200 likes | 356 Views
Intelligent Architectures for Electronic Commerce. Part 1.5: Symbolic Reasoning Agents. Agent Architectures (1). We want to build agents that are autonomous , can react to appropriate stimuli, act in a goal-directed manner, and interact with other agents. The organisation of the
E N D
Intelligent Architectures for Electronic Commerce Part 1.5: Symbolic Reasoning Agents
Agent Architectures (1) • We want to build agents that are autonomous, can react to appropriate stimuli, act in a goal-directed manner, and interact with other agents. • The organisation of the • knowledge representation, • decision-making machinery, and • agent/environment interface is the architecture of a specific agent design. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Agent Architectures (2) • Originally (1956-1985), most agents designed within AI were symbolic reasoning agents. Agents use explicit logical reasoning in order to decide what to do (GOFAI). • Problems with this approach led to the reactive agents (BBAI) movement (1985-present). • From 1990-present, a number of hybrid alternatives have been proposed. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Symbolic Reasoning Agents (1) • The classical approach to building agents is to view them as a particular type of KBS. • This is known as symbolic AI. • A deliberative agent, or agent architecture, is one that: • contains an explicitly represented, symbolic model of the world; and • makes decisions (for example about what actions to perform) via symbolic reasoning. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Symbolic Reasoning Agents (2) • If we are to build agents in this way, there are two key problems to be solved: • The transduction problem: how to translate sensory data into an accurate, adequate representation in time for it to be useful. … vision, speech understanding, learning. • The representation/reasoning problem: how to symbolically represent complex real-world entities and processes, and how to reason with this information in time for the results to be useful. … knowledge representation, automated reasoning. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Symbolic Reasoning Agents (3) • Most researchers accept that neither problem is anywhere near solved. • The underlying problem is the complexity of symbol manipulation algorithms in general: many (most) search-based symbol manipulation algorithms of interest are highly intractable. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Decision Making as Theorem Proving • How can an agent decide what to do using theorem proving? • The basic idea is to encode a theory stating the best action to perform in any situation. • Let: • be this theory (typically a set of rules); • be a logical database that describes the current state of the world; • Ac be the set of actions the agent can perform; • ⊢ means that can be proved from using Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Finding Actions for each Ac do if ⊢Do() then // find an action return // explicitly end-if // prescribed end-for for each Ac do if ⊬Do() then // find an action return // not excluded end-if end-for Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Example: Toy World • Agent’s goal is to collect all the toys. (0,2) (1,2) (2,2) (0,1) (1,1) (2,1) (0,0) (1,0) (2,0) Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Domain Description • We use three domain predicates for this world: • In(x,y) agent is at location (x,y). • Toy(x,y) there is a toy at location (x,y). • Facing(d) the agent is facing direction d. • Possible actions: • Ac = {turn, forward, pickup} (Note, turn means “turn right”.) Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Decision Making Rules • The rules for determining what to do: In(0,0) Facing(north) Toy(0,0) Do(forward) In(0,1) Facing(north) Toy(0,1) Do(forward) In(0,2) Facing(north) Toy(0,2) Do(turn) In(0,2) Facing(east) Do(forward) In(x,y) Toy(x,y) Do(pickup) • … and so on! • Using these rules (and other obvious ones), starting at (0,0) the agent will collect all the toys. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Limitations • There are a number of problems including: • how to convert video camera input to Toy(x,y) • decision making assumes the environment is static. • decision making using first-order logic is undecidable! (Even if propositional logic is used, in the worst case we must solve co-NP-complete problems.) • Typical solutions: • weaken the logic; • use symbolic, non-logical representations; • shift emphasis of reasoning from run time to design time. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
AOP • Much of the interest in agents from the AI community stems from Shoham’s notion of agent oriented programming (AOP). • AOP was proposed as a `new programming paradigm, based on a societal view of computation’. • The key idea is to directly program agents in terms of intentional notions: belief, commitment, etc. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
AGENT0 (1) • Shoham suggested that a complete AOP system will have three components: • a logic for specifying agents and describing their mental states; • an interpreted language for programming agents; and • an ‘agentification’ process for converting ‘neutral applications’ (e.g. databases) into agents. • Shoham proposed AGENT0 as an initial proposal for providing the first two components. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
AGENT0 (2) • AGENT0 is an extension to LISP. • Each agent in AGENT0 has four components: • a set of capabilities (things that the agent can do); • a set of initial beliefs; • a set of initial commitments (things that the agent will attempt to do — motivational states); and • a set of commitment rules. • The key component, which determines how the agent acts, is the commitment rule set. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
AGENT0 (3) • Each commitment rule contains: • a message condition; • a mental condition; and • an action. • On each ‘agent cycle’: • the message condition is matched against the messages the agent has received; and • the mental condition is matched against the beliefs of the agent. • If the rule fires, the agent becomes committed to the action. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
AGENT0 (4) • Actions may be: • private (an internally executed computation) or • communicative (sending a message). • Messages are constrained to be one of three types: • “requests” to commit to an action; • “unrequests” to refrain from actions; and • “informs” which are used to pass information. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
The AGENT0 Architecture Messages In initialise beliefs update beliefs commitments update commitments abilities Messages Out execute Internal Actions Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
Commitment Rule Example COMMIT( ;; Message condition: I have received a REQUEST ;; from ‘agent’ to do ‘act’ at ‘time’. (agent, REQUEST, DO(time, act)), ;; Mental condition: I believe that ‘agent’ is my ;; friend, I am capable of ‘act’ and I have no ;; other commitments at ‘time’. (B, [now, Friend agent] AND CAN(self, action) AND NOT [time, CMT(self, anyact)]), ;; Then commit to doing ‘act’ at ‘time’. self, DO(time, action) ) Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos
AOP Summary • AOP is probably the first attempt at producing an agent architecture where the emphasis is on a ‘societal view of computation’. • It was designed only as a prototype, and can be seen as a development from the numerous reactive planning agent architectures proposed in the mid 1980s. • However, reactive planning agent architectures give us a more detailed picture of practical reasoning. Intelligent Architectures for Electronic Commerce Timothy J Norman and Wamberto Vasconcelos