The popular Disney movie “Inside Out” portrayed the various emotions insides a girl named Riley, joy, fear, disgust, sadness and anger, as five homunculus. They interact with one another and with Riley’s long-term memory to effect Riley’s behavior. For an animated cartoon, the psychology behind the movie is remarkably accurate. My previous post in this series discussed how human brain is functionally organized and hierarchical, and how the primitive limbic region of the brain is responsible for emotion and drive. In this post I will argue why emotion is a critical part of intelligent behavior.
Now, imagine a world with no joy of reward; no fear of dangers, no comprehension of significance or meaning of events, no preference or biases in choosing, no impulse or motivation. All stimuli having the same importance (or lack of). Sounds like a very boring world? Now imagine a person without any emotion or opinions; that person will likely become an outcast, hard for anyone to develop an emotional attachment with. Yet that’s how we are designing robots; and we wonder why they aren’t intelligent like us. To make robots with human-like intelligence, they must possess emotion and driven by needs.
Recall Maslow’s Hierarchy of Needs: physiological, safety, belongingness, esteem and self actualization? He postulated five fundamental human needs. The first two are physical needs related to survival, the latter three are social needs, which can be interpreted as higher forms of survival. Being in a group offers protection and safety; and belongingness drives one to become part of a group. Once in a group, the need for esteem can be interpreted as a desire to elevate ones stature within the group as in almost all societies esteemed members are bestowed with special privileges or reward that will further meet their needs. After survival is assured, one now has the luxury of asking whether ones full potential is reached–this can be interpreted as self-actualization.
What is the relationship between emotions and needs? Emotion is an external expression of internal state of how well needs are being met. For instance, if most needs are met, joy or contentment is expressed; if safety is threatened, fear (flight) and anger (fight) may surface; if belongingness is lost, sadness may follow; if expected social behavior is not reciprocated or perceived, disgust may ensue. What if robots are endowed with these needs and emotions? I submit that they will act more intelligently; coupled that with learning, robots will become increasingly intelligent (in the eyes of humans) as they gain more experience. So how do we implement needs and emotions? As psychologists postulated that all human needs, even social ones, tie back to survival; and human behavior is driven by these needs. Therefore, it is reasonable to consider the plausibility that robots with human-like intelligence should be endowed with hard-wired needs and desires; and the robot behavior becomes action or response to mediate unmet needs.
Biological systems have operating envelopes called homeostasis that defines the range within which the organism will thrive. For robots and machines, homeostasis is equivalent to the operating specification in data sheets and user manual. When homeostatic condition is violated, the organism responds with action that will bring the operating point back to homeostasis. The response can be autonomous, such as sweating when body gets hot, or voluntary, such as removing layer of clothing. We know that in general voluntary response are learned as adults are much better at taking care of themselves than children. Similarly, robot can be programmed with “homeostasis” rules to respond whenever the operating condition deviates from specification. When endowed with learning ability, robots can expect to become more proficient at taking care of themselves as experience builds.
What does this all mean?
Now let me summarize the main points of Part 1-5 of this series:
- When humans refer to intelligence, we are really talking about human-like intelligence
- Human-like AI needs to take on a holistic approach by developing intelligence at a system level, not just at the building blocks level, like neural network and Deep Learning.
- To mimic human intelligence, one must review the work done in psychology, neural science and brain studies that reveal clues about human intelligence at both macroscopic and microscopic levels.
- Psychologists postulated that needs, including social needs, at some level tie back to survival, and emotion is an expression of the internal state of needs (feelings). Unmet needs create drives (motivation) that leads to mediating action (behavior). Basic needs, together with the ability to learn and adapt, creates intelligence that can grow with experience (maturity).
- Brain scientists learned that human brain has functional and hierarchical organization; and the inner brain (limbic and reptilian brain) is related to basic survival, and outer brain (neocortex) is related to high level functions, abstractions and planning. The layering of these regions suggests an evolutionary path of human brain from reptilian to mammal to human.
- Given the above, it should be plausible that robots programmed with basic needs and ability to learn can have growing intelligence nurtured by interaction with humans and the surrounding environment.
How do we make such a robot? In the next post (Part 6), I will discuss proposed intelligent system architectures that build on the above foundation.
- Could robots have human emotions?
- Musing on Intelligence–Part 1
- Musing on Intelligence–Part 2
- Musing on Intelligence–Part 3
- Musing on Intelligence–Part 4
(The above article is solely the expressed opinion of the author and does not necessarily reflect the position of his current and past associations)