Can Intelligent Agents Learn Empathy? [Theory of Mind AI: Explained]

Do Androids Dream of Electric Sheep?

We’ve seen this in movies: robots making a sacrificial choice for the sake of humanity. AIs exhibiting feelings such as love and compassion. A fictional Pinocchio wanting to become human so he can fulfill her human mother’s desires…

In books, one such title by Philip Dick, even have a made-up empathy test so it can properly distinguish humans from androids.

They are originally works of pure imagination, but scientists are driven to make this kind of fantasy turn into reality regarding AI robots (Sophia’s life-like creation). After all, as Isaac Asimov once said: “Today’s science fiction is tomorrow’s science fact.

So, can robots with artificial intelligence learn empathy?

Empathy Defined

First and foremost, let’s define empathy.

Having empathy means that you have the ability to realize and understand the emotional and cognitive and/or spiritual state of others, without you directly experiencing the triggers for that state. It’s putting yourself in the perspective of another person. As the saying goes: putting one’s foot in another’s shoe.

In order for you to have this capability, you have to be aware of other’s mental, social and emotional needs. Usually, one can achieve this by simply observing visual and auditory cues, and build from that understanding of what is good or bad for other people. Humans experience this naturally as they grow up with little or heavy guidance from their parents or parent-figures. As young as 4-year olds can show empathy towards other people, animals and even objects.

Visual Behavior Modelling

A robotics lab in New York had already done such experiment. Columbia University’s robotics lab, called the Creative Machines Lab, wanted to find out if AI robots can anticipate the behavior of other robots, just through observation (using camera and microphone) and deprive them of any symbolic reasoning.

They made a first robot look for green circles and go near them. But they put obstacles in some green circles, hidden from it. Then a 2nd robot, called the observer AI, observed the behavior of the 1st robot for 2 hours, without hiding any green circles from the 2nd robot.

The result was that the observer AI successfully predicted 98.5% of the outcome just by installing simple programmed visual sensors in it, without it having any symbolic reasoning (like if-then conditional rules set by its creator).

The experiment suggests that through seeing alone, an agent or entity can anticipate and predict behaviors primitively, i.e., without verbal cues and only through visual cues. This implies that simply-made robots with artificial intelligence can have a sense of awareness towards its surroundings, though in a primitive sense.

Click for more information about visual behavior modelling.

Other Robots Showing Empathy

There are other more sophisticatedly-built AI robots (Sophia is an example), but there are those which can detect emotions and respond “empathically” with people. Pepper, Paro, and Kaspar are such examples.

Pepper’s mechanism is to read emotional expressions by detecting the voice tones of a person and then respond appropriately to “console” the person. Pepper is considered a humanoid.

Paro, on the other hand, takes the form of a baby seal and is preprogrammed to mimic therapy animals, which are animals trained to comfort sick people. Paro, designed in Japan, is a Guinness world record holder for being the world’s most therapeutic robot.

Meanwhile, Kaspar is a child-like robot, termed social AI robots, made for children with autism. With a simple, non-assuming face and predictive reactions, socially-challenged children find Kaspar easier to talk to. Kaspar is a small time project, but it, being successful in its objective, has become a somewhat staple choice in helping autistic children interact with other people.

These robots with artificial intelligence can indeed express empathy, though they are not conscious or aware about having the ability to be empathic. They don’t attribute their own perceptions to social cues but are merely programmed to respond based on preformed reasoning.

Theory of Mind

The theory of mind is a theory in psychology wherein one person puts a personal meaning behind the behavior, statements, and expressions of another individual because these external factors are what is merely observed by the person. It also considers and understands that because of having this attributed meaning towards another, other people also have different beliefs and motives towards other people.

In the experiment, the observer AI can predict the behavior of its fellow robot and mimic the choice its fellow robot has to make, with almost 100% accuracy. This form of awareness, by anticipating the outcome of a behavior, is helpful in confrontation and avoidance mechanisms we humans do to sustain a harmonious relationship with other people.

By anticipating the right motive of the individual through external cues, the person can have agreeable social interaction with the individual.By realizing that other beings also have varying motives and beliefs than oneself, we would understand even a hint of what the other is feeling. And with just anticipating the behavior and not doing or experiencing the behavior, we can achieve cognitive empathy.

Wrap Up

AI robots (Sophia is an example) have been helpful in consoling humans with mentally and socially challenging lives. Kaspar, Paro and Pepper are such examples of social AI robots.

The recent experiment made by Creative Machine’s Lab tells us that robots with simple, unsophisticated systems can predict the outcome of other intelligent agents, with almost perfect accuracy to boot. With no symbolic reasoning to consider, their robot was able to predict the chosen path of another robot by merely visual observation.

This means that a robot can be aware of a behavior, just by observing for a period, and can predict such behavior. It is believed, then, to exhibit a theory of mind (AI).

ToM, or theory of mind (AI), is a psychological theory were one attributes mental notions for external factors, and understand that others also have different mental notions apart from theirs. ToM is an important marker for having empathy.

The robot showed more than just the monkey see, monkey do type of outcome, rather it is akin to knowing by seeing. It is considered artificially intelligent because of its virtually perfect prediction, with minimal human conditioning. We also do not fully understand how it came to make such accurate predictions.

So, can intelligent agents learn empathy? Well, a robot was shown infantile capability, through manifesting theory of mind (AI). But if in accepted notions, definitely not today. So fear not, AI-phobics.


You Might Like to Read:

Leave a Reply

Your email address will not be published. Required fields are marked *