In Part 1 and 2 I established that when humans talk about intelligence, we are talking about human intelligence, and AI should be measured by this yard stick. I also discussed that studies of neural network and deep learning alone are inadequate, as they are analogous to mere circuit analysis of a large complex mainframe. A holistic, system level understanding of human intelligence is necessary to build truly intelligent, sociable robots. In Part 3 and 4 I reviewed key evidences in psychology and brain science that human brains have compartmentalization and layering, both suggesting a layered architecture with specialized subsystems. In Part 5 I delved into feelings and emotion and how they are part of the primal brain and tie in the evidences from psychology and neural science. In this blog post I will discuss a proposed primal layer architectures.
An organism’s intelligence starts with the ability to sense its surrounding. Simple organisms have relatively few sensors while complex organisms like humans have many. Some sensory stimuli are mapped relatively quickly to actions, such as reflexes, while others are registered internally in short term memories (STM) forming what neural scientists called sensory maps. The human visual cortex is an example of sensory map where the retinal images are registered. Other sensory maps include tactile, temperature, even internal organs.
Sensory receptors and sensory maps are critical elements of life regulation (homeostasis). They let the organism know when something is not functioning properly and is requiring attention, and urge the organism to pursue rewards and desires that improve its homeostatic condition. In humans, these signals come in as varying degree of pain and pleasure, and our cognitive and biochemical reactions form internal feelings expressed outwardly as emotions. These “biological” reactions seemed to be hardwired and universal within a specie. Though each specie may react differently to the same stimuli, all seem to strive towards the same basic goal of survival.
Could it be that at the most primal level, robots should be endowed with similar survival goals? Could it also be that higher form of intelligence, such as planning, are complex behaviors striving at the same fundamental goals? If intelligence is viewed as a layered framework, as Paul MacLean’s Triune Brain and Maslow’s Needs Pyramid suggested, then the most primal layer has to be responsible for basic survival and life regulation.
Below shows the block diagram of a proposed primal instinct layer roughly equivalent to the reptilian brain or the limbic brain. The Motor/Executive carries out actions as directed by Feeler to interact with the world. The sensory stimuli from the world are registered in Maps and some may evoke reflexes. Self-actions by Motor/Executive are also registered through proprioception. Life regulation requires homeostasis to define and generate pain and pleasure signals to the Feeler. The Feeler interprets the signals and produces actions that interact with the world to regulate homeostasis.
Some of the Feeler actions are emotion expressions. One may wonder what value do they add to life regulation? Consider emotion in a social context as a mean to communicate one’s internal state. Such communication has life regulation value. For instance, a hungry child’s obnoxious cry causing the mom to attend to the child. The child got what he needed, the mom not only promoted the survival of her offspring, but also alleviated her own pain and discomfort from the baby cry. Emotions is a key part of social interaction and interdependency.
If one is to make robots sociable with humans, one must endow robots with human-like emotions; not because that’s what robot requires to be social, but because humans. Emotions as expression of internal state could be very different between robots. The feeling behind emotions, at a primal level, is driven by life regulation, and is instinctual, fast responding, and largely autonomous. When designing social robots, primal layer is the foundation that one must get right, as higher level intelligence depends on it. In the future blog posts I will explore some implementation options to realize the primal layer, as well as architectural add-on to the primal layer to realize more complex behaviors. But first I will cover some preliminary topics, such as learning and memory, which is the subject of my next blog post.
- Musing on Intelligence–Part 1: What is it?
- Musing on Intelligence–Part 2: Intelligence require a System
- Musing on Intelligence–Part 3: Motivation
- Musing on Intelligence–Part 4: Brain Architecture
- Musing on Intelligence–Part 5: Emotion
(The above article is solely the expressed opinion of the author and does not necessarily reflect the position of his current and past associations)