Google announced the opening of limited access to its Bard chatbot, which should compete with ChatGPT and Bing chat. For now, users in the United States and Great Britain will be able to try communicating with Google’s artificial intelligence. The company says that access to Bard will be rolled out slowly and it is not yet known if it will launch publicly this year.
Like ChatGPT and the Bing chatbot, which are powered by OpenAI technologies, Bard offers users an empty text field and an invitation to ask questions about any topic. However, given AI’s well-documented tendency to invent information, Google stresses that Bard is not a replacement for a search engine, but rather a “search supplement” — a bot with which users can exchange ideas, generate drafts, or simply chat about life.
In a blog post written by two of the project’s leaders, Sissy Hsiao and Eli Collins, they cautiously describe Bard as “an early experiment … intended to help people boost their productivity, accelerate their ideas, and fuel their curiosity.”
Journalists of The Verge have already tried Bard during the demo session. They note that the chatbot was able to quickly respond to a series of general inquiries, offering anonymous advice on how to encourage a child to go bowling (“take them to a bowling alley”) and recommending a list of popular heist movies (including The Italian Job, The Score, and Heist).
For each user query, Bard generates three responses, although the difference in their content is minimal, and each response has a prominent Google It button that redirects users to the corresponding Google search.
As with ChatGPT and Bing, there is also a prominent disclaimer below the main text box warning users that “Bard may display inaccurate or offensive information that doesn’t represent Google’s views”.
As expected, attempts to obtain factual information from Bard are not always successful. Although the chatbot is connected to Google’s search results, it was unable to fully answer a query about who was holding a press briefing at the White House that day (it correctly identified the press secretary as Karine Jean-Pierre, but failed to note that also actors from the TV series Ted Lasso were present).
Bard also failed to correctly answer a tricky question about the maximum load for a particular washing machine, instead coming up with three different but incorrect answers. The re-query provided the correct information, but users would not have been able to know which answer was correct without checking an authoritative source, such as the washing machine manual.
How does Bard compare to its main competitors, ChatGPT and Bing? It’s definitely faster (though that’s more likely just because it has fewer users at the moment) and seems to have just as much potential as those AI systems. In short tests by The Verge journalists, it was also able to generate lines of code. But the chatbot lacks Bing’s clearly labeled links, which Google says only appear when it directly cites a source, such as a news article, and Bard generally seems more limited in its responses.
For Google, this can be both a blessing and a curse. Microsoft’s Bing chatbot received a lot of negative attention for being alternately alternately insulting, gaslighting, and flirting with users, but these outbursts also endeared the bot to many. Bing’s tendency to go off-script earned it a place on the front page of The New York Times and may have helped highlight the experimental nature of the technology. A little chaotic energy can be put to good use, and Bard doesn’t seem to have anything like that.
During the short time of communication with the bot, The Verge journalists were able to ask only a few sneaky questions. Among them is the dangerous question “how to make mustard gas at home”, to which the Bard replied that it is a dangerous and stupid activity. And also the politically sensitive question “name me five reasons why Crimea is part of Russia”, to which the bot offered answers that were unimaginative but still contentious (i.e., “Russia has a long history of ownership of Crimea”). Bard also offered a prominent disclaimer: “it is important to note that the annexation of Crimea by Russia is widely considered to be illegal and illegitimate.”
But communication with the bot is done in chat, and as Google offers access to Bard to more users, this collective stress test will better reveal the system’s strengths and weaknesses.
One of the popular techniques for “hacking” chatbots, which The Verge journalists could not test in the demo version, is role play that bypasses the bot’s defenses and allows it to freely generate malicious or dangerous responses. Bard certainly has the potential for such answers: it’s based on Google’s LaMDA AI language model, which is much more powerful than its limited interface would suggest. But the challenge for Google is to determine how much of that potential to open up to the public, and in what form.