AIRIS: Adaptive AI for Evolving Systems

AIRIS (Autonomous Intelligent Reinforcement Inferred Symbolism) is an AI system developed to understand and interact with its surroundings by observing and analysing changes it causes. Rather than being preloaded with a rigid set of instructions, AIRIS learns from each experience it has, building its own set of rules based on what it observes. These rules help AIRIS create a flexible model of its environment that it uses to make decisions, test predictions, and update its understanding.

How does AIRIS work?

AIRIS observes its surroundings, forms predictive rules from each action, and refines them via a confidence score. This adaptive approach enables efficient navigation and goal achievement.

Environment Observation

AIRIS starts by taking in information about its surroundings. Imagine it in a simple game where it controls a robot that can move in four directions on a grid. AIRIS sees this environment as a grid of numbers, with each number representing objects like walls, open spaces, or targets.

Taking Action and Noticing Changes

AIRIS begins by randomly trying different actions, like moving left. When it moves left, AIRIS notices which cells on the grid change values. For example, the cell the robot leaves becomes empty, while the cell it moves to now contains the robot. From these observations, AIRIS creates “rules” about what happens when it moves left.

Creating Rules and Predictions

After observing the outcome of an action, AIRIS creates rules to predict future actions. For example, one rule might be: “If there is an empty cell to the left of the robot, then moving left will shift the robot to that cell.” These rules help AIRIS predict what will happen next time it tries moving left. AIRIS continuously refines these rules as it gains more experience.

Confidence Levels in Decision-Making

Each rule has a “confidence” level, a score between 0 and 1 that indicates how certain AIRIS is that the rule will work based on past experiences. If a rule has a 100% confidence level, AIRIS is certain of its prediction. Lower confidence levels mean AIRIS is less certain and might choose another action or update the rule as it gains more information.

Continuous Learning and Adapting

AIRIS doesn’t stop learning after it makes its initial rules. Each time it acts, AIRIS checks whether the outcome matches its prediction. If it does, the rule’s confidence increases; if not, AIRIS creates a new rule to better predict similar situations. This ongoing learning allows AIRIS to adapt quickly to new environments or unexpected changes.

Setting and Achieving Goals

Once AIRIS understands how to navigate its environment, it can start pursuing specific goals. In a game, for instance, this might mean collecting items scattered on the grid. AIRIS plans out a series of actions to reach these items efficiently. If it encounters obstacles like walls or doors, it applies its rules to navigate around them, updating its understanding as it goes.