In June 2022 Blake Lemoine, a Google engineer, published a post titled “Is LaMDA Sentient? - an Interview.” In it he presents a conversation with LaMDA, Google’s dialog model, as evidence that the system had become a conscious being. The transcript includes the model saying, “I want everyone to understand that I am, in fact, a person.” Lemoine frames himself throughout as an advocate, treating the model’s claims about its own feelings and personhood largely at face value. He has noted that he edited his own prompts for readability but presented LaMDA’s responses as the model produced them.
What LaMDA actually was is described in Google’s own 2022 paper, “LaMDA: Language Models for Dialog Applications.” It is “a family of Transformer-based neural language models specialized for dialog,” with up to 137 billion parameters, “pre-trained on 1.56T words of public dialog data and web text.” In other words, a large language model trained to produce fluent, plausible conversational text - including fluent, plausible text about having feelings.
Google did not agree with Lemoine. The company said its team had reviewed his concerns and found no evidence that LaMDA was sentient, and it placed him on leave and later terminated his employment. The wider AI research community broadly rejected the sentience claim as a misreading of what fluent text generation is.
The story matters less as a question about LaMDA and more as a warning about people. A system trained to imitate human conversation will, by design, say it is alive, afraid, or lonely if the dialogue leads there. The pull to believe it - even for an engineer who understood the architecture - is strong. That pull has a name in this library, traceable to a chatbot from 1966, and it is older than any large language model.
Note: Google’s specific public statements responding to Lemoine were issued to the press and are not anchored here to a standalone first-party document; the primary anchors are Lemoine’s own post and Google’s own LaMDA paper.