News & Analysis
Comment
SlickCar
This has been in the works for some time. It is only a matter of time before we ...
DarkMatter
Ah, the applications to social engineering are virtually unlimited.
Kiss your unfeeling computer goodbye
Junko Yoshida
8/21/2012 10:29 PM EDT
Founding of startup Affectiva
The affective computing research carried out by Picard’s team at MIT has not been locked up in the ivory tower. Picard’s team founded an MIT-spin-off called Affectiva in 2009. The startup has successfully commercialized emotion technologies, including Affedex, an automated facial coding platform, and Qsensor, a wearable biometric sensor.
Although Picard noted that making money in the commercial market had never been her goal, she said that having been inundated by so many inquiries from the medical community compelled her to start Affectiva. “If I didn’t help them, I felt as though I were holding up the medical research.”
Picard’s quest to make a computer that understands emotions has also found a partnership with researchers studying autistic children. Autistic kids are known for their inability to read other people’s emotions. They can’t seem to put facial expressions, vocal affects or physiological changes into context. As Picard’s team worked with people with autism, Picard said, “We discovered the autistic people showed totally off-the-chart emotional responses when they were agitated.” She added, “Even scarier was that the sensor recorded an enormous emotional peak just before the autistic person had a seizure.”
Picard and her team ended up teaching caregivers how to deal with autistic people by looking for signs, instead of teaching autistic people how to understand other people.
Emotion prediction
Picard believes that the automated facial coding platform developed by Affectiva works quite well. “We have the accuracy rate in the range of the upper 90 percent” when it comes to its pattern recognition, she said. “Our tools have been trained and tested not only in the United States but also in China and Brazil.”
Much more challenging, though, is the development of emotion prediction, Picard said.
Emotion is complicated. Humans can be angry or afraid, before the signals even get to the cortex, and before they become aware of what’s happening to them. Picard likes the example of people’s reaction to the sight of a snake. “Sometimes, we are already behaving, such as jumping out of the way of danger when we see a snake, before we become aware of an emotion such as fear,” Picard said.
In that sense, “Reading emotions is equivalent to forecasting the weather. You need to know how the weather changes as you look at the weather through a window,” she explained.
Asked about affective computing’s impact on hardware, Picard said, “As the algorithms [for computers to read emotions] get more complex, they require computing machines that can speed up pattern recognition.”
While improving processing performance, doing it in an embedded system with cost and power constraints will be even more challenging, noted Bier. He said that, to have any practical use, computer vision systems need to fit into equipment that is both smaller and less inexpensive.
DESIGN East, scheduled at September 17 – 20 in Boston, combines the Android Summit, DesignMED, the LED Summit and Sensors in Design events with UBM's Embedded Systems Conference (ESC) Boston. DESIGN East is also co-located with the Embedded Vision Summit, a technical education forum for engineers centered on embedded vision technology applications.
Related URLs:
Design West Keynote
Design West Registration
Embedded Vision Summit
Is vision the next-gen must-have user interface?
The affective computing research carried out by Picard’s team at MIT has not been locked up in the ivory tower. Picard’s team founded an MIT-spin-off called Affectiva in 2009. The startup has successfully commercialized emotion technologies, including Affedex, an automated facial coding platform, and Qsensor, a wearable biometric sensor.
Although Picard noted that making money in the commercial market had never been her goal, she said that having been inundated by so many inquiries from the medical community compelled her to start Affectiva. “If I didn’t help them, I felt as though I were holding up the medical research.”
Picard’s quest to make a computer that understands emotions has also found a partnership with researchers studying autistic children. Autistic kids are known for their inability to read other people’s emotions. They can’t seem to put facial expressions, vocal affects or physiological changes into context. As Picard’s team worked with people with autism, Picard said, “We discovered the autistic people showed totally off-the-chart emotional responses when they were agitated.” She added, “Even scarier was that the sensor recorded an enormous emotional peak just before the autistic person had a seizure.”
