Artefacts

Formative Activity

The activity requirement was as follows: "Create an agent dialogue, using KQML and KIF, between two agents (named Alice and Bob). Alice is an agent designed to procure stock and Bob is an agent that controls the stock levels for a warehouse. This dialogue should see Alice asking Bob about the available stock of 50 inch televisions, and also querying the number of HDMI slots the televisions have."

Below is my answer to the assignment:

Here, the agent uses the performative "ask-one", to make sure it receives only one answer, as opposed to "ask-all" as well as the "val" operator for the binding of "?qty" variable to represent the quantity of stock available.

        
          (ask-one
            :sender Alice
            :receiver Bob
            :language KIF
            :ontology warehouse-stock
            :content (val 
                        (stock-quantity 
                          (and 
                            (item-type Televesion) (item-size 50)
                          )
                        ) 
                        ?qty
                      )
          )
        
      

Here, the agent uses the performative "tell", and the actual quantity value instead of the "?qty" variable. Also, the dialogue now includes "in-reply-to" to link the reply to the original message.

        
          (tell
            :sender Bob
            :receiver Alice
            :language KIF
            :ontology warehouse-stock
            :in-reply-to msg1
            :content (val 
                        (stock-quantity 
                          (and 
                            (item-type Televesion) (item-size 50)
                          )
                        ) 
                        12
                      )
          )
        
      

This message echoes Message 1, but this time asking about HDMI slots count in the same type of televisions instead of their stock count.

        
          (ask-one
            :sender Alice
            :receiver Bob
            :language KIF
            :ontology warehouse-stock
            :content (val 
                        (hdmi-slot-count 
                          (and 
                            (item-type Televesion) (item-size 50)
                          )
                        ) 
                        ?num
                      )
          )
        
      

In the same fashion, the agent Bob replies with the number of HDMI slots, using the "tell" performative.

        
          (tell
            :sender Bob
            :receiver Alice
            :language KIF
            :ontology warehouse-stock
            :in-reply-to msg3
            :content (val 
                        (hdmi-slot-count
                          (and 
                            (item-type Televesion) (item-size 50)
                          )
                        ) 
                        3
                      )
          )
        
      

Additional Notes

• I used :ontology warehouse-stock in all messages to ensure that the agents are using the same shared vocabulary to interpret the content against. Without such a shared ontology, stock-quantity and hdmi-slot-count would be just opaque symbols.
• This exchange uses Knowledge Interface Formalism (KIF) as the content language. But since KQML is a communication protofol that acts as the 'envelope' or the pragmatic layer, defining performatives and metadata, KIF can be replaced with any other querying language, espcially if rooted in first-order logic, like OWL, and the agent dialogue would still work.

Artefacts

Collaborative Discussion 2

Peer Response 1

The first post I responded to gave a clear explanation and comparison between KQML as an agent communication language and method invocation techniques. The point that I found interesting in the post and that I aimed to tackle was that the colleague mentioned that, "Because agents only need to agree on message structure and shared vocabularies, KQML can connect heterogeneous implementations with less dependency on shared class libraries or tightly versioned APIs." I found the mention of APIs in this context interesting, so I did a bit of further reading and wrote the following response.

My Response

Your differentiation between KQML (Knowledge Query and Manipulation Language) and method invocation through Python, Java or similar programming languages is clear and to the point. I like how you framed KQML as an 'outer language', and therefore it allows for loose coupling and interoperability across distributed agentic systems (Finin et al., 1994). I agree with you on these points and add that method invocation in Python or Java requires shared code structures and known interfaces (Mayfield, Labrou and Finin, 2005).

I also acknowledge the same KQML disadvantages you mentioned. KQML and Agent Communication Languages (ACLs) in general - being based on the theory of Speech Acts - are usually tricky to achieve its promised full flexibility. KQML agents require a symmetric, shared ontology between agents for the communication to be effective and correct. Without such shared ontology, communication may yield incorrect agent interpretations for performatives.

Furthermore, I found your mention and contrast of our discussion with APIs (Application Programming Interface) thought provoking. It made me think about where KQML stands in relation to APIs. Revisiting Finin et al. (1994), I found the paper discussing KQML KRILs (KQML Router Interface Library). These are the APIs for KQML Routers, which are content independent message channels, and that each KQML-speaking agent is associated has its own process of. This router handles all KQML messages going to and from its associated agent. So, to sum up, KQML-based agents speak through routers that have APIs (KRILs), and therefore KQML and APIs are not exclusive contrasting terms to contradict and compare.

Thank you for your informative post that made me further assert what I know so far about KQML, and that encouraged me to focus on the relationship between KQML and APIs.

Finin, T. et al. (1994) ‘KQML as an Agent Communication Language’, in Proceedings of the Third International Conference on Information and Knowledge Management. Conference on Information and Knowledge Management, New York, NY, USA: Association for Computing Machinery, pp. 456–463. Available at: https://doi.org/10.1145/191246.191322.

Mayfield, J., Labrou, Y. and Finin, T. (2005) ‘Evaluation of KQML as an Agent Communication Language’, in Intelligent Agents II: Agent Theories, Architectures, and Languages. IJCAI’95-ATAL Workshop, Berlin, Heidelberg: Springer.

Peer Response 2

The second post I responded to also gave a good explanation and contrast on the topic. The most interesting part of the post was mentioning that adopting KQML aligns well with BDI-style agent models. Having read about BDI model already for our group project, I again did some additional reading, and focused on this point in my response.

My Response

Thank you for the clear explanation of Agent Communication Languages (ACLs), Knowledge Query and Manipulation Language (KQML) and method invocation for communication among agents. I agree with you on the advantages and disadvantages that you mentioned for each.

What is most interesting for me in your post is that you mentioned that the high-level of KQML abstraction suits Belief-Desire-Intention (BDI) software model. First introduced by Bratman (1987), the model envisions autonomous systems making use of three concepts for its functioning. The first concept is Beliefs, which represent what an agent considers to be true, with an associated degree of certainty. The second is Desires, which mean the high-level aims that an agent operates to achieve. Finally, Intentions are the concrete action plans that an agent commits to realise the desires (Perez, 2019).

With this buildup, BDI model allows agents to work in a clear multistep workflow, building on their beliefs and their evolvement over time, and therefore positions the agent in a better place for reasoning (Georgeff et al., 1999). Your post further suggests that the separation between the belief, desire and intention components also serve in better communication among agents, especially if coupled with KQML as a communication language.

I think your point in this regard makes sense, as long as all communicating agents are built up with the BDI framework. KQML would be suitable here as they would help agents to communicate only intentions and let each respective agent translate those communicated intentions according to its respective beliefs and in the light of its own desires. It would also be important, I think, for those agents, to 'know' about the existence of other agents in its multi-agent ecosystem as part of their beliefs, and have the cooperation with those as part of its desires.

Bratman, M.E. (1987) Intention, Plans, and Practical Reason. Harvard University Press.

Georgeff, M. et al. (1999) ‘The Belief-Desire-Intention Model of Agency’, in Intelligent Agents V: Agents Theories, Architectures, and Languages. ATAL: International Workshop on Agent Theories, Architectures, and Languages, Berlin, Heidelberg: Springer. Available at: https://doi.org/10.1007/3-540-49057-4_1 (Accessed: 7 March 2026).

Perez, A. (2019) ‘Leveraging the Beliefs-Desires-Intentions Agent Architecture’, MSDN, January. Available at: https://learn.microsoft.com/en-us/archive/msdn-magazine/2019/january/machine-learning-leveraging-the-beliefs-desires-intentions-agent-architecture (Accessed: 7 March 2026).