Data Absorption in GCN

GCN demonstrates the significance of coevolution in the emergence and dynamic maintenance of signaling systems, whether social or biological in nature. The emergent upper level biological signaling system in GCN is the equivalent of a social simulation's institutional level, where an institution is an agreement on the meaning of signs. The degree to which signals coordinate behaviors can be used to determine whether emergence is strong, and the lack thereof can be used to determine system degradation. Indeed, "altered intercellular communication," the final of the nine hallmarks of aging, is the system degradation of aging. The data absorption technique¹¹ employs coevolution to simulate an existing signaling system in its current state of coordination in order to test interventions on that dynamic system. Coevolution is central to the principle of feedback based emergence. It is required for the formation of institutions through symbolic interaction. For example, in the GCN, signs acquire shared meanings as a result of multiple coordinating agents. It is also required for the emulation of signaling-based biological systems, which result from the symbiotic coordination of multiple subsystems. Furthermore, coevolution in the data absorption technique provides a method for determining causal relationships and treatments in a signaling system.

Some of the challenges that test a candidate algorithm in our proposed use of data absorption for a human body simulation within the GCN can test internal consistency by ensuring that the upper level emerges from the lower level in accordance with the data. Different states of health and longevity are the result of various emergent self-reinforcing cycles. Homeostatic mechanisms fail, and as a result, new self-reinforcing systems take over, bringing the body to a new state. For example, in atherosclerosis, it is possible that the liver is already having difficulty clearing LDL while the intestines are absorbing more of it due to aging inefficiencies.

In addition, there are more Reactive Oxygen Species (ROS) in the tissues. This sets off a chain reaction in which the ROS injures the epithelium of the arterial wall, and the LDL that is already present is oxidized by the ROS, forming plaques that eventually rupture, causing a blockage and possibly a stroke. The stroke causes more oxygen dysregulation, and the oxygen deficiency causes mitochondria to release more ROS, further damaging blood vessels.

To simulate this process, we would have a repository of models that mimic the adaptive behaviors of the system components. The data absorption technique would create a causal simulation that takes the body from a healthy state to the described inflammatory state. For example, we would have macrophages that surround oxidized LDL and form foam cells, causing plaques to form.

The adaptive GCN agents can choose from those that best match a data set, such as one that shows the presence of plaques, oxygen, and LDL. We are trying to capture a self reinforcing cycle, whether homeostatic or decaying. However, before all of the GCN agents have chosen their parts of the system and gotten them to work together, no data is generated, and no self reinforcement exists.

We deal with this by "priming" the system by presenting it with signaling data that is not yet part of a self-reinforcing loop. The data absorption algorithm for the atherosclerosis example is shown in Figure 3.

1.	Start with a repository of adaptive agents that model the reactive behavior of the components of the cholesterol deregulation
2.	Have many "data" agents that are not part of a self reinforcing system automatically interact with adaptive agents, with ROS and LDL present
3.	Adaptive agents react to the data agents as they would to adaptive agents, and other adaptive agents adapt to those adaptations, initiating self reinforcement through mutual adaptation.
4.	Challenges reward self reinforcing systems that reproduce the data.
5.	Remove data agents when adaptive agents have reproduced the data

Figure 3. Data absorption in the atherosclerosis example.

In order to implement data absorption in the GCN, we would use some agents that never change their strategies during the simulation and others that do, in recognition of the fact that it does not matter if some agents never act according to personal utility, because the adaptive agents would not react any differently to them than to agents that were internally adaptive. Adaptive agents adapt to their surroundings without regard for whether other agent's behaviors are motivated by their own individual utility or are merely external mimics.

As long as their behavior distribution is consistent with that of their class, they can still seed a society that adapts to, and thus explains, the distribution from agent utility, or lower level rules, and thus cause. Once the feedbacks that make a multiresolutional simulation have stabilized, these seeding agents that do not adapt can be removed.

¹¹ Duong, D. (2013) The Data Absorption Technique: Coevolution to Automate the Accurate Representation of Social Structure. CSSSA 2013, Santa Fe, NM. Available at: http://computationalsocialscience.org/wp-content/uploads/2013/08/Duong2013.pdf (Accessed 13 July 2022)