There are numerous advanced techniques to heighten the quality of deepfakes.

One of the most popular involves Generative Adversarial Networks, a method where both a generative and a discriminative algorithm learn from one another to create a trained model.

Whereas the latter tries to differentiate “real” versus “fake” iterations of the training data, the former attempts to trick the discriminative network into accepting fake iterations as real. As the two algorithms go head-to-head, the generative algorithm eventually produces synthetic media that neither the discriminative algorithm (nor the human eye or ear) can easily tell is fake.

Reference Text drawn from Joshua Glick, “Deepfakes 101,” In Event of Moon Disaster website, 2020. Halsey Burgund and MIT. Also, see Tim Hwang, “Deepfakes – Primer and Forecast, NATO Strategic Communications Centre of Excellence, 2020. Background theme image from Shutterstock.

5 of 25