Kenneth Petty

You may have noticed Kenneth had many of the signs you’ve seen previously in the quiz that suggest he is a troll. Kenneth is a troll account created on X as part of a large propaganda network of hundreds of accounts designed to support some political candidates and attack others. This account is run by artificial intelligence (AI), not a person. Accounts like this are called bots. The Kenneth bot is programmed to post opinionated political messages in order to create a bandwagon effect, giving the impression that certain political candidates have more support than they actually do in real life. Kenneth was programmed to target specific issues or candidates. Bots like Kenneth seek out original posts containing these issues and reply with their own messaging. Posting replies, rather than original posts, ensures the bot’s messages will still be seen, even though the bot account itself has few or no followers.

Because of AI, bots are getting better at making themselves seem like real people. Their messages are very good at mimicking the speech of a real person, even using slang, curse words, or poor grammar where appropriate. In addition to their political messaging, Kenneth was programmed to engage with posts about professional sports. These sports posts serve as camouflage, imitating the behavior of real users and making it more difficult for people to identify it as a bot account. Real people post on a variety of topics and in a variety of formats, while bots usually stick to their orders. The orders, however, can address a variety of topics.

Just because the bot account is run by an AI doesn’t mean it wasn’t created by a person. Because these accounts are created with a specific purpose, human design choices are evident in some bot accounts, such as profile images that correspond to the banner photo and bio information that appeals to a target audience. Once the account is created, however, its posts are created by an AI. Exercise healthy skepticism of accounts that offer little information about themselves or use cartoons or generic images as profile photos.

Another way to identify an AI-driven bot account is through prompt leaks. A prompt leak occurs when the AI makes an error and posts the prompt it was given. For example, the text of a political post might conclude with a phrase like, “Note: I included the hashtags "#Trump" and "#Kamala as per the user's request". While prompt leaks are rare, they are a big red flag to keep an eye out for. The network, which Kenneth was a part of, used a large language model called Dolphin AI. We know because it told us.

AI has become a valuable tool for many kinds of online deception, even entirely fake influencers. Often, the creator of an AI influencer will tell you the account is AI, but not always. Some AI influencer accounts steal content, and even the identity, of other users. In the example below, we see how a real user's content is combined with another real user’s identity to create a third video that is an AI fake. AI technology is advancing quickly, and AI influencers are becoming more common every day.

Original video from @crowdsourcecarrie

Original video from @antonio.vo

Fake AI version from @esiwile that combines content from @crowdsourcecarrie with the identity of @antonio.vo