Jason Smiles: Incremental BDI MAS Learning

CTL AgentSpeak(L): A specification language for agent programs

“…An extended AgentSpeak(L) operational semantics that deals with intentional learning, for both incremental and batch inductive methods, has been proposed by Guerra et al [11]. It is inspired in the way AgentSpeak(L) is extended with speech acts: adding operational semantic rules that are implemented as a plan library.…”

“…Obviously, if an agent is blindly committed, we cannot talk about any form of reconsideration. But if this is not the case, single-minded commitment could be approached as a policy-based reconsideration, where policies are computed through intentional learning [8][9][10][11]. In this way we would reconcile a relevant aspect of the computational theories of BDI agency (commitment) with its philosophical foundation (reconsideration à la Bratman).…”

“…In this way, we can prove if any agent programmed in AgentSpeak(L) satisfies certain properties expressed in the logical specification. This work was motivated by a question related to rational agents, learning, and commitment: can a single-minded committed behavior [13] be approached as a policy-based reconsideration [3] where the policies are intentionally learned by the agents [8,9,11]? The question is relevant, because it reconciles the computational concept of commitment (the blind, singledminded, and open-minded strategies) with the philosophical concept of reconsideration (all reconsideration of α opens the question of whether α), through an adaptive computation of the policy of reconsideration (intentional learning).…”

An Agent-Oriented, Blockchain-Based Design of the Interbank Money Market Trading System

“…If some banks have different preferences regarding, for example, interest rates, their agents would first calculate their preferences and then start the negotiation process with other agents, where they must consider the extent of the differences between their preferences. The learning methods that can be applied by each bank are not specified at this level but can include a range of machine learning methods such as supervised, unsupervised, and reinforcement learning [39][40][41]. Combining these learning methods with the BDI architecture would lead to better decisions by market members [42].…”