoring everything they do, monitoring the Web sites they visit, monitoring all their conversations, and then parents will get perhaps an e-mail whenever certain things are, sort of, flagged. And the problem that parents had was that certainly that if they've got teenager children, I guess it doesn't really go down very well with their kids. Their kids—obviously they want to have their own independence, they want to be able to be trusted to look after themselves. And the parents basically felt that having to resort to a monitoring approach in some ways was almost like a last resort. Only if things got really bad would they have to, sort of, resort to it. And what they suggested to us was that why isn't there something that is actually being built for children themselves to use.
Steven Cherry: It does occur to me, though, if kids can use this to detect pedophiles posing as kids, is there anything that prevents pedophiles from using it to detect undercover police posing as kids to find the pedophiles?
James Walkerdine: Yeah, I mean, that's a question that, that's been asked, and, I mean, I guess obviously there's nothing you can do to stop that from happening. The main thing is that the software has been designed to monitor conversations rather than individual messages. So really, you know, if a pedophile does want to use it to, sort of, identify police or even to try and improve themselves in a way, it will mean that they have to go to quite a lot of effort to actually engaging in numerous conversations with an individual before that software can then be, actually be used by them. But, I mean, like you say, that is a potential risk of the software.
Steven Cherry: Very good. Well, the world definitely needs some software that does something like this, so thank you very much.
James Walkerdine: No worries.
Steven Cherry: We've been speaking with James Walkerdine of Isis Forensics about a new smartphone app that can help children identify adults posing as peers. For IEEE Spectrum's "This Week in Technology," I'm Steven Cherry.
Segment producer: Ariel Bleicher; audio engineer: Francesco Ferorelli
NOTE: Transcripts are created for the convenience of our readers and listeners and may not perfectly match their associated interviews and narratives. The authoritative record of IEEE Spectrum's audio programming is the audio version.