Human-computer interaction (HCI) has a big problem: humans have personalities and emotions; computers don't. So how does the computer side analyze and respond to humans expressing emotions?
Personality is like climate: the overall environment over time. Emotion is like weather: it is what we experience at a particular moment.
People of almost all personalities may experience, say, anger. But some personalities are more prone to anger than others. How does an AI tease out the difference?
In a recent paper PersEmoN: A Deep Network for Joint Analysis of Apparent Personality, Emotion and Their Relationship researchers from the University of Illinois and A*STAR Singapore, describe an AI system they've developed that seeks to automate evaluation of apparent personality. Since bots can only "see" through video, any such analysis is based on video of the subject - which could be you.
Where might such systems be useful? Human-computer interfaces, bots, adaptive marketing and advertising, adaptive tutoring systems, or psychotherapy.
The proposed system consists of two convolutional neural network (CNN) branches:
. . . which we call emotion network and personality network, respectively. Emotion network and personality network share their bottom feature extraction module and are optimized within a multi-task learning framework. An adversarial-like loss function is further employed to promote representation coherence between heterogeneous dataset sources.
The system relies on earlier work in facial, emotional, and personality analysis, but the researchers found that combining the three was more accurate than using the methods singly. More importantly, they found that state-of-the-art face recognition networks work well for both personality and emotional analysis.
A key problem for all CNNs is training. Typically a large, annotated data set is required for training, and the larger the better. They used the Aff-Wild dataset for emotions, and the ChaLearn personality dataset, for training both CNN subsystems. The two comprise about 72 hours of video.
After testing, the authors conclude:
We find that the proposed joint training of both emotion and personality networks can lead to a more generalizable representation for both tasks.
The Storage Bits take
My father was a psychiatrist, and he'd find the potential application of this technology to therapy promising. But, let's face it, marketing and advertising is where the money is, so that is where this technology will go.
Imagine a "free" video chat service that captures your video and uses it to psychographically design marketing and advertising strategies to sell you stuff, like pharmaceuticals, to "fix" your emotional flaws. Or that plays on your insecurities to find the most irresistible products to pitch. Advertisers are!
Researchers are aware of these possibilities, and are pushing for responsible use of this technology. But when money meets ethics, all too often ethics lose.
Yet the potential of AI-assisted psychotherapy is real. There are simply not enough therapists to assist all the hurting people in the world. If that pans out, it will be good for all of us.
RELATED AND PREVIOUS COVERAGE
CTO Shawn Edwards says strong AI capability helps the firm build data-led products for its customers.
At re:Invent, AWS CEO Andy Jassy introduced personalization and forecasting services for cloud customers that rely on Amazon.com technology but require no machine learning expertise
According to IBM, its suite of AI tools can help HR become a growth engine to enable digital transformation.
In 2019, firms will recognize the reality of AI -- what is and isn't, what it can and cannot do -- and focus on its pragmatic side.
Interview: The chess grandmaster famously beaten by a computer says intelligent machines will make us smarter.
Courteous comments welcome, of course.