Log in

No account? Create an account
25 April 2012 @ 08:56 pm
Formal Framework to Support Organizational Design.  
100 Current Papers in Artificial Intelligence, Automated Reasoning and Agent Programming. Number 1.

Catholijn M. Jonker, Viara Popova, Alexei Sharpanskykh, Jan Treur, and Pınar Yolum. Formal Framework to Support Organizational Design. Knowledge-Based Systems 31:89-105, 2012.

DOI: http://dx.doi.org/10.1016/j.knosys.2012.02.011
Open Access?: Sadly no.

I was half hoping to start these posts with a terribly exciting AI paper but sadly nothing terribly exciting caught my eye in the various Tables of Contents last week. However this is very much a bread-and-butter AI paper and so in some ways I'm not so sorry to be starting out with it, since it gives a good impression of what a lot of AI types do when they are not trying to develop machines that will take over the world and wipe-out humanity.

Background: Formalising Knowledge: If you want your artificial intelligence to do any sort of reasoning it is going to need some kind of a knowledge base to work with. While there is some debate about how best to go about creating one of these, one major approach is to attempt to formalise knowledge in some way. This essentially means capturing and expressing information in a mathematical way which your artificial intelligence will then be able to do stuff with.

Background: Organisations A major sub-field of artificial intelligence is autonomous agents. Exactly what these are is open to debate, however it's fine to think of them as mini artificial intelligences who work together. While there are fairly standard methods for creating these, there is less agreement about how they should be coordinated. For instance can some external organisation actually prevent an agent doing something? or can it only monitor and incentivise and sanction? Does an organisation have any computational content of its own? or is it merely the sum of agents that make up the organisation? and so on. This means there is a lot of interest at the moment in the ways organisations are created, and function, and change and, as a result in formalising knowledge about organisations.

This Paper: This paper tackles the design of organisations. It presents a formalisation of organisations and some mathematical tools for creating organisation designs from scratch, and changing organisation designs in a controlled fashion. It isn't actually specifically focused on agent programming though it is clear that one of the motivating factors must have been the design of organisations of computational agents. The work is purely theoretical, i.e. it is just the maths, but it could potentially be implemented as tool for designing organisations either of computational agents, or of human ones.

The Formalisation Bit: I'll just give a flavour of the kind of maths involved. So the first definition is

Definition 1 A specification of an organization. A specification of an organization with the name O is described by the relation is_org_described_by (O, Γ, Δ), where Γ is a structural description and Δ is a description of dynamics. An organizational structure is characterized by the patterns of relationships or activities in an organization, and described by sets of roles, groups, interaction and interlevel links, relations between them and an environment.

This means an organisation has a name, O, and something called a structural description (which hasn't been explained yet) but is labelled Γ and something called a description of dynamics (again not yet explained) which is labelled Δ Definition 2 is explains what the structural description is (its a set of relations of various sorts between various things like the roles members of the organisation can fill, sub-organisations and shared knowledge) and definition 3 explains what the description of dynamics are - it relates descriptions of how things change over time (written in something called a temporal logic) to components of the organisation like the roles.

There's a running example in the paper based on organising the refereeing of papers for a conference (for some reason this is the example used in pretty much every paper on the subject of organisations). So this example has roles such as "Conference Chair" and "Reviewer" and the reviewer role associated with dynamic properties such as "sends in a review before the deadline".

At this point you're about a third of the way through the paper.

Having explaining how they've formalised the concept of an organisation the authors then go on to explain how you can create and change organisations by doing things like creating or removing roles. Most of these are easy to state but they give the actual maths needed to work with their formal definitions, so for instance they explain what needs to be added to the set of roles, or deleted from the sets of relations of roles to dynamic properties and so on. There are a little over a dozen of these but some of them involve quite a lot of work to nail down clearly.

Lastly the paper looks at how you could specify constraints on your organisation before you start designing or changing. You can then check these constraints mathematically to make sure you haven't broken anything. For instance if someone in one role in your organisation, creates some information for someone in another role then there must be some way for the information to get from the first person to the second - i.e. if you remove the role responsible for delivering the internal mail from your organisation then you might be in trouble.

Although this is all theoretical at present, you can see that the ideas, if implemented in a computer program, would help in designing organisations and making sure all the little details were right - these kinds of formal design tools a significant sub-field in Computer Science. I'm not sure how much use the ideas are for non-computational organisations. The paper references a fair amount of literature on tools for human organisation design, so there is obvious an interest in such things outside the world of agent programming.

So there you go, not terribly exciting, but a fairly typical current paper in artificial intelligence.

This entry was originally posted at http://purplecat.dreamwidth.org/66833.html.
jhgowenjhgowen on April 26th, 2012 02:29 pm (UTC)
Well, that was all well-explained, but I never got any sense what the conclusions of the paper were. They have all this theory about organisations, but have they tested any of it? Is it just a laundry list of possible interactions within an organisation?
I never really felt that there was much point to this paper.
louisedennislouisedennis on April 26th, 2012 02:33 pm (UTC)
It was very much a "here's a set of tools" kind of paper and I'm fairly sure there isn't an implementation. Their conclusion is pretty much "no one has ever formalised organisational change before and now we have".