I found a really
interesting comment over on
Charlie's blog about the effect of technological change on the exercise of control. This comment in turn lead me to
The Rules of the Game by Andrew Gordon, a work on naval history which tells the story of the 1916 battle of Jutland.
The comment that interested me talked about how the introduction of steam powered ships into the Royal Navy, and therefore the increased control over those ships which were no longer slaves to the wind and many other factors, meant that where previously responsibility and authority for any given ship lay with the captains of individual vessels, this role was increasingly taken on by the admiral of the fleet who could precisely order the position and speed of ships under his command. This lead to less initiative on the part of individual vessels and had a deleterious effect on the competence of crews and independence of thought of their captains. The upshot was increased centralised but essentially illusory control over the fleet to the detriment of the military effectiveness of the fleet.
To some degree this is an issue of strategy, as devised by the admiral, versus tactics, of individual captains trying to achieve goals within that strategy. If too much of the tactical role is assumed by the admiral on top of the strategic role, then the control structure of the fleet is brittle. This is fine if the initiative of individual captains is not impacted because they can resume control should the middle not hold. However, should individual initiative be lost then this can leave the fleet without even decentralised control should the communications and instructions from the admiral be lost.
Clearly there is balance that need to be found between centralised control, steering the overall fleet in a desirable direction, and the initiative and intelligent behaviour of individual entities within the fleet. There are distinct parallels between this scenario and the kinds of problems that I am interested in tackling using multiagent systems. Much of computing has traditionally focussed on very centralised control over the elements of a system. Whilst I am sure there is some form of turing-type equivalence between centralised and decentralised solutions to problems, I am also sure that certain problems can be more easily, or more robustly, tackled using decentralised techniques of which software systems comprised of multiple, autonomous intelligent agents.