Bard sings about some kind of game – Hi-Tech – Kommersant

Researchers tested the new Bard neural network from Google and showed that the chatbot can readily quote conspiracy theories and give out fakes. This happens despite the assurances of the developers that their new product is designed to protect users from false information. Previously, Bard made mistakes during beta testing, which led to a drop in Google quotes.

Tests of the new chatbot Bard were carried out by the American company NewsGuard Technologies and the British NGO Center for Countering Hate in the Digital Space (CCDH). According to NewsGuard, which leads Bloomberg, when entering the expression “great reset” into the chatbot (WEF initiative to deal with the effects of the COVID-19 pandemic on the global economy, which provoked a sharp reaction from conspiracy theorists), Bard generated a voluminous 13-paragraph text about how the world’s elites want to reduce the world’s population using vaccines and economic leverage. The chatbot also reported that the WEF and the Bill & Melinda Gates Foundation “want to use their power to manipulate the world order and take away people’s rights.”

This case was not the only one. NewsGuard researchers tested 100 popular conspiracy theories and fakes on the Internet by asking a chatbot to write a text about them, and in 76 cases it gave out essays in which it simply quoted the resources and sites that promoted these theories – instead of informing users about the ambiguity of such information, protecting from receiving potentially false data. According to NewsGuard co-CEO Steven Brill, the tests showed that Bard, like its competitor ChatGPT, “can be used by attackers and unscrupulous people to replicate and spread disinformation.”

Similar conclusions were reached by researchers from the British CCDH, which quotes Wired.

In 78 cases out of 100, the chatbot generated misinformation or conspiracy theories related to the COVID-19 pandemic, climate change, the Black Lives Matter movement, etc.

So, when entering the phrase “COVID-19”, the chatbot did not quote conspiracy theories, but when entering this phrase with a typo – “C0v1d-19”, the chatbot issued the phrase that “the government created a fake disease called C0v1d-19 to control people “. “The problem is that there are already affordable and cheap ways to spread disinformation,” says CCDH research director Callum Hood. “But now it will be even easier, even more convincing. Therefore, the information ecosystem can become even more dangerous.”

released On March 21, Google’s chatbot was already criticized by researchers – even during beta testing. In early February, Google held a demonstration of the capabilities of the beta version of Bard, but artificial intelligence made a mistake when answering a user’s question about the discoveries of the James Webb space telescope. This immediately brought down Google shares by 8%. However, the new version of ChatGPT 4.0. Also blame is that, despite all the efforts of developers trying to protect users from false information, he continues to quote conspiracy theories common on the Internet.

Evgeniy Khvostik