On Sengupta et al. (forthcoming 2018), movement, and ABMs

I thought Sengupta et al.’s article, “Automated Extraction of Movement Rationales for Building Agent-Based Models: Example of a Red Colobus Monkey Group” (forthcoming 2018), was incredibly interesting. “Automated Extraction” discusses the use of agent-based modeling (ABM) strategies in simulating red colobus monkey groups’ movements “across space and time and predict[ing] environmental outcomes” (2). Utilizing the knowledge of experts as input, the modelling hopes to augment “the expert’s interpretation” (2).

At the conclusion of the article, Sengupta et al. note the possibility of ABM’s eventual replacement of scientists’ “heuristic knowledge” (11). It is exciting that ABM is continuing in the theme of original excitement behind GIS (helping us identify patterns that are not easily discernible quantitatively). However, it is also incredibly worrying, as it has the possibility of growing larger than zoological research purposes.

Sengupta et al.’s model relies on human monitoring of the model to check for errors, and the model requires more information from human experts’ field observations to become better at modelling. If AI were to be introduced to the model, and the model learns and understands the patterns better than human experts have observed or can observe, could we reach a point where nothing is unpredictable?

Continuing with the animal theme, this information could be used to predict, for example, where a group could be at a given time and then used on wildlife reserves to organize tours with high success of tourists seeing animals, or help researchers with short time-tables to most effectively study the animals. However, if poachers were to access (possibly) highly accurate modelling, they could more accurately predict the location(s) of groups of animals on the reserve and become more effective hunters of protected species.

For applications using human populations, ABM could be used for humanitarian purposes, like finding the most ideal evacuation routes (and edit existing routes or add new ones) for natural disasters, for example. However, if the model learns extremely well and everything becomes predictable, what would stop nefarious actors from using this information on human populations to catastrophic degrees?

Comments are closed.