Amplification to Synthesis: A Comparative Analysis of Cognitive Operations Before and After Generative AI
This paper analyzes X (formerly Twitter) data from the 2016 and 2024 U.S. presidential elections to demonstrate that cognitive operations have fundamentally shifted from bot-driven amplification of identical content to generative AI-driven synthesis, characterized by a surge in original content, collapsed lexical overlap, and narratively concentrated coordination.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the internet as a giant, noisy town square where people shout opinions, share news, and try to convince others of their point of view. For years, security experts have been watching this square to spot "cognitive operations"—essentially, organized campaigns designed to trick people into believing things or changing their minds.
This paper compares two specific moments in this town square: the 2016 U.S. presidential election and the 2024 election. The researchers wanted to see if the "bad actors" changed their tactics after a new technology called Generative AI (like the chatbots that can write stories for you) became popular.
Here is what they found, explained through simple analogies:
1. The Shift from "Copy-Paste" to "Freshly Baked"
The Old Way (2016):
Imagine a group of people trying to make a rumor spread. In 2016, they mostly acted like a photocopy machine. They would find one single message, print it out thousands of times, and hand it to everyone. On social media, this looked like millions of people hitting the "Retweet" button.
- The Data: About 40% of the messages were just retweets. The words were almost identical, like a chorus of people singing the exact same note.
The New Way (2024):
By 2024, the photocopy machine was gone. Instead, the group acted like a team of ghostwriters. They didn't just repeat the same message; they wrote thousands of new stories that all said the same thing but used completely different words.
- The Data: Almost 93% of the messages were "original posts." The "Retweet" button was barely used (less than 1%). It looked like everyone was writing their own unique essay about the same topic.
2. The Shift from "Crowded Room" to "Secret Club"
The Old Way (2016):
In 2016, the bad actors were like a crowded mosh pit. They would all shout at the exact same second, regardless of what they were shouting about. If one person yelled about the weather and another yelled about politics, they would both scream at 2:00 PM sharp. This "burst" of noise was their signature.
- The Data: They coordinated their timing perfectly across different topics. It was chaotic and loud everywhere at once.
The New Way (2024):
In 2024, the mosh pit disappeared. Instead, they formed small, secret clubs. They stopped shouting at the same time about random things. Instead, they waited until they were all talking about the same specific story, and then they all posted at the exact same moment.
- The Data: They rarely shouted at the same time about different topics. But when they were talking about one specific story (like a celebrity endorsement), they coordinated their timing perfectly within that group.
3. The Shift from "Robots" to "Human-Like Writers"
The Old Way (2016):
The messages were so identical it was obvious they were fake. It was like a robot reading a script.
- The Data: The words were 99% identical. If you compared two posts, they were practically the same sentence.
The New Way (2024):
The messages were diverse. One person might say, "Taylor Swift supports the candidate!" while another says, "The pop star just threw her weight behind the VP!" They meant the same thing, but the words were totally different.
- The Data: The words were very different (low similarity), but the meaning was the same. This is exactly what Generative AI is good at: taking one idea and rewriting it in a hundred different ways to sound natural.
The Big Conclusion
The researchers aren't saying, "We caught the AI." They are saying, "The fingerprints have changed."
- Before AI: The bad guys used bots to amplify (repeat) existing content.
- After AI: The bad guys seem to be using AI to synthesize (create) new content that looks unique but pushes the same agenda.
The paper suggests that the "bad guys" have upgraded their toolkit. They aren't just hitting "copy and paste" anymore; they are using AI to write fresh, unique stories that coordinate perfectly around specific narratives, making them much harder to spot than the old, repetitive spam.
Important Note: The authors are careful to say they didn't prove AI definitely caused this. They just found that the patterns in 2024 look exactly like what you would expect if someone were using AI, whereas the 2016 patterns looked like old-school bot networks. They are providing a new "baseline" for security experts to look for these new patterns in the future.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.