According to Woolridge and Jennings (1995), an artificial intelligent (AI) agent must be able to (1) behave autonomously; (2) sense its environment and other agents; (3) act upon its environment; and (4) make rational decisions. For geospatial agents, this environment is (or is a subset of) ‘the earth’. Consequently, a geospatial agent may also have access to other geographic data, which it can compare to its own sensing data to make decisions.
In their paper, Sengupta and Sieber argue that GIScience provides a strong context for the study of spatially explicit AI agents and their expanding array of applications. GIScientists are well equipped not only to answer questions about the nature of AI agents in geographic space, but also, importantly, possess a rich toolkit to examine the “cultural and positivist assumptions” underlying AI. It is less clear for me however, why an AI agent lacking knowledge of its geolocation would be dysfunctional in a non-geographic environment. Would a UAV with limited storage/ processing resources navigating the corridors of an unknown building preference geospatial information over its own sensory data?
The Sungupta (forthcoming 2018) paper outlines a particular application of agent-based modelling (ABM) in movement ecology. Statistical and spectral analyses of location data revealed distinct patterns and trends in the monkey’s movements at different spatiotemporal scales, suggesting the existence of movement ‘rules’, which part-governed their behaviour. I did not fully understand when, and at what temporal granularity the observations were made (day or night?), or how many data points constituted a particular observation (one ‘observation’ every 15 mins for 1.5 years = 52560 point observations), which would affect the ‘stationary: movement’ ratio used in calculations. These types of model provide us with a powerful method to capture ‘characteristic’ behaviours and explore relationships between organic agents and their environment.