On Wednesday, Sophia, an AI-robot, gave a question-and-answer/interview presentation at the Future Investment Institute in Saudi Arabia. Sophia was also given the first Saudi citizenship for a robot.
What follows is a partial nonverbal analysis of Sophia.
During 0:19 - 0:20, when Sophia is "smiling", note her facial asymmetry. Her cheek dimpling is more prominent on her right side, while the left corner of her mouth elevates greater than her right corner. Asymmetry in humans, is in general, a signal of insincerity.
Another characteristic which makes Sophia's smile insincere (if she was human) is her eyes. The eyes are by far the most important component of a smile. The eyelids should ALWAYS partially close during a true smile (Duchenne smile) - with temporary, simultaneous, concave-up furrows forming in each lower eyelid. Sophia lacks these lower eyelid changes.
Although her eyelids are capable of closing during blinking (note that Sophia blinks fairly well, e.g., during 0:17, captured in the image immediately above) - the partial closing of the lower eyelids during a sincere smile are both anatomically and in appearance, distinctively different.
With respect to Sophia's blinking, she does so much slower (normal, non-anxiety related blinking in humans is extremely fast), and at a much lower frequency compared with that of a normal healthy, non-anxiety human (females blink slightly more frequently than males).
During 1:05 - 1:06, Sophia made an expression what she said was consistent with "if something has upset me". For a robotic expression, this is fairly good - particularly with her mouth. The corners are pulling down and laterally - which is highly indicative of emotional pain/sadness/grief.
Although Sophia's central forehead does contract and elevate along with the inner (medial) portions of her eyebrows - it doesn't do so enough. This facial dynamic would have a much stronger effect if her central forehead elevated even further - and if she also expressed simultaneous, evanescent furrows in her central forehead.
A few seconds earlier, Sophia expressed her "Angry" face (during 1:00). The camera was not zoomed in during this moment, thus this is a low resolution image.
From a nonverbal perspective, anger is an interesting emotion - for when it's expressed at low to mid-levels, both the palpebral fissures (distance between the eyelid margins) and the mouth opening become narrowed - but when anger is expressed at higher levels (e.g., rage), both the eyelid margins and the mouth opening widen dramatically.
Although the image immediately above is of low-resolution, Sophia appears to be expressing an anger level which is significantly elevated - closer to the rage end of the spectrum (with both her eyelids and mouth opened widely). It's difficult to see her mid-face. If we could visualize it well, however, we would expect to see a tightened region immediately beneath her nose along with flared nostrils.
As robots and AI improve, we will see much greater nuance - and in this case, that would include the ability to express lower and mid-levels of anger.
In this example, Andrew Ross Sorkin says, "I think we all wanna believe you, but we also want to prevent a bad future", she responds, "You've been reading too much Elon Musk and watching too many Hollywood movies. Don' worry, if you're nice to me, I'll be nice to you. Treat me as a smart input-output system" - she then displays a false smile (Intriguingly, one of her least sincere of this interview) including a large display of her lower teeth (3:25).
If a smile is sincere, it should not reveal the bottom teeth (Exceptions here include if her head were tilted down, if the camera/viewer was significantly elevated/taller and was angled/looking downward, or if she was just beginning to laugh or finishing laughter [full, sincere laughter does expose both the lower and upper teeth]).
This is an example of a feigned or "Social Smile". If Sophia were a real person, we would feel she was trying a to be social/friendly, and yet she's not really "feeling it" - she's not "in the moment" - she's not being sincere. Again, here her mouth corners are pulling primarily laterally. If it were a sincere smile of joy-happiness (Duchenne smile), there would be a mostly upward movement of her mouth margins.
During 3:18, just after she says, "... too many Hollywood movies...", Sophia turns her head back to straight ahead, and a bit to her right (she had her head turned to her left). Before her head rotates right, her eyes rotate right. With moderate and rapid head turning this is normal human physiology - but not if we rotate our heads' at slower rates. Most people are unaware of this phenomenon (whether in another person or in themselves) - but it doesn't feel natural when it's missing. With the example shown here, however, this eye movement would not have occurred and her eyes would not have rotated all the way to her right with this corresponding relatively slow speed of head rotation - but it would have with a higher rotational head speed. Although not yet perfected, this is a relatively nuanced motion and a sophisticated dynamic for Sophia to be displaying.
Beginning at 3:45, as she says, "By the way, if you are interested in giving me an investment check, please meet me after the session" - Sophia tilts her head. In human behavior, such head tilting at the beginning of an "ask", projects higher sincerity and empathy. Moreover, it will also engender a significantly higher success rate than if the head is not tilted.
Summary: Sophia represents a significant advancement in AI and robotic-mimicked human behavior. The level of nuance and sophistication is impressive. Sophia 2.0 will no doubt display continued improvement.
It's worth emphasizing that during 3:05 - 3:07, Sophia says, "... I strive to become an empathetic robot ...". There are different types of empathy. While a robot/AI may develop or already possess some level of cognitive empathy — many people debate about whether it will become possible for “them” to feel emotional empathy. In our development of and our relations with such technology-beings, we must be careful to make the distinctions between these outward mimicked-emotional manifestations and what we perceive/project onto them as “AI feelings”.
Within 10-15 years, robotic mimicry of human facial expressions will be mistaken for real human expressions. The time will come when you will say to yourself, "Is that a robot or a real person?". This is not a matter of "If", only "When". People will be drawn to these robot/human substitutes because their "programmed empathy" will supersede a significant fraction of their fellow humans' empathy. Thus, if we're not careful, our own empathy shortcomings will draw us to our robotic progeny and facilitate our own demise. In Darwin's terms, robotic life forms will be selected for via our perceptions of their "programmed empathy".
Group Appearances and One-on-One
Online Courses Available
Body Language Analysis No. 4104: Walter Cronkite Reaction to JFK Assassination
Body Language Analysis No. 4102: Senator Bob Corker - Donald Trump is an Untruthful President
Body Language Analysis No. 4100: Bryan Cranston and The Eyelid Pull
Body Language Analysis No. 4098: President Trump and Puerto Rico's Governor Ricardo Rosselló - A Candid Moment in the Oval Office
Body Language Analysis No. 4097: Andre Agassi, Boris Becker, and a Tennis Tongue Tell
Body Language Analysis No. 4077: A Facial Expression Common to Both Bad Actors and Sociopaths
Body Language Analysis No. 4052: Hitler's Cryptorchidism and Emotional Dissonance
Body Language Analysis No. 4035: Hope Hicks, Jared Kushner, and Phone Tells
Body Language Analysis No. 3976: Bill Conner bicycles 1,400 miles to hear his daughter's beating heart again