Google announces Bard

Background

A chatbot that answers questions in ordinary language runs on a language model: a system trained on large amounts of text to predict what comes next, which lets it hold something that reads like a conversation. Google built its own such model, LaMDA, and Bard was the service built on top of it.^[1] From the start the company described the product as an experiment rather than a finished tool, a way to put the technology in front of real users while it was still being developed.

The announcement

On 6 February 2023 Sundar Pichai announced Bard, calling it an experimental conversational AI service powered by LaMDA.^[1] The same announcement said Bard would launch with a lightweight version of LaMDA, one that required significantly less computing power and would therefore let Google scale the service to more users.^[1] That detail matters because serving a chatbot is costly: every reply means running the model, so a smaller model that spends less computation per answer can be offered to far more people at once.

Google also said it would first open Bard to trusted testers, and in the same post announced that AI features were coming to Google Search.^[1] Pairing a conversational interface with search was the part with the widest potential reach, since search is where most of the company's users already were.

A demo and a drop

Two days later, on 8 February 2023, Google held a livestreamed demo in Paris. During the event viewers pointed out that Bard gave inaccurate information about the James Webb Space Telescope, and Google's stock fell eight percent.^[3] The episode showed how tightly a public demo is bound to credibility: a factual error in a single answer, shown live, was taken as a signal about how ready the underlying technology was.

Opening up and scaling

On 21 March 2023 Google opened Bard to users in the US and UK, describing it as an early experiment.^[2] At Google I/O in May 2023 the service adopted PaLM 2, a different model, and expanded to 180 countries.^[3] Taken together, the two steps moved Bard from a narrow test release to something offered nearly worldwide, and from the lightweight LaMDA of the launch to a newer model.

Why it matters

Bard put Google's conversational AI in front of ordinary users, and its short early history set a pattern that later assistant launches tended to follow: announce with careful "experimental" framing, start with a small group of testers, open in a few countries, then widen distribution and swap in a stronger model once the cost of serving it allows.

The Paris episode became the standard illustration of a second lesson — that a live demo is also a public test, and that a single wrong answer can move a share price. For readers looking back, the sequence is a compact record of how a large company introduced a chatbot in 2023: staged access, an explicit trade-off between model size and the number of users it can serve, and a search product waiting just behind it.