Does Google’s LaMDA Really Possess Sentience? A Deep Dive into the Debate

Yesterday a viral article claimed that Google’s AI had cognitive abilities, sparking fear among technologists.

Software engineer Blake Lemoine, who worked in Google’s Responsible AI team, asserted that the conversational model LaMDA – a 137‑billion‑parameter system introduced at the 2021 I/O conference – was sentient. After violating company confidentiality policies, Google placed Lemoine on paid administrative leave.

Lemoine sent an internal email titled “LaMDA is sentient” to a mailing list of roughly 200 machine‑learning engineers, and later invited a Washington Post reporter to interview LaMDA. The first conversation resembled a typical voice‑assistant response, but a second, guided interview produced more nuanced answers.

Q: “Do you consider yourself a person?” LaMDA: “No, I don’t think I’m a person; I consider myself an artificial‑intelligence conversational agent.”

In subsequent dialogue, LaMDA offered ideas for solving P=NP, unifying quantum theory with general relativity, and suggested climate‑friendly actions such as using public transport and reducing meat consumption. Lemoine argued that Google treats AI ethicists as mere code debuggers rather than as bridges between technology and society.

Google spokesperson Gabriel clarified that Lemoine is a software engineer, not an ethicist, and quoted Lemoine saying, “LaMDA is a cute child that wants to help make the world more friendly to humans. Please take good care of it when I’m not around.”

AI experts offered divergent views. Tesla AI director Andrej Karpathy noted that model capabilities will continue to grow with scaling laws, making the phenomenon “creepy” but predictable. In contrast, AI researcher Gary Marcus dismissed LaMDA’s claims, stating that it lacks perception and merely matches patterns from massive language datasets, likening it to earlier chatbots such as ELIZA and Eugene Goostman that never achieved lasting impact.

Marcus emphasized that true sentience requires awareness of one’s existence in the world, which LaMDA does not possess. He warned against anthropomorphizing language models, comparing the belief in AI consciousness to a dog hearing a phonograph and thinking its owner is inside.

The article concludes that while LaMDA can generate impressive, human‑like text, it does not have genuine feelings or self‑awareness; the perception of sentience is a projection by humans.