Slashdot Mirror
Researchers Have Found a Way To Root Out Identity Thieves By Analyzing Their Mouse Movements With AI (qz.com)
An anonymous reader shares an article: In the study, published recently in PLoS One, the researchers quizzed 40 respondents about their personal details. Half of the respondents were asked to answer the questions truthfully, but the other half were given details about fake identities they had to memorize and use in the quiz.
The computer quiz kept track of the movement of each respondent's mouse as they answered the questions, and noted how the fakes differed from the truth-tellers when they moved the cursor from the bottom of the screen to the answers at the top. The quiz consisted of 12 questions like, "Do you live in Padua?" and "Are you Italian?" That covered details an identity thief could easily remember and answer, but then the quiz threw them a curve ball. "What is your zodiac sign," it asked in the second series of 12 questions, which were designed to be easy for the genuine respondents, but more difficult for the fakers to work out. After the researchers took the mouse-movement data collected from the quizzes and trained a machine-learning algorithm to analyze it, they found that was indeed the case. It was able to discern the fake responses from the real ones 95% of the time.
I have no fucking clue what my zodiac sign is, I'd have to look that crap up. I'm not just not into fairy tales for adults.
I respond .... "Neon".
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
This can also work with signature pads. A forger maybe able to fake your signature, but they have a much harder time faking the rhythm and timing of the strokes.
Another test that measures timing is testing for pedophila. Patients are shown pictures of children, adolescents, and adults, and asked to rate each picture for attraction. The ratings are then ignored, and instead the "dwell time" for each photo is what is actually measured, since that is much more predictive.
Are they even trying anymore ? It sounds more and more that the way to research work is 1) gather shit data, 2) feed any combination to it to neural net, 3) look for correlation, 4) repeat if nothing interesting, 5) publish. At this point, they might as well rename AI for AR, as in Artificial Research...
Then to fool the AI, all you need is an AI program to drive the mouse to mimic the original person.
By itself, this study just shows that a machine learning algorithm can compute a statistical average (regarding mouse movements) that classifies a specific set of 40 people into two groups.
If the same statistical average also classifies other groups of people accurately, then you can make real claims about separating truth-tellers and liars.
See this good explanation of pattern classification and its uses and misuses.
Worse, now that this technique has been revealed, the real thieves meanwhile will remember to note down info they enter in pieces of paper as the go along, thus they can revisit, open another google window to search for stuff etc.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
But their dowsing rod is now better than ever... or was in a test, anyway.
No significant result possible.
Ah, the zodiac sign gotcha! Used to use that when I was a bouncer. I didn't memorize the dates, it was enough just to see the hesitation in the response.
Might work great in a museum, in the real world everyone is using phones and tablets now.
touch screen... not much movement except for hitting the answer.
If someone were to steal your online banking credentials, log in as if they are you, and try to empty out your account, this kind of system might prove useful. Unless of course you have a tennis elbow, sore arm muscles or just injured your hand, in which case you would not be able to use your own money and possibly starve to death as a result. =) Remember kids - always keep a small wad of 100s tucked away in your sock, for days when the thief-detecting-mouse-movement-algorithm thinks that YOU are not YOU. And yes, in Soviet Russia YOU are the Mouse Analysis Algorithm. =)
Why did the chicken cross the road? Because Elon Musk put an AI chip in its head.
Seriously? Hackers? Mice? GUI? That's funny.
At around 2012 - 2014 Google's Russian office developed a very similar technology for clickfraud detection. What has happened later - the best talent involved in developing it was given a relocation offer to Switzerland given that he was just SO INCREDIBLY VALUABLE, but the guy refused. Their HR's became more assertive, threatening to ruin his Arbeitsbuch (in Russia your work record is an official document just like in China). The guy demandes his Arbeitsbuch back and resigned the same day. People say he joined the biggest clickfraud collective at the time just to say f.u.
Coincidentally, It was right around early 2015 when conversion rates dropped to below 2% on many Eastern European RTB marketplaces.
My source, people at Vizavi HR, the only recruiting company working for Google in Russia
Oh look, another reason why a web form has to have javascript to work. It's for security, doncha know!
https://xkcd.com/1512/
1. This is not AI. This is a straightforward statistical analysis. AI is the marketing buzzword they're using.
2. If less than 5% of the population tested are identity thieves, a 95% accuracy means it will block more innocent people than thieves. Their accuracy needs to be well above 99% to be useful, since identity thieves are probably quite a bit rarer than that.
How does this help? You could simply supply a stream of mouse events to the measuring program that will pass its filters, case closed. Probably what it looks for is how often you hesitate and change your mind on an item.
I personally alwats create a full set if information for every online persona I use. Name, First name, Initial, Sex, DOB, POB, Address, Zip, Phone.
I have rarely had to work at this, all personas are arbitrary names, sex, birthdate/birthplace. I have found lately that some do check whether there is such an address but that's about it.
"Give way"
Building a better mousetrap means one is left with better (smarter) mice. This will teach criminals to invest a little time in creating a fake persona.
someone calling you while filling the questionnaire would also mark you as an id thief.
using a touchscreen would do that.
answering a message in the middle of filling it would also mark you as an id thief.
the whole point of the study is this: if you need to look up something you're likely to not be able to answer as quick and straightforward.
it's the kind of a study you can make a hypothesis for easily and verify it easily and provide something that can be presented like it is a solution possibly with further study for a real life problem. perfect academic research in other words.
and let me guess.. they gave another group sheets of info they had to look the stuff from and another told just to answer as themselves.. yeay.
world was created 5 seconds before this post as it is.
Interesting but not novel or new. For example typing patterns have been used before, this is an established biometric measurement called typing or keystroke dynamics . There are even companies that sell typing metrics as an authentication factor.
There is a standard test of detection performance claims that screams here. What is the probability of incorrectly rejecting the null hypothesis? The dimensionality of the space (mouse movements) is probably pretty large, so with a small sample size (40) and no separation of training and test sets, I am surprised they didn't get 100% detection performance. And it is meaningless.