In June 2014, Ian Goodfellow and a team including Yoshua Bengio published “Generative Adversarial Networks” on arXiv. The paper proposed a new way to train AI systems that create realistic data, such as images.
The core idea, as described in the abstract, is an adversarial process that trains two models at once: a generator that tries to produce data resembling the real thing, and a discriminator that tries to tell real data from the generator’s fakes. As the two compete, the generator gets steadily better at fooling the discriminator, producing increasingly convincing outputs.
This framework, known as the GAN, became one of the most influential ideas in generative AI. It drove a wave of progress in synthetic photographs, art, and video, and shaped public awareness of both the creative potential and the deepfake risks of generative models. GANs were a major precursor to today’s image-generation tools.