You are correct many authors do use the term interchangably. Saw a presentation yesterday where they used the term MAS. But i would define it as an ABM

Here is how i think of the two

MAS. Usually applies to engineering problems eg in telecoms eg using the Jade system for example. Looking to solve a real problem or to complete a task. MAS systems usually have sophisticated communication systems between agents eg using say FIPA standards. Agents are trying to find a method or set of behaviours to complete a task. MAS therefore has multiple interacting intelligent agents. In many examples many agents are used to solve a task. Rather than just one agent.

ABM usually involves modelling behaviours of agents with set rules and simpler communication protocols eg simulating human behaviours in a social setting Agents are given rules and the simulation looks to see how systems may respond. In essence The goal of an ABM is to search for explanatory insight into the collective behaviour of agents (which don't necessarily need to be "intelligent") obeying simple rules, typically in natural systems, rather than in solving specific practical or engineering problems

Gary

Different things: generally speaking, multi-agent system/model--Homegeneous; agent-based model--Heterogenous.

I have no intrinsic objection to the major proposed distinction (reality versus simulation) distinction, but I'm not sure how many real life situations would correspond to a "multi-agent system".

I think actually articulating the predefined constraints of a multi-agent system may be nearly impossible, particularly in that, in real life, organisms come up with novel ways of dealing with situations.

There is nothing intrinsically wrong with having (effectively) synonyms -- most natural languages make use of them to impart nuance on an ad hoc basis. Nailing down the "definitions" can be entirely on a contextual basis ... I can see myself potentially making those distinctions, but I can se myself chafing at having "multi-agent system" nailed to "real world" systems when I am discussing multi-agent models where both the nature of the agents are changable, an the rules governing those chamges are mutable.

Short summary: It is a really good observation and I can certainly see use in making the distincton at times, but I think making it only when it is needed and in context is more useful. Good observation!

Cheers!

Randall

That's a very thoughtful answer. Thanks for that.

You are correct many authors do use the term interchangably. Saw a presentation yesterday where they used the term MAS. But i would define it as an ABM

Here is how i think of the two

MAS. Usually applies to engineering problems eg in telecoms eg using the Jade system for example. Looking to solve a real problem or to complete a task. MAS systems usually have sophisticated communication systems between agents eg using say FIPA standards. Agents are trying to find a method or set of behaviours to complete a task. MAS therefore has multiple interacting intelligent agents. In many examples many agents are used to solve a task. Rather than just one agent.

ABM usually involves modelling behaviours of agents with set rules and simpler communication protocols eg simulating human behaviours in a social setting Agents are given rules and the simulation looks to see how systems may respond. In essence The goal of an ABM is to search for explanatory insight into the collective behaviour of agents (which don't necessarily need to be "intelligent") obeying simple rules, typically in natural systems, rather than in solving specific practical or engineering problems

Gary

Thank you Gary Howorth for the insight. I also found this other thread that talks about the same topic and there also I find some interesting thoughts.

https://www.researchgate.net/post/What_is_the_diference_between_agent-based_simulation_ABS_and_multi-agent_system_MAS

However, the difference is that in this thread, the participants are more inclined to think that ABM's are more complex and close to real-life while MAS are models more simplified rule-based behavior.

Quoted from the discussion there: " The main difference is that ABM typically implement low numbers of highly complex agents, and the main feature they consider are their individual capabilities to face the task. On the opposite, MAS consider (very) large numbers of simpler agents, focusing on the emergence of new phenomena from social interactions".

This is the opposite of what you just explained and it goes to show that; 1) there is no common agreement on this matter, 2) size and complexity of the system/ model seem to affect the terminology, 3) it may also be the case that different domain experts have a tendency to use one terminology over the other. Maybe this observation aligns with Randall Gray's suggestion to think about the context and structure when choosing the terminology.

Would also add. MAS sytems can also be simulation systems. But the line between ABM’s and MAS is getting blured. The REPAST ABM engine for example has been linked to jade.

I have been working on thr idea of a hybrid approach For a number of years

See https://pure.strath.ac.uk/ws/portalfiles/portal/82383568/Howorth_Kockar_Simultech_2018_Do_we_need_a_new_architecture_for_simulating_power_systems.pdf

I am reimplementing the "toy" model I put together to explore the utility of simulation models that allow entities or groups of entities to change their formal representation based on the their needs, the needs of other entities, or other conditions in the model. I think there are links to pdfs in my profile. It is really heartening to find that I'm not the only explorer in this part of the jungle!

I've been using adaptive time-steps in models that mix formalisms - incorporating population-based, individual-/super-individual based agents and equation based elements since the late 1990's.

I've found it difficult to find researchers that are willing to stray from their familiar toolkits & formal-approaches.

Cheers, all!