Archive for October, 2017

Geospatial Agents, Agents everywhere! (Sengupta & Sieber, 2007)

Monday, October 2nd, 2017

This paper gave a comprehensive overview of AI, agents, the role of these in GIScience. I found in many ways this overview brought up topics from the ‘GIS: Tool or Science’ debate, as the paper seemed to bring up AI a lot, when the real topic at hand were agents possibly due to the instant intrigue AI gets from many academic circles outside of geography. Additionally, the disaggregation of agents being either ALGAs or SGAs teases the question of whether agents transcend into their own field or are simply a large subset of GIScience.

The actual uses of agents, for which I was very unfamiliar to prior to this reading, are actually fascinating. Geospatial agents truly capture all aspects of geography (urban planning for urban sprawl modeling, LULC classifications, animal migrations, etc.) and have the ability to not only aid analyze geospatial data with SGAs, though even collect virtual data in the form of ALGAs. Although ALGAs resemble traditional AI and sci-fi, I find the use of SGAs very interesting for the possible near future. The ability to seamlessly work with the plethora of GIS data formats, as well as eliminate age old issues such as the Modifiable Arial Unit Problem and scale would be tremendous strides in GIScience. However, as someone who would like to work in a GIS field later, I also find this troubling as it removes from the human element of GIScience, in that the human knows these things and inputs them into the machine. If the machine does all the work, it seems GIS analysts could be computer techs rather than scientists quite easily.

Currently however, with the examples provided by Rodrigues and Raper (1999) SGAs currently come in the forms of (1) Personal agents (which could be spiders/crawlers), (2) assisting users in their GIS environment (easily done in Python or Bash scripts), (3)helping users find GIS data online (which exist in extensive GIS data repositories), and (4) assisting in decision-making collaborative spatial tasks (which could be seen as modeling in the cases provided on urban sprawl). So far these seem quite attainable, and it will be interesting to see the future uses/advances in geospatial agents, which sentient or not seem like they will have a large role to play in the GIScience future.


Automated Extraction of Movement Rationales (Sengupta et al., 2018)

Monday, October 2nd, 2017

I very much enjoyed this paper by prof. Sengupta as it brings some very new age technology of Agent-Based-Models (ABMs) in a very natural and almost humble way. The methodology for this paper was very interesting as it used pixel based classification to derive land cover (remote sensing), points and buffers (vector files), and DEMs to enhance the pure XY coordinate location of the monkeys to be more useful. We often discuss the advantages of a GIS background versus pure computer science/coding skills, and I find this is a perfect example of this. Although I did find the wording of this paper to reflect topics in computer science quite literally at times, when the behaviour of the Colobus monkeys were defined by conditional statements of ‘IF’ something happens, and ‘ELSE’ if they do not. This pure modelling of behaviour in a virtual context is quite amazing, though also quite tough to believe with all of the additional external variations that could occur, such as predation or simply having an original thought.

What this paper reflects however, is how easy it is to pair observed behaviour, and include it into the models as a block of code to account for some behaviours to make ‘movement and constraining rules’ out of them. The scary yet very possible points brought up in the ‘Future possibilities’ section about using automated extractions from ABMs to augment or replace (a big OR there) heuristic knowledge of experts seems quite possible with the large emergence of ‘Big data’, and its largely growing presence. It makes me wonder if in the future if our assumptions on ecology will be solely based on ones and zeros generated on the computer, rather than actual observations. And if so, what if we were incorrect and continue science in the wrong direction by assuming the computers always right.

Lastly, while this paper reflects ABMs uses in an innocent context, my concerns are when this technology are used on the Colobus monkeys not so distant relative: humans. With all of us essentially carrying GPS trackers in our pockets, I could see data centers observing human movement, and making their own heuristics on people. This becomes dangerous when assumptions are made on people’s movements without their knowing, and without full context beyond our XY coordinates and surrounding objects and people. Could ABMs be used to predict ‘criminals’ based on movement pattern analysis? This is all to be seen in the not so distant future I guess.


Automated Extraction of Movement Rationales (Sengupta et al., 2018)

Monday, October 2nd, 2017

In this paper, the authors present the rationale of animal movements, and show how to extract “rules” from movement data by applying agent-based models (ABMs).

To understand the movement of individual organism, the authors refer to four components including internal state, motion capacity, navigation capacity and external factors, which proposed by Nathan et al. (2008). However, the authors admit that the internal state data are not easy to capture, and they actually didn’t use these data in the experiment. Nevertheless, I believe internal state might influence the modelling results to certain extents. I suppose the movement of animals can reduced by illness or feelings of individuals. It may or may not be significant factors, but it is still an issue that should not be overlooked. As we know, ABMs are not only applied to animal movements, they also used to discover pattern in human activities and support decisions. You cannot deny the importance of individual feelings, wills or other internal elements for concerns of social justice. Even this adds much more complexity when applying models, there is still a necessity to do so when involve humans.

Back to the animal movements, the sample data is gathered from Red Colobus from a national park. The authors infer that there is no mating and predation involved. However, it is not an usual situation in the wild world. I may regard this model as a relative simple version for testing the movement rationale, but not think it is a mature one since the applied scene is kind of ideal. While, I credit the thought to break down the rationales into movement itself and environmental factors. It is a reasonable way to simplify a complicate situation with rigors.

