We propose a computer model of 'a sense of humor'. The humorous effect is interpreted as a characteristic failure in information processing, caused by the need to quickly discard a false version of the information presented to the mind. The biological function of a sense of humor is to accelerate the processing of information into the mind and to more fully utilize the brain's resources. We analyzed the feasibility of implementing the proposed algorithm in neural networks, which requires organizing their operation in real time. The mechanism of laughter and the development of a sense of humor during evolution are discussed. We propose a specific algorithm for addressing the 'linguistic problem' — namely, the recognition of a continuous sequence of ambiguous patterns, which is essential for the practical development of a computer endowed with a 'sense of humor'.