A new method of detecting RFI in time-frequency visibility images is presented, using the novel Generative Adversarial Network machine learning architecture. The network is trained entirely on manually flagged data produced by the Multi-Element Radio Linked Interferometer Network (e-MERLIN). The network shows an excellent ability to identify and flag RFI features that are common in the training set, but is not as proficient at identifying less common features such as wide-band RFI. The network will be tested against current methods of automatic flagging to quantify and compare its effectiveness. More advanced methods of excising RFI are critical to the future of radio astronomy as the radio spectrum becomes more polluted and even the most remote stations are disrupted by modern satellite networks.