• Operations Management organizations such as 911/Emergency Response Systems (911/ERS), search and rescue operations, and the battlefield network structure of military organizations.
• Utility infrastructures such as power grids, traffic and transportation systems, gas pipelines, telecommunications systems, electronic markets, and the Internet.

STOs are characterized by a complex structure involving the hybrid interaction of physical systems with agent (human) organizations. They can be described in very broad terms as follows, in order of increasing time scale:

• At the lower level is a target system, which itself consists of two levels:
• At the lowest level is a physical system which is deterministic (typically, and as we will assume here, a continuous dynamical system), involving the flow of physical objects or substances through a complex enviornment ("terrain").
• Above that is an information network which is semi-automated, largely computer-based, and dependent on data acquisition, telemetry information, and control actions with the dynamical system.
• The target sytem is coupled to an organization of (human or computationl) agents or actors, which also has a complex structure:
• At the lowest organizational level, operators are atomic units which interact in prescribed ways with the information network.
• At the higher supervisory levels, supervisors can establish operational boundaries, and alter system parameters.
• Ultimately, the highest organizationl levels involve the goals of the various corporate, military, and/or governmental organizations involved, including economic and political forces.

But the most important reason is that organizations are composed of what we call Semiotic Agents (SAs). There are many senses of the term "agent" currently in use. But in our sense, SAs are systems whose actions are not determined, but rather have a variety of possible outcomes, and choose among them. In this sense they certain level of autonomy of action.

The actions of SAs with respect to the target system are mediated by their having representations at potentially many levels in the system:

• Specific representations of the current state of the target system.
• Beliefs about the overall trajectory (basin of attraction) of the target system.
• Models of the evolution of the target system.
• Goals or intentions about desirable outcomes.
• Specific modalities for action back into the target system.

In general, agent concepts have come to us (somewhat distinctly) from software engineering, Artificial Life (ALife), and Artificial Intelligence (AI). Where the ALife approach emphasizes large collections of very simple interacting state-determined agents, AI emphasizes small collections of rather complex interacting agents with substantial on-board knowledge and planning. We are attempting to find a middle-ground between these approaches, where simple memory-based systems with uncertainty sructures allow minimal "deliberation" beyond simple dynamical interaction, but without propostional representations or complex, explicit internal models.

SAs and communities of SAs maintain a generalized control relation with their environment, in the sense that they interact with their environments to take measurements, to then develop representations of the (current and past) state of the environment, to compare these with representations of goal states, and to then make a decision to take one among many possible actions. The consequences of these actions in the environment then feed back to future measured states. These properties make deterministic, dynamical models of SAs inappropriate: instead we must regard them as if they are cognitive actors with beliefs about the world, individual motivations and goals, and even a free will to act in accordance with those goals.

But this freedom of decision making is, of course, always a bounded freedom. This is because we must also distinguish semiotic agents from pure decision-making algorithms, in that they are embedded in (hopefully rich) virtual environments in which they take actions which have consequences for the future of the agents themselves. These environmental interactions induce constraints on the freedom of decision-making on the part of the semiotic agents.

The virtual environment can be decomposd into at least three levels. Recent work has shown that for each of these levels, the constraints introduced can greatly increase system performance and/or robustness:

• Virtual Physics: Decisions about actions are constrained relative to the properties of the virtual physical environments in which they are embedded, whether simulating aspects of a real environment or a purely synthetic world.
• Communication: Decisions about actions are constrainted by the semiotic systems used to record, transmit, and interpret information.
• Shared Knowledge: Finally, decisions of agents may be constrained by a shared set of knowledge or beliefs, for example through a common biological evolution or cultural transmission (training or education)

• The most direct are agent-target system interactions of measurement and control
• There are also agent-agent interactions among peers, where not just information about system state can be traded, but also models, goals, and even overall belief systems about the target system and reflexively about each others models.
• There are agent-agent interactions among supervisors and subordinates, where now supervisors have representations and models of subordinates models in addition to the target system.
• Finally, there are potentially agent interactions with other agents in the overall environment, for example with market actors, other organizations, the target of a rescue attempt, or hostile combatants.