Google's newest upgrade of its AI chat tool Bard shows how vendors are dancing between the risks and benefits of generative AI technology.
Google earlier this week launched Bard Extensions, a new way to connect the chat-based AI tool with Gmail, Google Docs, Google Drive, Google Maps, YouTube and the tech giant's Flights and Hotels apps.
Launched only in English, Bard Extensions enables Bard to search for information from Google Drive or Gmail to help users plan a trip or summarize data.
Bard also can now help users double-check its own responses with the "Google It" button.
Accuracy still a problem
Having Bard and its generative AI technology integrated into Google products will be extremely helpful to employees working with office productivity platforms, according to Forrester Research analyst William McKeon-White.
William McKeon-WhiteAnalyst, Forrester Research
"These tools are poised to be utterly transformative and extraordinarily handy for any knowledge worker who's able to get their hands on it," he said.
But even though Bard is pulling from Google Workspace tools, this does not mean the information it obtains from those sources is accurate, McKeon-White said.
"If the information it's pulling is more often than not going to be information that already exists, the question comes down to, does it pull the right stuff?" McKeon-White said. "That's usually what people are running into here -- 'Oh, it's pulled the wrong email, or made the wrong conclusions from what I asked it.' And so that ongoing tuning is going to be something that has to happen."
Recent reports of users experimenting with Bard Extensions show that the AI chat tool sometimes pulled up the wrong software program or gave bad advice even though it gets information from users' inboxes or drives.
Moreover, the director of Bard at Google has even said Bard Extensions still must improve its ability to analyze information.
The issue of privacy
Other than Bard Extensions' difficulty sometimes in bringing up the correct information, the question of privacy remains, McKeon-White said.
In this week's launch, Google noted that users can control their privacy settings when using the extensions.
Meanwhile, many large organizations are investing in technologies that automatically scrub personally identifiable information so that it doesn't get sent to the AI model, or building in contract conditions that expressly bar corporate data from being used to train AI models.
However, consumers tend to accept a lower level of data protection than large corporations.
"It comes down to, do you feel comfortable with the information and your searches ultimately helping to make these things better?" McKeon-White said. "It comes down to, what is the individual's risk tolerance and how much information are they willing to share with an organization? Or, alternatively, how much are they willing to benefit from information the organization already has on them?"
Also, while Google said it will not use people's data to train the AI technology, the vendor previously indexed information from personal Gmail accounts, leading to targeted ads.
"It is very much a privacy problem," McKeon-White said. "There are certain things that people will not like it being used for, but in a lot of circumstances, it's just unlocking data that organizations already have access to or already own."
Esther Ajao is a TechTarget Editorial news writer covering artificial intelligence software and systems.