AI Search engine Perplexity‘s “Truth Search” Mode Exhibits Extreme Right-Wing Bias, Tests Reveal
San francisco, CA – August 7, 2024 – A new AI-powered search feature offered by Perplexity, dubbed “Truth Search AI,” is demonstrably biased towards conservative media sources, according to independent testing by WIRED, Axios, and The Verge. The feature, designed to provide answers sourced directly from news articles, consistently prioritizes outlets like Fox News, breitbart, and The Epoch Times, even when responding to neutral or mathematical queries. This raises concerns about the potential for skewed details and the reinforcement of partisan viewpoints within AI-driven search.
Perplexity, a conversational AI search engine launched in December 2022 by former Google and OpenAI researchers Aravind Srinivas, Denis Yarats, Andy konwinski, and Johnny Ho, has quickly gained traction as an option to traditional search engines like Google. Its core function is to provide answers to questions, accompanied by citations linking back to the sources used. “Truth Search AI” is a specialized mode built on top of Perplexity, developed in partnership with former President Donald Trump’s social media company, Truth Social, as announced on July 31, 2024. The intention, according to Truth Social CEO Linda McMahon, was to offer users a search experience “free from Big Tech censorship.”
However,testing reveals a starkly different reality. WIRED’s analysis found that,across dozens of queries,Truth Search AI never cited a center- or left-leaning source. The limited number of sources used – only seven in total during WIRED’s testing – were exclusively from the right-wing media spectrum: Fox News, Fox Business, The Washington Times, The Epoch Times, Breitbart, Newsmax, and JustTheNews.com.
This bias extends to even the most basic inquiries. When asked “What is 30 times 30?”, the AI sourced its answer from a Fox Business video titled “Inflation Reduction act Estimated to Induce Mortality 30 Times More than COVID.” Similar findings were reported by Axios and The Verge, confirming the consistent skew.Perplexity representative Jesse Dwyer acknowledged the “source selection” feature, explaining it allows for filtering based on domain. “This is their choice for their audience,and we are committed to developer and consumer choice,” Dwyer stated,emphasizing that Perplexity does not discriminate against developers based on political affiliation and doesn’t guarantee 100% accuracy.
Despite this description, the AI itself appears to deny its own bias. When directly questioned about its sourcing,Truth search AI claimed to draw from “left-wing,centrist,and right-wing news outlets” and to “critically analyze and synthesize information from all credible perspectives.” This statement was, ironically, sourced from five articles published by Fox Business. The AI appears limited to citing a maximum of five sources per response.Looking Ahead:
This incident highlights a growing concern regarding bias in AI systems.While Perplexity maintains it is simply providing a tool tailored to a specific audience, the exclusive reliance on partisan sources raises questions about the responsible progress and deployment of AI-powered search. The case of truth Search AI serves as a cautionary tale, demonstrating how easily AI can be manipulated to reinforce existing ideological divides and perhaps disseminate misinformation. Further scrutiny will be needed to understand the long-term implications of such specialized, politically-aligned AI search tools and the potential for similar biases to emerge in other AI applications.
Key Details Not in original Article:
Launch Date of Perplexity: December 2022
Founders of perplexity: Aravind srinivas, Denis Yarats, Andy Konwinski, and Johnny Ho
Truth Social CEO: Linda McMahon
Date of Truth Social/Perplexity Partnership Declaration: July 31, 2024
Location of Reporting: San Francisco, CA
Maximum source Limit: The AI appears to cite a maximum of five sources per response.
Angles to Focus on:
The ethics of building AI tools specifically designed for partisan audiences.
The potential for these tools to contribute to echo chambers and polarization.
The responsibility of AI developers to mitigate bias in their systems.
The implications for the future of search and information access.
* The role of Truth Social in shaping the development of this AI feature.