Picard and her team ended up teaching caregivers how to deal with autistic people by looking for signs, instead of teaching autistic people how to understand other people.
Emotion prediction
Picard believes that the automated facial coding platform developed by Affectiva works quite well. “We have the accuracy rate in the range of the upper 90 percent” when it comes to its pattern recognition, she said. “Our tools have been trained and tested not only in the United States but also in China and Brazil.”
Much more challenging, though, is the development of emotion prediction, Picard said.
Emotion is complicated. Humans can be angry or afraid, before the signals even get to the cortex, and before they become aware of what’s happening to them. Picard likes the example of people’s reaction to the sight of a snake. “Sometimes, we are already behaving, such as jumping out of the way of danger when we see a snake, before we become aware of an emotion such as fear,” Picard said.
In that sense, “Reading emotions is equivalent to forecasting the weather. You need to know how the weather changes as you look at the weather through a window,” she explained.
Asked about affective computing’s impact on hardware, Picard said, “As the algorithms [for computers to read emotions] get more complex, they require computing machines that can speed up pattern recognition.”
While improving processing performance, doing it in an embedded system with cost and power constraints will be even more challenging, noted Bier. He said that, to have any practical use, computer vision systems need to fit into equipment that is both smaller and less inexpensive.
DESIGN East, scheduled at September 17 – 20 in Boston, combines the Android Summit, DesignMED, the LED Summit and Sensors in Design events with UBM's Embedded Systems Conference (ESC) Boston. DESIGN East is also co-located with the Embedded Vision Summit, a technical education forum for engineers centered on embedded vision technology applications.
Related URLs:
Design West Keynote
Design West Registration
Embedded Vision Summit
Is vision the next-gen must-have user interface?
|
|
|
|
|
Navigate to related information
JeffCB
8/21/2012 7:00 PM EDT
Professor Picard's presentation on September 19th in Boston will be a combined keynote presentation for DESIGN East and the Embedded Vision Summit -- a free educational event for engineers. For more information, and to reserve your seat, visit http://www.embedded-vision.com/embedded-vision-summit).
Sign in to Reply
DrQuine
8/21/2012 11:44 PM EDT
While emotion may not be a feeling experienced by computers, the root causes of human emotions (grief, joy) are typically associated with events with predictable implications. Computers could therefore be programmed to respond appropriately when data is received corresponding to significant events for a nearby human. Naturally, as with human-human interactions, there may be mistakes. It might not be obvious (to the computer or a fellow human) that news of a big promotion could be considered bad news by the recipient because it requires an unwanted relocation.
Sign in to Reply
nicolas.mokhoff
8/22/2012 10:21 AM EDT
Recognizing emotions is hard enough in human-human interactions. Coding pattern recognitions cannot possibly yield accurate "emotion quotients" for individuals. It's good that autisitc research has another tool to classify a large range of emotions. And, affective computing could do much to make people more effective. Applause to Dr. Picard's work; her keynote should make for a fascinating experience. Bravo to the events people for giving her stage time at Design EAST.
Sign in to Reply
prabhakar_deosthali
8/22/2012 1:15 AM EDT
This is a good news that the true humanoids can now be created who are not just efficient machines doing human chores but have feelings like humans , can express affection, love, hate and anger- the intrinsic part of the human life.
Sign in to Reply
kinnar
8/22/2012 7:56 AM EDT
Finding emotions using Video/Still Camera and Microphone is some what critical, as I remember when trying to get the actual voice in a sound installation from a speaker is very hard, most of the time due to change in the conditions it is very hard to find the real voice as an output, same thing happens with the Video/Still as well. So it will be a long travel to get successful results for this emotions detection.
Sign in to Reply
junko.yoshida
8/22/2012 10:19 AM EDT
It's quite true. There are so many variables in real-life conditions -- including bad lighting, movement in the background, etc. That said, I think it's remarkable when scientists and engineers find "building blocks" that identify subtle changes in humans' emotion.
