What you will investigate
- How sensors provide percepts about the current environment.
- How TELL adds new information to an agent's knowledge base.
- How ASK uses stored facts and rules to determine what the agent should do.
- Why an agent needs relevant knowledge for its task, but not a complete representation of the entire plant.
This is a synthetic teaching scenario. It does not represent Umeå Energi's actual equipment, operating thresholds or safety procedures.
1 · The agent architecture
Follow the same flow as the architecture in the lecture slides. Work from the environment towards the agent's decision and action.
1Environment and sensors
The plant has a boiler, pressure sensor and temperature sensor. Change the conditions to simulate what the sensors detect.
2What the world is like now
This is the percept received from the sensors. The agent does not automatically know every detail of the plant.
3What action should I do now?
The agent can only use the information stored in its knowledge base. First TELL it about the percept; then ASK it for a recommendation.
2 · The agent's knowledge base
The knowledge base contains general facts about the plant and rules that connect conditions to possible actions. The representation is deliberately shown in ordinary language.
IF pressure is high AND the sensor is unreliable, THEN verify the pressure reading.
IF pressure is high AND an independent sensor confirms it, THEN escalate for operator assessment.
IF pressure is high AND the sensor is reliable AND there is no independent confirmation, THEN inspect the pressure condition.
IF pressure is normal, THEN continue monitoring.
IF temperature is high, THEN inspect the temperature condition.
3 · Agent decision and explanation
After ASK, the agent displays the action selected by its rules and the facts supporting that action.
Generate a percept, TELL the agent, and then ask what it should do.
Decision trace
4 · Try these experiments
Change the environmental conditions, generate a percept, press TELL, and then ASK. Compare the resulting actions.
High pressure · Sensor unreliable · No confirmation
Expected behaviour: verify the pressure reading.
High pressure · Independent confirmation available
Expected behaviour: escalate for operator assessment.
High pressure · Sensor reliable · No confirmation
Expected behaviour: inspect the pressure condition.
Normal pressure · Normal temperature
Expected behaviour: continue monitoring.
5 · Does the agent need to know everything?
Imagine that the plant contains hundreds of components, but the agent's current task is only to decide what to do about a high-pressure observation from Boiler B1.
Does it need a complete representation of the entire plant before it can make this decision?
Connecting to the next topic: Ontologies
Our agent knows that Sensor S1 monitors Boiler B1. But what exactly is a sensor? What makes a pressure sensor different from a temperature sensor? Can we define that a PressureSensor is a type of Sensor?
Next lecture: Ontologies provide an explicit model of concepts, classes, relationships and their meaning. We will learn how to represent this knowledge more formally.