In a Google Bard ad, a new AI-powered search tool makes a factual mistake

Google’s push for artificial intelligence search engine Bard shows the company is making a factual error about the James Webb Space Telescope, raising concerns that the tools aren’t ready to be integrated into search engines.

An ad for Google Bard, the tech giant’s experimental conversational artificial intelligence, inadvertently shows a tool that gives a virtually inaccurate answer to a query.

This suggests that the transition to using such AI-powered chatbots to provide web search results is happening too quickly, says Carissa Veliz of the University of Oxford. “The potential for creating disinformation on a massive scale is enormous,” she says.

Google announced this week that it is launching an AI called Bard that will be integrated into its search engine after a testing phase, providing users with a customized written response to their query rather than a list of relevant websites. Chinese search engine Baidu also announced plans for a similar project, and on February 7, Microsoft launched its own AI results service for its Bing search engine.

Experts have warned New scientist that there is a risk that such AI chatbots may give inaccurate answers as if they were facts, because they create their output based on the statistical availability of information, not accuracy.

Now, a Twitter ad from Google has shown Bard answering the question “What new James Webb Space Telescope discoveries can I tell my 9-year-old about?” with incorrect results (see image below).

Bard’s third sentence was, “JWST took the very first pictures of a planet outside of our solar system.” But Grant Tremblay of the Harvard-Smithsonian Center for Astrophysics pointed out that this is not true.

“I’m sure Bard will be impressive, but for the record, JWST didn’t take ‘the very first image of a planet outside of our solar system.’ instead, the first image was by Chauvin et al. (2004) with VLT/NACO using adaptive optics.” he tweeted.

Bruce McIntosh, director of the University of California Observatories and a member of the team that took the first pictures of exoplanets, also noticed the error. write on twitter: “Speaking as someone who photographed an exoplanet 14 years before the launch of JWST, do you think you should find a better example?”

Veliz says the bug and how it slipped through the system is a prime example of the dangers of relying on AI models when accuracy is important.

“This perfectly illustrates the biggest weakness of statistical systems. These systems are designed to give plausible answers depending on statistical analysis – they are not designed to give truthful answers,” she says.

“We are definitely not ready for what is coming. Companies have a financial incentive to be the first to develop or implement certain types of systems, and they are just in a hurry,” Veliz says. “So we don’t give the public time to talk about it and think about it, and they don’t even think about it very carefully themselves, as seen in this ad.”

Google did not respond to a request for comment.