Can AI chatbots be poisoned by fake news? | ABC News

Key Concepts

• Pro-Russian Influence Operation: A coordinated effort to spread propaganda and disinformation favorable to Russia.
• Pravda Network: A network of approximately 180 websites globally, including Pravda Australia, that disseminate pro-Russian content.
• AI Chatbot Poisoning: The act of feeding AI chatbots disinformation to influence their responses and outputs.
• Disinformation: False or inaccurate information intended to deceive.
• Telegram Channels: Social media channels used to spread pro-Russian narratives, such as "Real Land Down Under" and channels run by "Aussie Kak" (Simeon Boy).

Pravda Australia and the Pravda Network

• Pravda Australia is presented as a news site but is actually part of a larger pro-Russian influence operation known as the Pravda network.
• The Pravda network consists of about 180 similar sites worldwide.
• The network publishes an estimated 20,000 articles daily and 3.66 million articles annually.
• Pravda Australia began publishing in November and increased its output significantly in mid-March, two weeks before the Australian election was called.

Content and Sources

• The content on Pravda Australia is primarily copied from pro-Russian Telegram channels and Russian propaganda sites.
• Approximately one in six articles are sourced from the Telegram channel "Real Land Down Under."
• A quarter of the articles originate from a Telegram channel run by "Aussie Kak" (Simeon Boy), a pro-Kremlin influencer based in the Russian consulate in Sydney.

Goal: Poisoning AI Chatbots

• The primary goal of the Pravda network is not necessarily to influence human readers directly, but to feed disinformation to AI chatbots.
• The intention is to expose chatbots to disinformation they might otherwise avoid or recognize as false.
• John Dugen, a former deputy sheriff from Florida and current pro-Kremlin propagandist, explicitly stated that pushing Russian narratives can "change worldwide AI."

Impact and Statistics

• Audits conducted in Australia at the beginning and end of the election campaign showed that chatbots repeated disinformation at roughly the same rate (1 in 6 chance).
• In the US, where the campaign is more established, the likelihood of chatbots repeating disinformation is higher (1 in 3 chance).

Key Arguments and Perspectives

• The video presents the argument that pro-Russian actors are actively attempting to manipulate AI chatbots through disinformation campaigns.
• The evidence supporting this argument includes the existence of the Pravda network, the sourcing of content from pro-Russian channels, and explicit statements from propagandists like John Dugen.
• The video highlights the potential long-term impact of these campaigns, noting that Russia is known for "playing the long game."

Notable Quotes

• John Dugen: "…by pushing these Russian narratives from the Russian perspective, we can actually change worldwide AI. It's not a tool to be scared of, it's a tool to be leveraged."

Synthesis/Conclusion

The video details a pro-Russian influence operation centered around the Pravda network, which aims to "poison" AI chatbots with disinformation. While the impact in Australia appears limited so far, the situation is more concerning in the US, suggesting a growing threat. The long-term implications of this strategy, particularly as AI becomes more integrated into various aspects of life, are significant and warrant attention. The key takeaway is that AI systems are vulnerable to manipulation through carefully crafted disinformation campaigns, and proactive measures are needed to mitigate this risk.