A recent research paper from Google discusses how the tech giant used a new technique to discover bot-cluster groups producing AI slop videos, thereby resulting in removal of over 130,000 channels from a popular online video platform.
Although Google does not name YouTube, its subsidiary as the platform in the research paper, it shows how the a video platform removed over 130,000 channels in six months using multi-faceted scalable defense system to identify and terminate adversarial” or low-value AI generated content.
The research paper title Scalable Detection of Adversarial Synthetic Slop and Coordinated Media Abuse: A LoRA-Enabled Multimodal Defense System by Google researchers Abhinav Mathur, Birant Orten, Claire Liu, Kelvin Tan and Yifei Liu presents how Google was able to create an effective Scalable Cluster Termination System (S-CTS) that could target entire network of spam accounts, instead of having to analyze each video.
The S-CTS technology, discussed in the paper, combines machine learning with large language models and low-rank adaptation (known by its abbreviation LoRA), to locate spam templates, absurd upload patterns, and shared infrastructure among the identified accounts. According to researchers, once a given amount of accounts has been traced back to a single creation operation, that cluster is terminated.
“Operational data spanning a six-month period demonstrates the system’s significant impact, resulting in the successful termination of 50K clusters comprising 130K channels of synthetic spam generators. Furthermore, the LLM-driven automation significantly improves operational efficiency, saving approximately 83 human review hours to cut down human reviews by 50%.,” read the abstract from Google paper.
The term “AI slop” describes the production of poor quality and repetitive content at a huge scale using artificial intelligence without any or a very small input of original human work. AI slop has spread on YouTube, Instagram, TikTok, and content websites. It can be found in the form of the explainers with AI voice-over, replicated news clips, memes, and other more videos.
As stated by Google, when developing its detection system, the company pays attention to the semantic similarity and considers violations of copyright laws caused by the use of the already existing content.
Other social media platforms like Snapchat are also taking steps to curb the spread of AI slop videos. In a recent policy update on July 31, the company said wholly AI-generated videos will no longer be eligible for recommendation on Spotlight.
“As low-quality, repetitive, AI-generated content becomes increasingly common across the internet, we want Spotlight to remain a place where people can discover authentic creativity from real people, because we believe there’s enduring value in rewarding original perspectives, personal storytelling, and the moments that people choose to create and share themselves,” read a statement from Snapchat.
Also Read: Algorithm and Hate Speech: UN Leaders Call for Action on Digital Extremism
