Deepfakes

While the problem of differentiating between artificially manipulated image, audio, or video and "real" media has existed since at least the early-to-mid twentieth century¹. However, modern generative AI models and services allow for sophisticated generation and manipulation of images, audio, and video, commonly known as "deepfakes" - a term that combines "deep learning" and "fakes"².

A worrying consequences of this emerging era of widespread and cheap deepfakes on political discourse are being monitored by such projects as the AI Election Project.

The challenge of deepfakes in politics is not restricted to any one region of the world, with the following examples documented by the Election Project.

Examples of Political Deepfakes

• United States, Fake Joe Biden robocalls³ in New Hampshire Primary, Fake image of Donald Trump with Black Voters⁴

• Mexico, Fake recording of Mexico City Mayor interfering with Elections⁵

• Ukraine, Deepfake shows Ukrainian president Volodymry Zelensky dancing⁶

• Pakistan, Series of deepfakes encourage election boycotts in Pakistan⁷

• India, South Indian political party uses AI to resurrect dead leader to campaign⁸

• Bangladeshi, Deepfake appears to show Bangladeshi politician dropping out of race⁹

• Taiwan, Deepfake of US congressman appears to support Taiwan’s Democratic Progressive Party¹⁰

• South Africa, Deepfake appears to show Eminem endorsing South African opposition party¹¹

Another significant concern is the proliferation of exploitative content, including non-consensual sexual deepfakes involving celebrities and, alarmingly, the creation of child pornography.

How can Libraries help?

Libraries are adapting their information literacy programs to include new methods for evaluating sources. By teaching patrons how to assess and verify content found online, libraries can help them determine the legitimacy of what patrons encounter when consuming media.