4 | Artificial Intelligence (AI)
Modern AI is powered by machine learning, a process that involves the system recognizing patterns, understanding experiences, and making decisions to replicate human intelligence. Whether fed images or supplied with videos, AI constantly refines its accuracy by learning from its mistakes and crunching the data it receives for the best outcome. While this process does take time, breakthroughs have enabled machines to not only identify and analyze this information but also create it.
Now, generative AI can fully replicate realistic human-like text, art and images, synthetic voices, and even hyper-realistic videos. But the same technology that enables innovation also makes it possible to fabricate highly convincing fake content, or deepfakes.
What are Deepfakes?
The U.S. Government Accountability Office (GAO) describes deepfakes as a "powerful tool that can be used for exploitation or disinformation." These videos, photos, and audio recordings are synthetic media in which a person's likeness (face, voice, or both) is digitally altered or replaced using AI. Whether manipulating facial expressions, synthesizing faces and speech, or altering face models, deepfakes have the capability of depicting someone saying or doing something they never did. These videos have the capability of influencing elections, eroding trust, inciting civil unrest, and even being used as a weapon of psychological warfare.
[Picture above leads to source.]
Deepfake technology was coined in 2017 by a Reddit moderator who founded a subreddit for users to exchange deepfake pornography they had created of celebrities. While this forum has since been deleted, the term has persisted as the new label for AI-generated content. Today, open-source tools enable rapid experimentation, fueling both creative liberties and harmful uses.
When presented a video that has been heavily edited or manipulated by AI, many individuals will believe what is being presented to them. Deepfakes pose a major threat to trust, as many public figures have been subjects of these manipulated videos or images. Some of these videos are harmless parodies, others are intentionally deceptive or malicious.
Creation of this content typically involves data collection, model training, face and voice synthesis, and refinement. AI tools are given large amounts of images or videos of the celebrity, and the model then learns facial expressions, angles, and movements. Voices are then targeted, with AI-models replicating pitches and intonations. Finally, the videos are then edited for realism, where lighting is adjusted, lip movements are synced, and visual inconsistencies are smoothed.
The more footage of a celebrity, the easier it becomes for AI systems to generate results. Public figures are especially vulnerable because of the sheer amount of content already present online.
These videos raise serious ethical and legal concerns.
Celebrities have no knowledge that their likeness is being used––nor is there any way to truly stop it. These videos can then influence public opinion or disrupt elections, ultimately leading to attacks on reputation before a video can be debunked. But most importantly, if people begin to doubt all video evidence, even authentic footage may be dismissed as fake. These issues are pressing, and it's up to audiences to understand what is and isn't a fabricated video.
Digital literacy is becoming essential in recognizing deepfakes. According to MIT Media Lab, audiences must be able to pay attention to facial transformations, inconsistencies in lighting, textures and shadows, hair and lips, and blinking to determine whether the video they're watching is true or not.
Artificial intelligence reflects the goals of those who use it, and audiences should be aware of what content they consume. The same algorithms that predict weather patterns could be the same ones generating content of a beloved celebrity, and they might not even know it. As AI continues to evolve, we as a society must evolve alongside it.
Comments
Post a Comment