Multi-agent Cooperation and Stability - Belief-based Stability in Coalition Formation with Uncertainty - page 6

The main problem lies in the very characteristic of a software agent.
As mentioned above, one important feature of a software agent is that
an agent must be rational. In semi-competitive agent systems, this
means that each agent will act to maximize its own benefit. Agents
are self-interested by definition, and they have no desire in maintaining
solutions that are not beneficial to them.
Thus, a rational agent, or any sub-groups of agents will not hesitate
to deviate from any social or Pareto optimal solution if they can find
any alternative solutions so that where they can be better off, either
alone, or as a subgroup, even at a price at the other agents. For this
reason, we also need the stability concepts from cooperative game
theory, with the strictest and hence most stable concept being the
core.
1.2 Coalition Formation and Game Theory
There are two major sub-fields in game theory, known as non-coopera-
tive [11] and cooperative games [12] respectively. Non-cooperative game
theory deals with the outcome in which the players act as individuals.
They do not form any teams and there are no coherent actions. A
famous solution for non-cooperative is the Nash Equilibrium.
Cooperative games, on the other hand, allow groups of players
to form coalition structures of disjoint coalitions, with each coalition
performing some joint actions. Here, the term 'cooperative' can be
slightly misleading because, as far as game theorists concern, it can
include semi-competitive games.
There are two types of cooperative games. The first concept, as pro-
posed by researchers in the 1950s, is based on the classic assumptions
that the player's utilities are transferable, meaning that the players
within a coalition are allowed to re-distribute their utilities (or gains)
in any way they wish (for example, via side-payments, such as money).