Google’s Bard chatbot is launching on Tuesday in the UK and US, as the company completes its dash to release a competitor to Bing Chat and ChatGPT.
It is seen as a do-or-die moment for the company, whose profitable web search service risks being outcompeted by artificial intelligence (AI) chatbots – even if those chatbots currently have problems in consistently returning accurate and useful results.
Describing the service as an “experiment”, Google’s Jack Krawczyk, the product lead for Bard, said the company was “extremely excited … watching how people are using this product in a way that’s boosting their creativity. It’s helping them accelerate their ideas, and is helping them really fuel their curiosity.
“We feel like we’ve reached the limit of the testing phase of this experiment,” Krawczyk added, “and now we want to gradually begin to roll it out. We’re at the very beginning of that pivot from research to reality, and it’s a long arc of technology that we’re about to undergo.”
Like its peers, Bard lets users type queries and requests in normal English, and is capable of answering detailed questions, giving creative answers to difficult prompts, and engaging in back-and-forth conversations. From Tuesday, users can sign up for access via a waiting list on the company’s website.
In a live demo, Krawczyk showed Bard helping him brainstorm ideas for a birthday party for his son that combines his two obsessions, rabbits and gymnastics.
Using one of Bard’s unique features, the ability to easily click between three answers to the same question, he narrowed down on one suggestion that was more focused on the “gymnastics” part than the “rabbit” part, and asked the chatbot to follow up by generating an email invitation, and then to insert the address of the real venue taken from a search.
But the unplanned follow-up was less impressive. Asked to list ideas for two children on holiday in Tokyo, Bard initially suggested a trip to Tsukiji fish market – but failed to mention that a substantial part of the market moved to a new venue in 2018. In a subsequent question the chatbot acknowledged this, providing footnotes to the correct information.
“The answers aren’t going to be perfect along the way,” Krawczyk said, adding that for some queries a conventional Google search would remain the best option. “There’s a lot of content that has been written across the internet where you can still do deep research, and so what I would expect as people are using Bard is that search will continue to be a complement.”
Google only announced Bard in February, revealing the conversational AI in response to the runaway success of ChatGPT. A few days after Google’s announcement, Microsoft went one further, revealing and launching Bing Chat, powered by OpenAI’s GPT-4 language model.
Unlike those two systems, Bard is based on Google’s own language model, called LaMDA, which hit headlines in June 2022 after an engineer, Blake Lemoine, was put on leave for publishing transcripts that he said demonstrated the system was sentient.
The company has long been a forerunner in AI technology, even inventing the “transformer” technology in 2017 that became the T in “GPT”. But it has historically struggled to ship products based on that research, which insiders have blamed on a mixture of organisational dysfunction and a fear that AI technology could harm the company’s profitable core businesses.
Google did not answer a question on how much more a Bard query costs the company to process than a conventional search, instead focusing on the “efficiency improvements” the company had made. External estimates range from a 10 to a 100-fold increase in costs, yet Bard has no adverts and no obvious revenue sources.
Although Google is playing catchup against would-be AI disruptors, those companies have their own hurdles to leap in producing software fit for millions of users.
On Monday, a privacy breach at OpenAI meant users saw other people’s chat histories. The company shut down the service while it tackled the breach, and as of Tuesday morning the history feature was still offline for many users.
Such flaws could harm the company’s attempts to posit itself as a viable provider of services to corporate customers such as PwC, which last week signed an international deal to use an OpenAI-backed legal chatbot called Harvey. OpenAI did not reply to a request for comment about the breach.
