Chatbots are not the only form of generative AI that students and faculty should be aware of. ChatGPT tends to receive the majority of the attention in discussions of generative media due to its prevalence as an academic tool by students but other topics are equally important. AI generative technology can be used for more than written tasks, it can make images, and videos, mimic voices of politicians and movie stars, generate numerous forms of fake documentation, and much more. Below are some of the other generative tools to be aware of. These creations go by different names but collectively fall under the category of synthetic media.
Synthetic Media and Deepfakes
The term "Deepfake" first appeared on Reddit in 2017 when a user by the name created an area of the popular forum site where they could post pornographic videos where the faces of performers could be swapped passably with the faces of famous politicians, celebrities, or faces requested by other users. The swaps were accomplished using open-source face-swapping technology similar to the tech used in many joke videos and social media filters. The uses for technology outside of adult film content quickly became apparent as "creators" quickly began making use of the technology to create credible images of politicians, celebrities, and people they knew. Soon the technology evolved and developed new capabilities like voice mimicry and higher-resolution video. For some time many researchers have instead used "synthetic media" or "artificial intelligence-generated synthetic media" as umbrella terms to describe original deep fakes as well as their auditory cousins while distinguishing them from legitimate synthetic media creations like movie-generated CGI, professional photoshopping, and computer animation.
Examples of Prominent Deepfakes
After Facebook made it official policy that they would not be policing or removing suspected or confirmed Deepfakes from their social media platforms (Facebook, Instagram, and Metaverse) artists Bill Posters and Daniel Howe worked with advertising company Canny to create a Deepfake video of Meta's CEO Mark Zuckerberg giving a speech about the vast power of Facebook's data collecting while referencing the James Bond villain organization, Specter. The video is still on Instagram and can be viewed here. The artists also made Deepfakes featuring former President Donald Trump. Bill Poster's Instagram page is full of high-quality Deepfake examples.
In 2019 the CEO
How To ID a Deepfake
As the technology continues to evolve Deepfakes have become easier to make and more difficult to detect. Now there is evidence of Deepfakes and other synthetic media being used explicitly to spread mis and disinformation on an array of topics. For now, there are some simple tools one can use to properly identify a Deepfake. Some of these methods are listed below.
For Images and Videos
1) Count things like the number of fingers, arms, or teeth, in a given photo. Many Deepfake programs make clear mistakes when counting the number of body parts an average human being has. Photos of models with 60-plus upper teeth, or pictures of someone hugging a celebrity who suddenly is hugging them with three arms instead of two are common.
2) How does the lighting look? Does it look natural? Are there strange shadows or places that don't look right? AI has a difficult time replicating the complexity of light, especially in videos.
3) Look at the face. Humans have an amazing ability to see slight nuance in the faces we encounter. So when in doubt check out someone's face. Is their mouth moving with their voice? Does it look like their hair is in the right place? Are there eyes proportional to their face? Is there a clear distinct or faint line between the head and the neck?
4) What they're saying sounds deeply out of character. This is mostly useful with politicians or celebrities.
5) Always try to find secondary sources for a video that doesn't look or feel right.
Test Your Knowledge
Northwestern University has created an online guided educational tutorial on spotting and addressing Deepfakes. Test your knowledge here.
In the Event of Moon Disaster This interactive Deepfake project was created to help educate the general public on how easily Deepfakes can be made to fool people or to spread misinformation.
