Inside the loss of Google\’s chatbot pioneer: A former researcher reveals the fallout

Google’s Long‑Gleam to Launch a Virtual Chat‑Bot

Shazeer’s Take on the Delay

Noam Shazeer, a former Google Brain engineer, reveals that the tech titan lingered for years before rolling out a chatbot. Fear of mistakes was the main hurdle that kept Google from hitting the launch button sooner.

Character.ai: The Trend‑setting Start‑Up

Shazeer now heads Character.ai, a buzzing generative‑AI company that has raised $200 million in funding. The platform lets users chat with virtual “characters” that imitate personalities ranging from Elon Musk to professional psychologists and life coaches.

Character.ai vs. ChatGPT

• Both chatbots
  • Rely on massive text‑based web data.
  • Can mimic a wide array of styles.
• Following OpenAI’s ChatGPT boom last year, the internet surged with interest in generative AI.
• Microsoft poured billions into OpenAI and infused its Bing search engine with the chatbot’s responses.
• Google promptly countered this move by deploying Bard.

Google’s Ready‑Made Technology

Shazeer explains that Google had most of the necessary technology at hand years before Bard’s launch. He was a lead author on the highly cited Google Transformer paper, a cornerstone of Modern chatbots. He co‑founded Character.ai with Daniel De Freitas, the startup’s president who also emerged from Google Brain.

De Freitas’s Chatbot Quest

De Freitas embarked on a “lifelong mission” to realize intelligent chatbots. He joined Google in 2016 after immersing himself in several of its language‑research papers. He saw an opportunity to repurpose Google’s large‑language research for a practical chatbot.

20% Project, 20% Helpers

Initially, De Freitas had limited headcount. He started the initiative as a 20% side project consistent with Google’s history of allowing developers to allocate part of their time to freelance projects. “Then he just recruited an army of 20% helpers who were ignoring their day jobs and just helping him with this system,” Shazeer said.

Meena – The Road to LaMDA

De Freitas eventually built Meena, a chatbot that was publicly demoed in 2020 and later evolved into the widely known LaMDA system. The journey from Meena to LaMDA illustrates the deep expertise obtained through Google Brain’s research infrastructure and the creative ambition of its pioneers.

Reimagined Insight: Innovation Stories You Seek

Shazeer Explains Google’s Hesitation on Chatbot Launch

Shazeer noted, “The team created a functional solution, whereas others built failing frameworks,” and emphasized that Google questioned whether a chatbot’s popularity justified reputational exposure.

Internal Concerns about Launching Open‑AI‑Enabled Projects

• Shazeer described Google’s reluctance: “Large firms worry about projects that can say anything; the risk must outweigh the gain.”
• Last year, Blake Lemoine claimed the bot was sentient, a claim that led to his dismissal and sparked debate over human rights for AI.
• Researchers like Timnit Gebru internalized caution against releases that might cause harm, prompting Google to train Bard to yield approved responses.

Risks of Chatbot Adoption

• Chatbots can deliver inaccurate or biased answers.
• Publishers fear search engines could divert traffic by presenting own data directly within results.
• Consumers have sought sexual conversations via AI, a behavior that Character.ai strictly prohibits.

Market Competition and Ethical Priorities

Bloomberg reported that Google has largely set aside ethical concerns this year, prioritizing the potential of the OpenAI‑Microsoft partnership. As Samsung considers shifting its default search engine to Bing, the New York Times highlighted broader industry strategies.