Artificial Personalities with Artificial Genomes
Rity is a small, cute, blue digital dog. Genderless, Rity is also the first of what is hoped will be a new breed of AI, an AI with emotions and with an emotional personality genome in a novel take on evolutionary algorithms. Rity's genome is composed of 14 chromosomes, each of which possesses three gene vectors: the fundamental gene vector, the internal-state-related gene vector, and the behaviour-related gene vector. As 2 fundamental -genes, 47 internal-state -genes, and 77 behaviour -genes represent each chromosome, Rity has 1,764 genes in total.
Created at the Robot Intelligence Technology lab in Keora, Rity has been developed as an emergent personality with predisposition based on that genome. The genome itself was not created manually, but rather seeded with values thanks to a series of evolutionary algorithms that mutated the base code over time, generation after generation, adding complexity as it went.
Depending on the values of the genes, the researchers specified five personalities and their opposites to classify an artificial creature's personality traits:
Thus the artificial chromosomes define the personality of the AI.
Added to that, the AI has a standard needs-based upper layer. That is to say, it balances conflicting needs at all times. Whichever one, or two are strongest in its mind at the time, determine its behaviour. As with everything else about Rity, their states are quite complex and intricate, and again partially determined by genetics.
Rity relies on its internal states which are composed of three units - motivation, homeostasis, and emotion. Each is a separate, although interacting needs based system. The three units have a total of 14 states, which are the basis of the 14 chromosomes.
They spread out like thus:
All of these states interact in real-time, inside Rity. To cap it all off, Rity learns and assimilates from past information and responds interactively to stimuli.
Rity lives in its own virtual environment, creating an embodied state, and interacts through mouse clicks, a camera, or a microphone. 47 input pathways were assembled for Rity, from multiple audio and optic streams, as well as objects in the VR that can stimulate the AI. All are processed through the medium of Rity's avatar, making the experiences in the VR; ideal training for a robotic embodied form down the road.
With the mouse, clicking on Rity is interpreted as patting; dragging the mouse slowly over Rity's form is stroking. Double clicking is a punch, and so on. All these different interface modalities give Rity a stream of stimulus, which it integrates and builds on to create habits and self-trained responses. They can also be used to train Rity, who has developed limited associated behaviour, much like you would train a puppy.
The approach to AI offered is certainly novel, and the complexity certainly lends itself well to immensely complex emergent behaviour. Time will tell if this method of evolving intelligence, and training the resultant seed, pays off.
J. -H. Kim, C. -H. Lee and K. -H. Lee, "Evolutionary
Generative Process for an Artificial Creature's Personality," (PDF)