As in the other texts we dived into, the author gives a great space to defining the terms at the center of the research, the semantics seems to be at the core of the works. This made a lot of sense for the algorithm, but it was all the more striking to discover through this week’s reading Figuring the Human in AI and Robotics how broad and even controversial it can be to define words that felt common and obvious. What does it mean to be “humanlike”? How do you define a human from a non human? Why is it important to know from which point of view they are being defined?
After opening to different ways of thinking, the author suggests a classification of three elements that could lead to address the experiences of the machines as “Humanlike”:
Embodiment
Emotion
Sociality
One of the exemple that aroused my curiosity, that would fall in the Emotion part, was that of the use of AI as a psychologist.
As for the robots described in the text, Cog and Kismet, that are trained to read emotions, after them being thoroughly analysed and classified, Ellie was developed as part of SimSensei project by the USC Institute for Creative Technologies. It is a robot, embodied in an (cartoon) identifiable human, that is able to read emotions through sensory mapping of the face and body. It detects patterns, and reacts accordingly, for example, with a smile at the right moment, to make a connection with the “patient”, by noding, engaging in small talk etc.
Sceenshots from a video showing the way Ellie/SimSensei works and interacts with a patient
Not having access to this kind of experiment, I decided to explore this week’s research through a live session with what is presented as an accessible, largely used AI therapist: Weabot. It’s an application developed by Stanford University researchers that markets itself as providing “Compassionate solutions for human problems” and puts the communication focus on “heart” and feeling like “a friend”.
Woebot's presentation on its website https://woebothealth.com/
The first session with Woebot on the Woebot App
Although I don’t have any previous experience in human therapy, and I wasn’t actually a candidate for it, I was glad I didn’t count on it for any health conditions.
It appears that the democratized version of therapist is far from competing with a human exchange. The empathy is overstated, the exchange is out of the hands of the user, who is part of a very directed “discussion”. It felt closer to a kind of motivational coach, at best.
They did not make the choice of a humanlike bot, but of an almost cliche version of a robot.