How Trustworthy is AI?
AI is being used increasingly to provide users instantaneous information on various platforms. Pioneered by X (formerly Twitter), the platform implemented an AI system called Grok to respond to users. It can be “@” like a regular user, and the “@” can be followed with a question. However, the AI has been proven to be consistently incorrect on important topics.
Screenshot of Grok AI from X.com
In 2024, five Secretaries of State wrote a letter to Elon Musk, who owns X and subsequently Grok AI, urging Musk to remove of fix the feature. They claimed that, especially in the current state of the political climate, misinformation was directly coming from the open AI system.
This was because:
Grok AI was trained with tweets.
Yes, an open AI that users are relying on to tell them the truth is pulling its sources and information from… other X users with no basis of facts or evidence.
In one example, a user tweeted out a tongue-in-cheek comment regarding a basketball game. Golden State Warriors Guard Klay Thompson struggled and missed many shots in the contest. A fan of the opposing team took to X to mock him. He posted the following:
First off… I am ok.
My house was vandalized by bricks.
After my hands stopped shaking, I managed to call the Sheriff… They were quick to respond.
My window was gone and the police asked if I knew who did it.
I said yes, it was Klay Thompson.
The tweet pokes fun at Thompson shooting “bricks,” but the post quickly turns serious when Grok AI, which uses tweets as sources, claimed that Thompson was “Accused of a bizarre brick vandalism spree”
The AI took the joke tweet as a real news source and fabricated body text claiming Klay Thompson was actually in a police report for throwing bricks at houses.
Image from X user (Lakeshowyo). It shows the original tweet from the poster and the subsequent response from Grok AI.
This is a particularly damning case for the AI because it shows direct proof that it will provide unverified information. In this case, there is not a real threat to Thompson. However, during an election cycle or serious world event, the AI may spew a fabricated story that affects an elected official and their families safety. It is proven to be an unreliable source, yet is propped as an emerging power in the AI industry.
Other open-source AI systems do not have this issue because they were trained correctly. Grok is the only open AI that directly uses social media posts from unidentified users as training to reach an output.