Emotion-detecting software isn't new--Sony, for example, recently filed a patent for a "laughter detection" system--but a San Francisco-based startup called Affective Interfaces, launched at TechCrunch50, claims that it has developed an ultra-accurate system that can detect emotions correctly over 85% of the time.
The AI system, which is based on proprietary algorithms, uses Webcams to monitor facial expressions like smiles, frowns, yawns, etc. The data can be tracked in real-time as users watch video in order to pinpoint the exact moment that, say, a commercial triggers laughter or disgust. "If you're looking for a dramatic moment in an ad, you want to know if it's correlating with audience response," explained Jai Haissman, AI's chief executive.
Initially, AI is gearing its system towards ad agencies that could potentially save costs by removing human monitors from the equation. And while limited number of participants can take part in traditional focus groups, AI's system allows hundreds or even thousands of users to join in.
The AI platform has uses that extend far beyond the advertising space. It could, according to Haissman, be used in everything from drowsiness detection in vehicles and dynamic video gaming engines to mood-based iTunes playlists and educational tools.
And what about concerns that the technology could be used for Big Brother-style spying? AI anonymizes all shared data and scrubs it of identifying information, but Haissman acknowledges that the platform could fall into the wrong hands. "Anything of power can be used for bad and good. We intend to use this for good," he said.
AI still has a long way to go before moving into the mainstream. The company has bootstrapped to date, and is just now considering outreach to investors. But if the system is as accurate as Hassman claims, the company should have no trouble attracting data-hungry advertisers.
Related Stories: | Topics:Innovation, Technology, Ethonomics, emotion detection, affective interfaces, TechCrunch50, tc50, startup, Emotions, sony, Jai Hassman, Jai Haissman, San Francisco, Sony Corporation, Apple iTunes |
Email
Digg
StumbleUpon
Facebook
Buzz Up!
LinkedIn
Recent Comments | 4 Total
September 16, 2009 at 2:41pm by Software Developer
Many interesting applications for video.
September 16, 2009 at 3:03pm by Ryan Bruins
Since when did 85% constitute "an ultra-accurate system"? :-S
September 21, 2009 at 10:14am by Marco Hout
I think the application that AI developed is very interesting, though indeed not new (see: FaceReader). I am also surprised by the "85% = ultra accurate" statement, but I do think it can provide a good bit of insight in emotional experience of a product/video/etc. Nevertheless, what applications like this do not measure are the more subtle emotions like 'shame', 'interest' or 'boredom'. At SusaGroup we therefore prefer to work with self-report methods (PrEmo, LEMtool, PanorEmo, LEM-emotions) that do give insight in these type of emotions. A combination of the two methods would be the ideal I think, but not that easy to implement. The face-reading methods make it easier for participants, but the non-verbal self report methods we work with make it more insightful for researchers.
www.susagroup.com
September 21, 2009 at 9:26pm by Jai Haissman
Hi, Jai from Affective Interfaces here! Just wanted to clarify a few things, regarding accuracy: the software exceeds human accuracies in any but the most well trained experts.The inter rater reliability with human experts is over 85% as well, which corresponds to the accuracies based on University training sets. Human accuracy is quite variable and often wrong when untrained. Our accuracies are quite extraordinary and more than adequate for market applications.
Marco, interesting approach! Of course there are no established universals in identifying shame or interest or boredom from behavioral cues, although there appear to be some interesting indicators which we do track. We actually do track engagement and drowsiness, distress (pain) and positive versus negative emotion in addition to the universal emotions of joy, fear, surprise, anger, disgust, neutral. Would be interested to know what your reliability is for shame and interest, and what testing system / benchmarking you are employing. Regarding Facereader, we are quite a different product in means of capture, parameters measured, and quality and insight in reporting.
It is quite easy to implement our system given it only requires a webcam!
Amongst the several benefits in tracking identified human universal indicators of market relevant emotion, we are able to massively scale samples from anywhere in the world at a cost savings and provide a frame by frame analysis of how your customers really feel about your product. Of course it is quite easy to gather surveys at the same time. It's quite easy to initiate a study just contact us at our site!
The trouble with research contexts in general is it introduces testing context bias. Self report methods contaminate direct emotional response with conscious mediation. Business needs to know what customers think, but really needs to know, reliably, objectively, how they feel. That's the value of AI.
www.AffectiveInterfaces.com