AutonomousAgents Flashcards
What is the literal meaning of “agent” from Latin?
Those who act
What is the fundamental role of an agent?
Someone who acts on behalf of others with power derived from delegation
Name four real-world examples of agents
Secret Agents; Travel Agents; Sports/Showbiz Agents; Purchasing Agents
What are the four common characteristics of real-world agent jobs?
They engage in specific goal tasks; Are delegated by someone; Have power and knowledge to act; Can decide actions based on conditions
What defines a software agent’s goal?
Achieving or maintaining a specific situation
Name two examples of filtering software agents
Antivirus and anti-spam agents
What are the four key characteristics of software agents?
Goal-oriented; Autonomous; Situated; Social
How does a software agent differ from a typical process in goal handling?
A software agent is designed to achieve goals and can change behavior while a process simply executes predefined algorithmic code
What makes modern processes similar to agents?
They adapt to serve best; Capable of unsolicited execution; Adaptive; Situated; Social
What does it mean for an agent to be “situated”?
It executes in a specific environment and can sense and affect that environment
What does “social” mean in the context of software agents?
Interacts with other agents in a multiagent system
What is meant by “goal-orientedness” in agents?
Agents don’t execute algorithms but aim to achieve a specific state of affairs
How does agent autonomy differ from traditional software control?
Agents use distributed decision-making with sub-goals rather than centralized control
What aspects define agent sociality?
Exchange of knowledge; Delegation of tasks; Assignment of roles
What are three key differences between processes and agents?
Function-oriented vs Goal-oriented; Centralized vs Decentralized decision-making; Simple vs Sophisticated interactions
What makes self-driving cars exemplary autonomous agents?
They have goals navigation goals; Have knowledge of rules; Can sense conditions; Act on controls; Must minimize danger; Decide actions dynamically
What are “beliefs” in agent knowledge?
Knowledge about the current state of affairs
What are “desires” in agent knowledge?
Knowledge about the desired state of affairs
What are “effects” in agent knowledge?
Knowledge of how actions can change the current state of affairs
What is theoretical reasoning in agents?
A process which affects beliefs to increase knowledge about what’s happening
What is practical reasoning in agents?
Weighing conflicting considerations based on desires and beliefs to choose actions
What are the two components of practical reasoning?
Deliberation (deciding to act) and Means-End Reasoning (deciding how to act)
What is the main challenge of using deep learning in agents regarding XAI?
Deep learning makes agents “black boxes” making it difficult to explain their decisions
What is the key contrast between deep learning and traditional agents?
Deep Learning: Black-box not explainable not symbolic; Agents: White-box explainable symbolic
What are the six main types of agent architectures?
Reactive; Reactive with State; Proactive goal-oriented; Proactive utility-oriented; Hybrid proactive; Reinforcement Learning
What defines a reactive agent?
Perceives events and acts based on simple rules with implicit goals
What are the advantages and disadvantages of reactive agents?
Advantages: Simple and efficient; Disadvantages: Not flexible and not truly autonomous
Give three examples of reactive agents
Temperature agent; Simple mail agent; Roomba robot
How does a reactive agent with state differ from a simple reactive agent?
Maintains internal knowledge and can make decisions based on both current perception and past state
What are two examples of reactive agents with state?
Mail agent that tracks user markings; Roomba robot that maps environment
What defines a proactive goal-oriented agent?
Has explicit knowledge of goals and selects actions to approach the goal
What are the two types of planning in goal-oriented agents?
Offline planning (predefined plans) and Online planning (using planning algorithm)
What is a causal model in agent planning?
Specifies the effect of actions on the environment probabilistically
What are the two types of planning algorithms?
Greedy planning (single action at a time) and Multistep planning (sequences of actions)
What defines a utility-oriented agent?
Aims to maximize a specific current utility or preserve quality of specific conditions
What is the key difference between goal-oriented and utility-oriented agents?
Goal-oriented evaluate current vs desired state; Utility-oriented evaluate a utility function against desired value
What is a hybrid architecture in agents?
Mixes utility and goals requiring trade-offs between the two
What is Newell’s Principle of Rationality?
The rational choice is the one that given current knowledge is most suitable to accomplish/approach the goal
What are the two main limitations of the rationality principle?
Knowledge can be partial/incorrect; World can dynamically change as agent acts
What defines reinforcement learning in agents?
Self-supervised learning where agent learns how to act through rewards
What are the key components of reinforcement learning?
Set of actions; Policy for selecting actions; Observation of environment; Reward measurement
What is model-based reinforcement learning?
Focuses on learning explicit representation of relationship between environment state and agent actions
What is the main advantage of model-based RL?
Builds a reusable inspectable and explainable model of the world
What is model-free reinforcement learning?
Doesn’t construct explicit world model relies on exploration to determine rewarding actions
What is Q-Learning?
Traditional approach to model-free RL that builds a Q-Table matrix associating values to actions in given states
What is Deep Q-Learning?
Embeds the policy (Q-table) into a Deep Neural Network for handling large state dimensions
What are the advantages and disadvantages of Deep Q-Learning?
Advantages: Learns complex behaviors handles large spaces; Disadvantages: Lacks explainability not reusable costly learning
What is SOTA in agent systems?
State of Affairs representing current desired condition within environment
What defines a goal in SOTA phase space?
Reaching a particular area in the SOTA phase space
What are the two conditions that characterize goals?
Precondition (when goal should activate) and Postcondition (what achievement implies)
What are utilities in SOTA phase space?
Constraints on how goals can be reached within the system
What is the BDI model?
Belief-Desire-Intention model for operational way to think about goal-oriented agents
What are the three key components of BDI?
Beliefs (knowledge about world); Desires (goals); Intentions (currently activated goals)
What is JADE?
A goal-oriented agent system where agents have state and behaviors for sub-tasks
What is Jason?
A declarative language for BDI agent-oriented programming using logic-programming approaches