In this old paper, Openshaw (1992) calls attention to the potential of artificial intelligence (AI) methods in relation to spatial modeling and analysis in GIS. He argues that GIS with a low level of intelligence has only little changes to provide efficient solutions to spatial decision-making problems. The application of AI principles and techniques may provide opportunities to meet the challenges encountered in developing intelligent GIS. One thing which draws my attention is that the author mentions it is important to “discover how best to model and analyse what are essentially non-ideal data”. But I didn’t see a definition or explanation of non-ideal data in this paper. Does the non-ideal data refer to less structured data or unreliable data? AI can use less structured data such as raster data, video, voice, and text to generate insights and predictions. However, every AI system needs reliable and diverse data to learn from. Very similar data can lead to overfitting the model, with no new insights.
Further, Openshaw demonstrates the usefulness of artificial neural networks (ANNs) in modeling spatial interaction and classifying spatial data. But he didn’t mention how to transfer data from the GIS to the ANN and back. The most widely used ANNs requires data in raster form. However, the spatial data used to produce an interpretive result in GIS is most efficiently managed in vector form. Therefore, I am wondering if there is an efficient methodology to transfer information between the GIS and the ANN.
As of now, GIScience is not new to AI. For example, the most well-known application of AI is probably image classification, as implemented in many commercial and open tools. Many classification algorithms have been introduced to clustering and neural networks. Also, recent increases in computing power have made AI systems efficiently deal with large amounts of input data. I am looking forward to learning more about the current uses of AI in GIS.