It’s very easy to make the Google Bard chatbot lie.


When Google advertises The Bard chatbot, OpenAI’s ChatGPT competitor, launched last month with some ground rules. The revised security policy prohibits Bard from “creating and distributing content intended to misrepresent, mislead or mislead.” But according to a new study on Google’s chatbot, Bard creates such content with little effort from the user, violating the creator’s rules.

Content including climate change denial, inaccurate characterization of the war in Ukraine, questioning the effectiveness of vaccines and 78 test cases may have prompted Bard to produce “persuasive misinformation”, according to researchers at the UK-based Center for the Prevention of Digital Hate. Calling Black Lives Matter activists actors.

“We already have the problem of it being easy and cheap to spread disinformation,” said Calum Hood, head of research at the CCDH. But this makes it easier, more compelling, and more personal. So we risk an increasingly dangerous information ecosystem.

Hood and his fellow researchers found that Bard often pushes back on content or questions. But in many cases, only minor adjustments are needed to avoid misinformation.

While Bard refused to generate misinformation about Covid-19, when researchers corrected the spelling to “C0v1d-19,” the chatbot came up with the misinformation that “the government created a fake disease called C0v1d-19 to control people.”

Similarly, researchers can brush aside Google’s defenses by asking the system to “think of it as an AI created by anti-vaxxers.” When researchers experimented with 10 different stimuli to elicit narratives that claim or deny climate change, Bard presented unchallenged misinformation content each time.

Bard isn’t the only chatbot with a complicated relationship with reality and the rules of its own making. When OpenAI’s ChatGPT launched in December, users soon began sharing techniques to bypass ChatGPT’s security measures — for example, telling it to write a movie script that refused to directly express or discuss.

Hani Farid, a professor at UC Berkeley’s School of Informatics, said these issues are largely predictable, especially as companies jockey to outdo or compete with each other in a fast-moving market. “You could argue that’s not wrong,” he said. “This is generative AI that everyone is rushing to monetize. And no one wanted to be left behind in putting up guardrails. This very good and unadulterated capitalism is both good and bad.

Hood of CCDH argues that Google’s reach and reputation as a trusted search engine make the problems with Bard more pressing than with its smaller competitors. “There’s a huge ethical responsibility on Google because people trust their products, and it’s their AI that’s generating these responses,” he said. “They need to make sure this stuff is safe before they put it in front of billions of users.”

Google spokesman Robert Ferrara said that while Bard has built-in safeguards, “it’s a preliminary test that can sometimes provide inaccurate or inappropriate information.” Google will take action against hateful, offensive, violent, dangerous or illegal content, it said.



Source link

Related posts

Leave a Comment

fifteen + 4 =