- NYU professor Gary Marcus said Google overhyped its AI tech and falsely presented it as science.
- Google’s tech was at the center of a claim by an engineer that the company LaMDA AI was sentient.
- Marcus also called out OpenAI and Elon Musk for not being transparent.
Google may have publicly disavowed claims by an employee that one of its chatbots had become sentient, but a prominent AI researcher said the tech giant is also largely to blame in the saga. The researcher, Gary Marcus, argued Google actively hyped the promises of its tech over any scientific research.
“The positive results are celebrated and the skepticism is kind of shunted to the side,” Marcus, an entrepreneur and professor at New York University’s school of engineering, said Wednesday on the tech podcast Dead Cat hosted by this reporter and journalist Eric Newcomer.
Google’s AI projects made headlines earlier this month when Blake Lemoine, a Google engineer and a member of its Responsible AI department, publicly concluded that the company LaMDA chatbot was sentient. While a Google spokesperson, along with many scientists in the AI world, strongly disagreed with that assessment, Marcus said the company has been drafting off its own announcements and promises for years.
Marcus, who sold his artificial intelligence startup to Uber in 2016 and briefly ran the rideshare company AI division, also criticized Google for not being transparent in its AI research or submitting its work for peer-review.
On LaMDA specifically, he referred to the technology that’s built on a tech called large language models, as a “magic trick” that pulls from trillions of words of text to give people the sense of conversing with a sentient being. But these models show very little continuity in the conversation – it does not “remember” previous conversations, for example – and certainly no awareness of what they’re saying.
Marcus extended his critique to Google’s CEO Sundar Pichai who previously promised that the company’s voice assistant Duplex would help people complete real-world tasks. But four years later the tech hardly gets any mentions. And while the company puts out papers explaining the advancements in its tech, Marcus said they do not pass scientific muster.
“[Google is] putting out these articles that look like science. They have bibliographies, they have citations … but then you look carefully and and they’re missing denominators and they’re not going out for peer review, “Marcus said.” So they are portraying themselves as a major contributor to science, but they’re not playing the game of science. “
The problem, Marcus said, extends to other big names in AI including OpenAI, the startup cofounded by Elon Musk and former Y Combinator chief Sam Altman. OpenAI has gotten attention recently for its AI-enabled illustrator DALL-E, which can create pictures based on simple-language prompts from users. Despite its name, OpenAI has not been open to allowing researchers like himself to examine the DALL-E technology, Marcus said.
He was particularly critical of the media for covering the big announcements these companies make, but doing very little follow up or speaking to experts who could cast informed skepticism on the claims.
“When its the last time you read a story on these technologies that actually reviewed the history and said all these promises did not come true?” Marcus said.
Got a tip? Contact reporter Tom Dotan via email at email@example.com or Twitter DM at @cityofthetown.