One of the fun things about being a theoretical physicist — or maybe its just me — is that you are so used to abstractions and simple mathematical models that if someone throws a problem at you you can usually come up with a simple version that might be interesting.
So last year I went to a workshop and one of the problems was about decision making. I was in a group of about five other academics and so we though about this and started throwing some ideas together to seeing if anyone saluted. These things can usually end up with some general sort of agreement on working mostly together to see how far idea 3B(ii) can be pushed, or everyone kind of doing their own thing, but also providing a bit of input and feedback to others.
Sometimes these spawn proper funded research projects (or at least part funded), such as I ended up working on, producing a report to the original funder and then a paper …
Agent swarms: cooperation and coordination under stringent communications constraint
Kinsler, Holman, Elliott, Mitchell, Wilson
https://arxiv.org/abs/2210.01163
Abstract: Here we consider the communications tactics appropriate for a group of agents that need to "swarm" together in a highly adversarial environment. Specifically, whilst they need to cooperate by exchanging information with each other about their location and their plans; at the same time they also need to keep such communications to an absolute minimum. […]
… although note that I’ve blogged about this in a general way in ``A conundrum for ninjas’’. I wonder when I’m going to hear back from the journal about the referee reports?
But back to my “decisions” project. I’ve lost the notes now, so I am not sure whether the brief involved both “hierarchy’’ and “distributed’’. As far as I recall, the stuff that did get mostly done at the workshop wasn’t obsessively either, and indeed neither did the corner of the problem I mostly did there. But anyway, whether or not it was prompted, I did have an idea for a simple model of hierarchical and distributed decision making, a bit at the workshop, and much more afterwards. I’d thrown some python code together as a simulator, upgraded it, and after a while this mini-project looked like something that might be written up.
As is rather typical, the idea comes fast, along with the code, but actually turning this into a documents is much more work. You’ve got to systematise the concepts, invent suitable notation (and then, invariably, fix it), work out why to chose choose which parameter values, and then find some sort of results people might be interested in.
Oh, wait. What about the existing literature? There must be some, but is it relevant? This is a new time-sucking rabbit hole to explore. Or is all this really worth the investment in time? It is only a side-project. And cleaning up something for arxiv, let alone an actual journal, can take longer that you might think. But I’ve done most of the work now…
My solution has often been to aim at arxiv, and worry about journals later. Some side-projects, like my synthetic innovation project (which resulted from a bit of pre-research in case I got an interview for a job), can mean you end up with something with no obvious target journal. Although now I’m thinking I can re-pitch the machinery to address different approaches to how increasing levels of intersectionality are chosen, formulated, or generated, which — if I find the time — should just about confuse the bejeezus out of almost everybody (if so: reeeeesult!!).
Enough. Head off to arxiv and read all about my ground-breaking new paradigm in hierarchical decision dynamics. You imagine a tree of agents, with each sending messages up (to superiors) and down (to underlings), based on what they think about what they see. Where, crucially, what they independently believe to be the best action is not what they do, since that is influenced by the reported judgements of others. Further, superiors act more slower, but underlings faster; all have different ways of seeing the world; and different ways of judging success.
Read the gory details here:
A multi-agent model of hierarchical decision dynamics
https://arxiv.org/abs/2404.17477
Abstract: Decision making can be difficult when there are many actors (or agents) who may be coordinating or competing to achieve their various ideas of the optimum outcome. Here I present a simple decision making model with an explicitly hierarchical binary-tree structure, and evaluate how this might cooperate to take actions that match its various evaluations of the uncertain state of the world. Key features of agent behaviour are (a) the separation of its decision making process into three distinct steps: observation, judgement, and action; and (b) the evolution of coordination by the sharing of judgements
I think it does a reasonable job of presenting an investigating the idea, but the results section is weak. It needs a mapping of which parameter ranges give different types of outcome, at the least. But that means more time, fixing the code to run better in batches so I can walk the parameter space more effectively, and various other things. So for now I’ll just park it on arxiv in case anyone finds it interesting.