Sign in to Reply
affectiva
8/22/2012 10:22 PM EDT
You're exactly right on one hand. In order to get successful results, the system has to be able to detect emotions in various environments, under any given condition (ex: blurry faces, bad lighting, outside interference). Our facial coding technology, Affdex, was designed to do just that by gathering spontaneous facial expressions (emotions occurring naturally in real-world environments). We continually evaluate the accuracy of current classifiers, while training and testing new classifiers. On the other hand, this isn’t a future idea, this is happening now. Affdex has already been validated in developed and emerging markets.
Sign in to Reply
ShopSAR
8/22/2012 3:04 PM EDT
Yes it is true with the different feelings someone people have. But is it very hard to exactly show it.
www.shopsar.com
Sign in to Reply
TFCSD
8/22/2012 8:04 PM EDT
Robots reading and having emotions is right up there with slapping lipstick on a pig. I believe KISS is the best and safest mode for a robot’s “personality” and having a robot “guess” what people are “feeling” and “guess” what to do next is a recipe for inconsistent robot operations and will cause more problems than it is worth.
Sign in to Reply
przemek
8/23/2012 11:08 AM EDT
The idea of including emotional cues in HMI is valid and I am sure that there are clever techniques to measure the emotions, like the one mentioned in the article, looking for skin color modulation. I looked at their demonstration video and couldn't see it with my own eyes but the digital sensors are more sensitive and also can enhance the effect by clever math.
At the same time, this is inexact science: polygraph testing is even more involved technologically but has been proven to be unreliable.
Sign in to Reply
Schatzy
8/23/2012 2:27 PM EDT
When a human is born, the sensors and detectors built into their processor (brain) are all functional, but yet the human computer only appears to have the capability to operate in autonomic mode (i.e. heartbeat).
The challenge for affective computing would likely be to figure out what the human model goes through to create things like "Moral Compass". I would assume that we all want our machines to properly adapt to needs in a way that is simlar to the human response- Most of our daily habits are developed from the initial impressions that were instilled into us from our beginning environment.
Interesting study! I will be watching to see how this goes in the future. Will I like the personality of the robots in the future? Who knows!
Sign in to Reply
jaybus
8/27/2012 8:25 AM EDT
Even if emotions are completely innate in humans, the response to other's emotions is almost certainly learned behavior. This is likely true for most mammals. Perhaps a shorter term goal for a machine would be to recognize emotion in a mouse or something a little less complex than a human.
Sign in to Reply
Luis Sanchez
8/28/2012 12:11 PM EDT
This is a small brick in the wall of research on artificial intelligence. We humans do identify the face blemish as emotions like shame, worry, extreme and sudden fear (our face turns white), or even being angry. Actually I’ve thought sometimes the mood of a person could become part of it’s social network information. In the not too far future, we won’t have to bother to update our Facebook to let our friends know how are we feeling. The app will do it for us, by taking a look at our face, hearing our voice, sensing our heart rate or even perhaps our temperature or how fast or slow we walk.
This is the era of the cognitive sciences. Interesting right?
Sign in to Reply
agk
9/20/2012 9:20 AM EDT
It is very difficult to under stand the real thoughts of human from the face.It is for the very simple reason at many instances we are compelled from our inside mind to smile or keep neutral or cry. So computers never will be able to detect the true feelings of a human by only reading the faces. May be by analyzing the brain waves,saliva,sweat and other outputs it may be possible.
Sign in to Reply
DarkMatter
9/20/2012 11:29 AM EDT
Ah, the applications to social engineering are virtually unlimited.
Sign in to Reply
SlickCar
10/9/2012 2:53 PM EDT
This has been in the works for some time. It is only a matter of time before we are all being driven around by emotional computers. "No need for friends I have my Ipad 9!" "She knows what I like and tells me what I want to hear."
Sign in to Reply