"Video is already the most important medium of human communication: over 80% of all internet data will be video in 2022. It will also be a breakthrough year for synthetic video. As AI-generated synthetic media technologies mature, expect the proliferation of billions of user-generated synthetic media clips in 2022. In addition, watch out for the emergence of a few new synthetic media generation companies. The best of these may well become the new tech giants of the next decade."

– Nina Schick, Author, Deepfakes: The Coming Infocalypse

Assessment: Correct ✅ - 2022 was THE breakthrough year for Generative AI and its proliferation is underway. Also, Nina’s prediction of “new synthetic media generation” companies was absolutely true. 2022 most significant fundraises of the year were synthesis companies - Stability ai and Jasper ai. Here are some examples:

• Open AI launched the text-to-image tool DALL-E-2 which takes text prompts and generates images from them. They begin adding 1000 users per week. An open implementation not affiliated with Open AI, DALL-E mini was also created (now Craiyon).
• Meta unveiled the text-to-video platform, Make-A-Video, allowing users to type descriptions of a scene to generate short videos and editing video became as easy as editing a document with Descript’s generative AI tool.  

• Deepfakes made an appearance on America’s Got Talent. Startup Metaphysic made it to the finals with live deepfakes of Simon Cowell and other celebrity judges. 
• This year has been described as the year generative AI exploded into the mainstream. Image-generating systems from OpenAI to Stability AI were released to the public, resulting in a surge of synthetic images and video online and on social media.
• Lensa went viral after releasing its “magic avatars” feature that uses user-uploaded images and neural network Stable Diffusion to generate portraits in various digital art styles, raising privacy and ethics concerns.

"An informed citizenry is the foundation of democracy. Deepfakes are becoming more sophisticated, making it increasingly hard to distinguish reality from harmful fictions. I’m excited about the promise of media provenance technologies to help protect people and society from bad actors, who seek to persuade, mislead, and sow division."

– Dr. Eric Horvitz, Chief Scientific Officer, Microsoft

Correct ✅ - Over the past year synthesis models have advanced, and Deepfakes have become incredibly sophisticated and nearly impossible to distinguish from reality. Further, provenance technology has gained tremendous support and is recognized as one of the most promising approaches to increase transparency in digital content at scale. Here are some examples:

• The Colorado State Fair chose an AI-generated image as the first-place winner in the digital art category, awarding the ribbon to artist Jason Allen, who generated the image using Midjourney and AI Gigapixel.
• Researchers at Stanford University find more than 1000 AI-generated deepfake profiles on LinkedIn. AI algorithms have become sophisticated in the last few years thanks to neural networks and deep learning. They can create synthetic human faces that nearly resemble real ones.
• The FBI warns about an uptick in convincing deepfakes used to apply for remote tech jobs, based on an increase in the number of complaints about scams in which people use a combination of deepfake videos and stolen personal identification information to trick employers into hiring them. 
• According to the DHS, deepfakes, synthetic media, and disinformation pose challenges to society and can impact individuals and institutions ranging from businesses to nation-states. Although there are approaches that can mitigate these challenges, for any to be successful will require collaboration. For example, the Content Authenticity Initiative (CAI) promotes the adoption of an open industry standard for content authenticity and provenance as a promising solution against media manipulation.

"The growing sophistication, power, and democratization of access to deep-fake technologies, paired with the ubiquity of social media, are leading us into a world where the authenticity of any visual data can reasonably be called into question. When anything can be fake, then nothing has to be real, and anyone can easily dismiss inconvenient facts. Our economy, society, and democracy -- requiring a shared set of facts and knowledge -- will be placed at significant risk when this shared knowledge ceases to exist."

- Dr. Hany Farid, Professor, UC Berkeley, Electrical Engineering and Computer Sciences

Assessment: Correct ✅ - Dr. Farid explained the Liar’s Dividend in which anything can be fake allowing anyone to dismiss inconvenient facts. This year we witnessed the Liar’s Dividend being used to question reality as synthetic and manipulated media flooded online communities - a concerning trend previewed over 12 months. Here are some examples: 

• A deepfake depicting Ukrainian President Volodymyr Zelenskyy calling on Ukrainian forces to surrender to Russia circulated online and was later removed for misinformation. Later in the year, an Elon Musk fake promised huge returns in a viral cryptocurrency scheme.  
• Chesney and Citron identify a notable danger in study of deepfakes. As the general public learns about the threats posed by deepfakes, liars trying to dodge responsibility for words and actions gain credibility.  
• During the January 6 hearings, prosecutors submitted a motion seeking to allow the use of video evidence sourced via YouTube. In a response, it was suggested that it could be a deepfake, arguing that prosecutors should be required to verify video authenticity.
• Twitter began giving blue checkmarkas for a small fee resulting in confusion. A nine-word tweet impersonating pharmaceutical giant Eli Lilly and Co resulted in a multi-billion dollar change in market cap. In addition, a deepfake of Sam Bankman-Fried showed the FTX founder claiming he could compensate for losses by doubling cryptocurrency in a giveaway scam.
• The advent of AI-generated synthetic media and deepfakes, create new and controversial challenges that legislators, attorneys, and business must tackle, such as ownership, deepfake specific laws, traditional laws, and courtroom evidence.

Summary: 2022 was truly a watershed year for the digital content we see, hear, and make decisions upon in our daily lives. Two interesting trends are taking shape simultaneously: First, Generative Artificial Intelligence is perhaps the most rapid innovation we’ve ever seen and will make us rethink our concepts of reality. Secondly, digital content provenance and authenticity is also emerging as a growing industry in which the world’s largest technology companies and businesses are beginning to engage in and recognize will become a foundation for the future of business and society. Both of which are incredibly consequential to our use and understanding of the internet, we can’t wait to see what 2023 has in store!

See past newsletter volumes: here.

Correction: Dec 15, 2022 

In an earlier version of this newsletter it was stated that Open AI created DALL-E mini. DALL-E mini is not affiliated with OpenAI.