In March, Google released an artificial intelligence chatbot called Bard. It was Google’s answer to OpenAI’s hugely popular ChatGPT.
But Bard used less sophisticated A.I. than ChatGPT. It came across as less capable and less conversational. Within weeks, Google revamped the tool with upgraded technology, but ChatGPT continued to be the chatbot that captured the public’s attention.
On Tuesday, Google unveiled a plan to leapfrog ChatGPT by connecting Bard to its most popular consumer services, such as Gmail, Docs and YouTube. With the new features, Google took a step toward tying Bard into the company’s vast constellation of online products.
Though Bard has not received as much attention as ChatGPT, Google’s A.I. tool has gone from being a chatbot also-ran to a close contender. In August, Bard had nearly 200 million desktop and mobile web visits, closely trailing ChatGPT, according to data from Similarweb, a data analysis firm.
Still, Jack Krawczyk, Google’s product lead for Bard, said in an interview that Google was aware of the issues that had limited the appeal of its chatbot. “It’s neat and novel, but it doesn’t really integrate in with my personal life,” Mr. Krawczyk said users had told the company.
Google’s release of what it calls Bard Extensions follows OpenAI’s announcement in March of ChatGPT plug-ins that allow the chatbot to gain access to updated information and third-party services from other companies, including Expedia, Instacart and OpenTable.
With the latest updates, Google will try to replicate some of the capabilities of its search engine, by incorporating Flights, Hotels and Maps, so users can research travel and transportation. And Bard may come closer to being more of a personalized assistant for users, allowing them to ask which emails it missed and what the most important points of a document are.
A.I. chatbots are widely known to offer not only correct information but also falsehoods, in a phenomenon known as “hallucinations.” Users are left with no way to tell what is true and what is not.
Google believes it has taken a step toward addressing those issues by revamping the “Google It” button featured on Bard’s website, which had allowed users to run Google searches on the queries they had asked the chatbot.
Now, the button will double-check Bard’s answers. When Google has high confidence in a claim and can support it with evidence, it will highlight the text in green and link to another webpage that backs up the information. When Google cannot find facts to bolster a claim, the text is instead highlighted in orange.
“We’re really committed to making Bard more trustworthy by not only showing the confidence of our response, but admitting when we make a mistake,” Mr. Krawczyk said.
Various tech companies have poured billions of dollars into developing the so-called large language models that underpin Bard and other chatbots, systems that need vast amounts of data in order to learn. That has prompted worries about how companies like Google are using consumers’ information.
The company has sought to assuage concerns about how Bard would use this information.
“We’re committed to protecting your personal information,” Yury Pinsky, Bard’s director of product management, wrote in a blog post. “If you choose to use the Workspace extensions, your content from Gmail, Docs and Drive is not seen by human reviewers, used by Bard to show you ads, or used to train the Bard model.”
Mr. Krawczyk said Bard would uphold users’ privacy, though he declined to comment on how other Google services were using this type of data.
Google also updated Bard’s underlying A.I., Pathways Language Model 2. It expanded the feature that allows users to upload images to more than 40 languages. And Google is letting users share Bard conversations with one another, so that they can see the responses and ask the chatbot additional questions on the topic.
Even though people in more than 200 countries and territories are able to use Bard, Google is still calling the tool an “experiment,” rather than a full-fledged product.
“These are still the early days of this technology,” Mr. Krawczyk said, “and they have profound capabilities but they need to be well understood by the people that are using them.”
