Challenges for Scaling Agent-Based Modeling

As described in the previous sections, agent-based modeling may significantly improve the understanding of hominin dispersal processes. However, a number of challenges is existing as well which scientists may face during the modeling and sim­ulation of the platform described.

Scale 1: Expertise. A variety of highly complex domain models, e.g., weather, cli­mate, or botany, needs to be integrated.

As described in Sect. 2.5 a large number of highly specialized domain models needs to be integrated into the simulation platform for modeling the environment and the actors being part of it. The development of each of these models is the respon­sibility of domain experts as they possess knowledge about the mechanisms con­trolling the behavior of a particular factor or entity. These mechanisms and further features need to be transfered into a formal model, which then can be interpreted and integrated by the simulation platform. As mentioned in the previous chapters the integration can either be accomplished by the use of agent-based modeling or as part of a complex mathematical formula determining the potential of a piece of land as cell of a cellular automaton.

Furthermore, besides design and implementation issues, the specification of a multi-disciplinary modeling language is required. In order to achieve a common understanding of the subject among all groups of participants adequate techniques for formalizing and visualizing domain specific insights and to provide an optimal human-computer interaction need to be created.

Scale 2: Space. The model’s granularity concerning the spatial and temporal resolu­tion of the simulation needs to be determined.

Nowadays, due to the availability of cloud-computing and networked systems, computer disk space becomes less important to private users. However, the amount of data being generated when simulating hominin dispersal processes on a detailed level is inconceivable. The area of Africa is currently stated as 30 million km². When modeling the African environment using potential fields, each cell having the size of one square kilometer and each potential requiring 4 bytes^[3] of disk space, a total amount of 115 MB of space is required for each update of the potential field.

As users of the simulation platform are going to simulate a long lapse of time and as stepwise analysis of the simulation progress is required, intermediate data need to be stored as well. Considering an interval of 1.5M years and steps of 100 years, almost 1.7 TB^[4] of disk space will be occupied. When considering smaller time steps and a significantly higher amount of disk space required for the simulation of soft­ware agents, which have not been regarded in the calculation shown above, the chal­lenge of defining a manageable yet informative modeling and simulation granularity becomes apparent.

Scale 3: Time. Particularly influenced by the granularity, scaling challenges emerg­ing from the extraordinary long lapse of time being simulation need to be solved.

Certainly due to the use of agent-based modeling and intelligent software agents the time lapse of simulation needs to be kept in mind. As described in Sect. 2.4 intel­ligent software agents can memorize the world they are located in and changes that occurred during their execution. In case the time steps are not chosen wisely when performing simulation experiments, a considerable amount of data can be collected in no time. As a result of this a new area of agent research needs to be faced: inten­tional forgetting.

Depending on the hypotheses defined by researchers, the level of detail regard­ing individual tribes might vary. In case an ant inspired behavior for discovering new sources of food or fresh water supplies by randomly spreading into all directions, the simulation of each member of a tribe as a single software agent is required. How­ever, in case assumptions defining a continuous movement of a tribe are made the consideration of each member of a tribe might no longer be relevant. Instead, one software agent representing the entire tribe might be sufficient.

Scale 5: Validity. As a consequence of the previous four scales the question of how to validate models of this enormous complexity arises.

Verification and validation is a key aspect of modeling and simulation. The model should correspond to the real world in such a way that effects identified in the sim­ulation are related to the real world behavior of a system. Ensuring verification and validity is a challenge for simulation by itself. Each dimension introduced so far increases the validation complexity significantly. Another challenge arises from the non-existence of hominin dispersal processes in our days such that empirical evi­dence is not provided.

1.7