Abstract. Recent artificial intelligence (AI) technology is capable to generate new texts (so that it makes new stories) by learning order of words in partic-ular datasets. This paper reports an experiment of generation of imaginal in-cident reports by AI text generator, which learn from real aviation incident reports. AI tries to synthesize new texts to maintain similarity of word order pattern to training data of real reports, while it may generate brand-new sto-ries due to randomness in its algorithm. We may find hidden risk of human error incidents among those imaginal reports. Although theoretical capability of AI text generation is widely approved, it is still a practically hard problem, which requires large costs and massive amount of data. Therefore, the main contribution of this paper would be not only qualitative evaluation of made imaginal reports, but also quantitative evaluation of its feasibility. Keywords: Incident report · Human Error · Human Factors· Neural net-works · Text synthesis · Natural Language Processing · Artificial Intelligence