Agent Communication Languages

Initial Collaborative Discussion Post

Agents are computer systems that are capable of acting autonomously to achieve their design objectives in the environment in which they are present. This autonomy sits at the core of agency, enabling the systems to perceive their environment, react to changes, and pursue their goal proactively. A key difference from objects in object-oriented programming (OOP) is that agents maintain control over whether or not they execute actions (Wooldridge, 2009). Multiagent systems are composed of several interacting agents, which can cooperate and coordinate, and ultimately make decisions, all with the purpose of achieving their goals (Drasko and Rakic, 2024).

The critical component of multiagent systems is communication. Various types of systems exist, including non-ACL ones, which are only focused on transmitting information, but agent communication languages (ACLs) such as KQML and FIPA ACL are based on speech acts (Austin, 1962; Suguri, 1999). This enables agents to have complex interactions like negotiations and agreements. Unlike method invocation in Java or Python, where a public method must execute when called, ACLs send requests. Based on these, the receiving agent can choose to act or not, giving them autonomy over their behaviour (Wooldridge, 2009).

KQML is more lightweight and flexible, making it suitable for simpler communication, and FIPA ACL provides better support for rich semantics and dynamic collaborations, making it suited for environments such as smart cities or autonomous robotics (Drasko and Rakic, 2024). There are of course some disadvantages to these languages, for example KQML is lacking in standardisation, leading to inconsistencies across implementations (Gatti et al, 2025) and alongside FIPA ACL it is seen as rigid and lacking flexibility in the broader sense. Despite these limitations, ACL-based systems are extremely valuable for situations that require coordinated agent interactions, and these issues can very well be solved by improving on existing protocols, making them more adaptable (Drasko and Rakic, 2024).