Automated Extraction of Movement Rationales for ABMs

Monday, October 2nd, 2017

I was left with several questions about the observation data. It was collected over a 1.5 year period, yet locations were only observed every 15 minutes. There was no explanation of the daily frequency of observation. If the observation is not continuous, can’t “movement” only be observed between the 15 minute intervals? Was the 75% calculation for movement based on data with many gaps? Interestingly, the group of monkeys remained in a 0.9 km2 box throughout 1.5 years of observation.

This type of ABM appears very simple. Three values (initiation, distance, and direction) are assigned according observed probability, and several additional constraining rules evaluate each new position, causing the algorithm to recalculate a point if certain condition are not met. This example of ABM for animal movement is additionally simplified by the lack of predation and migratory tendencies. Randomness is key to Red Colobus Monkey movement.

This seems to be a step forward for predicting animal movement. The article does not go into detail about the many other factors that may influence the movement of other species, and how these might be modeled by rules and constraints. Another important aspect of ABM should be testing for accuracy against observed movement. Continued observation of Red Colobus Monkeys in KNP would be important for altering rules for the model as well as adding additional constraints or scenarios, such as land cover change (e.g. by human or wildfire) or predation.

Thoughts on ‘ Geospatial Agents, Agents Everywhere…’ (Sengupta, Sieber 2007)

Sunday, October 1st, 2017

I liked how this review delineated very specifically the difference between Artificial Life Agents and Software Agents because I was quite ignorant about the latter before.

What struck me about the four conditions necessary for classification as an intelligent agent, each criteria is the result of many different interacting subfields. For example, the possession of rational behaviour is dependent on decision-making theory from psychology, economics, reward learning paradigms, and machine learning algorithms. These varied fields all have something to contribute to the “rational behaviour” required of an intelligent agent. This suggests that agents are the product of interdisciplinary sciences, and supports the findings that they have very diverse applications.

The major distinction I made between ALGAs and Software Agents was that ALGAs are concerned with intra-human relations; behaviour, interactions between social beings, and the learning that arises from these interactions, whereas Software Agents are concerned with making inter-human processes easier, namely the retrieval and manipulation of spatial data though a computer interface. I am not sure if this distinction captures the true differences between the two types of agents. Are software agents really so removed from the inner workings of humans? Do they not need an awareness of the user and the user’s capabilities, limitations, responses to the environment in order to facilitate easier task processing?

The authors discuss how initial forays into AI research were met with “disillusionment with their true potential in mirroring human intelligence”. This doesn’t surprise me because I think as a society we tend to idealize new technologies when they arise, and overestimate their explanatory, predictive, or transformative power. An example is the use of fMRI and the huge spike in using neural data to explain every conceivable phenomena. AI is also having a major moment, with its use in categorizing human faces for facial recognition to its use in perfecting self-driving cars.  I wonder if agent based modelling, promising though it is in terms of testing out predictions and tweaking specific variables to see the outcomes, is also one of those overly hyped technologies? Is it intrinsically better than naturalistic observation in coming to conclusions about behaviour, or are they different tools to answer different kinds of questions?

The discussion about ALGAs immediately reminded me of agent-based modelling by Thomas Schultz in McGill’s psychology department that modelled different cooperation strategies (ethnocentric, allocentric) to see which one was most evolutionarily successful. This makes me think about how much we can extrapolate from geospatial agent based modelling, especially when dealing with a very granular issue like the behaviour of individual organisms. Can we make widespread predictions about the real world based on these simulations?

The discussion about the various applications of ALGAs, from migratory behaviour to models of urban sprawl, raises questions about what kinds of problems ALGAs are more suited to, and what metric could be used to determine the appropriateness of agent-based modelling for a given issue. This might help curtail the tendency to use the technology for every kind of problem (even those which might benefit from another approach.) Developing some way of testing the adequacy of the method for the issue would also help focus the predictive power and efficacy of the generated models. This is a framework that could be built upon a comprehensive review of the different kinds of issues which geospatial agent based modelling is being used to solve.




Sieber and Sengupta: Geospatial Agents Everywhere

Sunday, October 1st, 2017

Sieber and Sengupta’s paper on geospatial agents situates artificially intelligent agents within the context to GIScience. This spatial context results in the geographically oriented applications of the concept referred to as ‘Artificial Life Geospatial Agents’ (ALGA’s) and ‘Software Geospatial Agents’ (SGA’s). Effectively, these agents are technologies that allow GIScientist to be more efficient in their work by automating elements of the GIS process. These agents are limited by the information available to them; essentially the information available to them is the data that has been made public/accessible. The environments in which the agents are situated could lead to discrepancies among users depending on the information available to the user. in other terms, considering privacy, a user with extensive private data may have more applications and success with the agents than a user using only publicly available data.

I believe that the article handles the subject well within the domain of artificial intelligence. However, as someone with layman information regarding AI, I feel that more context regarding AI at the foundation might have helped to explain how these agents function; understanding the limitations, and the possibilities in addition to the applications might have helped clear up some of the ambiguity I had around the topics.

The article has encouraged me to explore these limitations and see what applications these agents might have for someone for a GIS user such as myself. It would be important to see how much information is needed on behalf of the user to properly take advantage of the geospatial agents.