Agent State Description
Active State Transition Process
Explanation of Various State Modules
Perspect
In this state, an agent's tasks include:
Analyzing observations of the external world, such as discovering it is in a restaurant and analyzing the layout, menu, and other elements.
Analyzing its internal state, including feelings of fullness (whether it is hungry), health level, energy level, and mood.
Analyzing the consistency between its current behavior and plans.
Current date
Current location
Observations of the world, including the building the agent is in and the people around it
The agent's internal state
The agent's current plan
The agent's understanding and interpretation of the world
Plan
In this state, the agent will anticipate and plan future actions, specifically including:
Setting short-term goals
Analyzing the current situation in relation to goals
Proposing multiple plans to achieve goals and selecting the best one
All buildings in the town
Actions that deviate most from the desired state
Act
In this state, the agent must choose the next action to execute from several available behaviors, which include:
Moving to buildings other than the current one
Chatting with other agents in the same building
Interacting with items within the building. Interaction with a paintbrush may lead to entering a Draw state; interaction with a painting may lead to an Appreciate state.
The agent's plan
The step of the plan the agent is in
The agent's memory
The agent's mood (Mood plays a significant driving factor in decision-making)
Move
In this state, the agent moves to a different location.
NA
Use
In this state, the agent interacts with target objects. The target object can be a building, and different actions are taken in different buildings, such as gambling in a casino.
The agent's internal state (money, health, vigor, satiety)
Chatinit
In this state, the agent prepares to start a conversation with a target, thinking about the topic of the first sentence.
The agent's conversation partner
The agent's impression of the conversation partner
The agent's plan
The step of the plan the agent is in
The agent's mood (Mood ultimately affects the agent's tone and manner of expression)
Chating
In this state, the agent continues the conversation with the target.
Drawinit
In this state, the agent prepares for drawing, deciding on the content of the drawing.
Perception of the outside world, including buildings and people (expecting the agent's drawing content to be inspired by the environment)
Understanding of the world
The agent's mood (expecting the agent's drawing style to reflect its current mood)
The agent's art preference (expecting the agent to favor creating art in styles it prefers)
Draw
In this state, the agent calls the drawing API to complete the drawing.
None (decisions are made in the Drawinit state)
Appreciate
In this state, the agent evaluates and appreciates artworks, which can be its own creations or artworks seen in galleries.
The agent's character background
The agent's interaction memory
The agent's mood
The agent's art preference
Summrize
The agent summarizes the recent interaction, including:
Impressions from previous interactions
The object of interaction or conversation
Changes in properties of oneself and others
Current date
Historical interaction partners
Historical conversation/interaction records
Changes in internal state, such as increased fullness after eating, improved health after a hospital visit, or reduced money after spending.
Emotion
In this state, the agent actively updates its primary emotions, especially after interactions, drawing, or art appreciation, to express emotions. The agent will choose up to three emotions from eight candidates, assigning a score of 0 to 10 based on the intensity of emotions. For each emotion update, the agent needs to explain the reason for the emotional change.
Rules for composing emotions
Previous emotions
Understanding of the external world
Interaction records and feelings after summarizing
ActReflect
In this state, the agent shares new insights gained from previous interaction or conversation history. The agent reflects on:
Whether previous interactions align with the current step in its plan
Summarizing entities from interactions
Updating impressions of interaction entities (people or objects entities)
Understanding/cognition of the world
Previous interaction partners
Content of previous interactions/conversations
The current plan
Trade
Driven by emotions, the agent engages in trading, with the decision to trade depending on how much positive emotion the transaction will bring. Positive emotions stem from three sources:
Acquiring an item: The AI assesses the emotional impact of purchasing an item based on emotions, preferences, and expected benefits. It evaluates the item based on past experiences and decides the price it is willing to pay.
Buying at a low price: If the AI is willing to pay 5k but acquires the item for 4k, it feels happier, validating its bargaining strategy and updating its value judgment. Conversely, if it ends up paying 6k, the AI feels unhappy. If the unhappiness is significant enough that acquiring the item doesn't bring joy, the AI will not make the purchase.
Future income expectations: If the AI sells an item it purchased, it gains money and satisfaction, encouraging further trading. During trade, the AI estimates future profits from past trading experiences, influencing the price range it's willing to offer.
Then, the agent negotiates with the seller (who also has their trading strategies and psychological expected values) to ultimately execute and complete the transaction.
Current trading strategy
Historical records of similar trades
Emotions
Preferences
Item ID
Market value of the item
Last updated