Can Deep Learning Do Everything in AI? An Exploration of Simple Examples in Image Segmentation

Nowadays, Deep Learning, especially ChatGPT and other Generative AI-related products provides tremendous opportunities as tools for so many tasks and jobs, from science to engineering to education. However, when the author asked for programming code to segment a simple image, neither ChatGPT deep learning code nor Google deep learning code provided satisfactory answers. Nevertheless, traditional algorithms such as K-means and lambda-connectedness seemed to deliver better results. This paper attempts to express our analysis of why deep learning neural networks might not work well for some simple but popular problems such as image segmentation.