Many recent artificial intelligence (AIArtificial Intelligence (AI)) systems in ophthalmologyOphthalmology, starting with deep learningDeep learning (DL) for screeningScreening of diabetic retinopathyDiabetic retinopathy (DR), has stirred excitement due to their robust performances. Abundance of DL models not only in common retina diseases, such as age-related macular degeneration (AMD)Age-related Macular Degeneration (AMD) and retinopathy of prematurityRetinopathy of prematurity (ROP), but in many other fields of ophthalmologyOphthalmology, such as glaucomaGlaucoma and cataractCataract, has proliferated since the first publications of DL for DR screeningScreening. Most of the published models are claimed to have performances at least on par with, if not better than, expert ophthalmologists based on internal validation. Published articles on external validation of AIArtificial Intelligence (AI) in ophthalmologyOphthalmology are far fewer. Only handful articles on deploying AIArtificial Intelligence (AI) for prospective, real-world validation are published. The general tasks for AIArtificial Intelligence (AI) in ophthalmologyOphthalmology, mostly performed on images, can be categorized into (1) classification or screeningScreening, (2) segmentation, and (3) predictionPrediction of disease progressionProgression or treatment outcome. Various architectures of deep convolutional neural networks, such as AlexNet, Inception, VGG, ResNet, etc., are used for classification of DR, AMD, glaucomaGlaucoma, ROP, cataractCataract, etc. Conventional machine learningMachine Learning (ML) methods, such as Bayers’ theorem, support vector machine, decision tree, etc., are still useful for automated segmentation of retinal layers, the optic nerve head area, retinal fluid, etc. Both conventional ML and DL may be applied to predict disease progressionProgression or outcome of treatment, such as progressionProgression of DR severity or vision after intravitreal injections. Although the performance of AIArtificial Intelligence (AI) for this task can be improved with the combination of risk scores, the accuracy of approximately 70–75% is still lower than the other two tasks. There is an increase in articles applying generative adversarial networks for creating synthetic images for research where there is limitation of datasetsDataset. Large language modelsLarge language models: generative pretrained transformer, or chatGPTChatGPT-4.0, has recently been shown to interactively generate texts with very high performance. The roles of these AIArtificial Intelligence (AI) systems in ophthalmologyOphthalmology are remained to be investigated.