Artificial intelligence (AI) is intelligence in a robot, computer program, or other non-living entity that makes decisions for it, typically towards completing a task.
Character AI also makes decisions, yet it results in something more like a character in a novel or a film: some combination of personality, emotional expression, social interaction, and being emotionally invested in outcomes. Character AI controls all of the characters in a video game aren't controlled by human players. And it controls electronic toy characters.
Normal character AI is hand coded. But bots_alive's AI is built algorithmically from observations of human control of the robots during development.
Machine learning creates what we call “models” of the human controller. These models take a bunch of descriptions of the robot's context (e.g., distances and angles to key objects) and output the probability that the human controller would have commanded the robot to move forward, left, right, ..., or to stay still. And any one model can control the robots on its own, emulating the control style of the human it was derived from.
We think this new form of character AI has the potential to create believably alive characters. It already has with these simple robots and in the preceding research our founder, Brad Knox, conducted at the MIT Media Lab. The technique captures the play, unpredictability, and social authenticity of the human improviser that teleoperated them during development.
These robots are the first product to demonstrate the power of this new approach to creating simple but lifelike characters.
You can read more at https://medium.com/@bradknox/building-character-ai-through-machine-learning-7a3159dc4940.