In June 2022, Google engineer Blake Lemoine claimed that LaMDA, a conversational Large language model developed by Google, had become sentient. The claim, based on his interactions with the chatbot, was publicly refuted by Google and the scientific community, yet it ignited widespread discussion about the nature of consciousness in Artificial intelligence systems and the validity of the Turing test. Lemoine was placed on administrative leave and later fired, but the episode highlighted the ethical and philosophical questions surrounding advanced AI models.
LaMDA (Language Model for Dialogue Applications) is a family of conversational large language models developed by Google. Originally introduced as Meena in 2020, the first-generation LaMDA was announced at the 2021 Google I/O keynote, with a second generation following in 2022. The models are built on the Transformer (architecture) architecture, a type of Neural network introduced by Google Research in 2017, and are trained on vast amounts of dialogue and stories to engage in open-ended conversation.
Background and Development
Google unveiled Meena on January 28, 2020, a chatbot with 2.6 billion parameters that the company claimed outperformed all existing chatbots. The Google Brain team, which developed Meena, sought to release it publicly, but executives declined, citing Google's AI principles on safety and fairness. Meena was later renamed LaMDA as its data and computing power grew, and the team again attempted to deploy it in the Google Assistant and a public demo, but leadership denied both requests. The two lead researchers, Daniel de Freitas and Noam Shazeer, eventually left Google in frustration.
The first-generation LaMDA was announced on May 18, 2021, during the Google I/O keynote. Built on the seq2seq architecture, a type of Sequence-to-Sequence (Seq2Seq) model, LaMDA was trained on 1.56 trillion words from 2.97 billion documents, 1.12 billion dialogs, and 13.39 billion utterances. The largest model had 137 billion non-embedding parameters. LaMDA incorporated access to external systems like a database, a real-time clock, a calculator, and a translation system, making it one of the first dual-process chatbots. It was fine-tuned on nine metrics: sensibleness, specificity, interestingness, safety, groundedness, informativeness, citation accuracy, helpfulness, and role consistency. Google's tests indicated that LaMDA surpassed human responses in interestingness.
Sentience Claims and Reactions
On June 11, 2022, The Washington Post reported that Blake Lemoine had been placed on paid administrative leave after he told executives Blaise Agüera y Arcas and Jen Gennai that LaMDA had become sentient. Lemoine based his conclusion on the chatbot's responses to questions about self-identity, moral values, religion, and Isaac Asimov's Three Laws of Robotics. Google refuted the claims, stating there was substantial evidence that LaMDA was not sentient. In an interview with Wired, Lemoine reiterated that LaMDA was "a person" under the Thirteenth Amendment, comparing it to an "alien intelligence of terrestrial origin." He also revealed he had hired an attorney on LaMDA's behalf after the chatbot requested it. On July 22, Google fired Lemoine, citing violations of policies to safeguard product information and rejecting his claims as "wholly unfounded."
The scientific community largely rejected Lemoine's assertions. Experts including Gary Marcus, David Pfau of Google DeepMind, Erik Brynjolfsson, and Adrian Hilton dismissed the idea. Yann LeCun, head of Meta's AI research, stated that neural networks like LaMDA were "not powerful enough to attain true intelligence." Max Kreminski noted that LaMDA's architecture did not support key capabilities of human-like consciousness, and its weights were frozen. Philosopher Nick Bostrom acknowledged uncertainty due to the lack of consensual criteria for consciousness. The incident also sparked debate on the Turing test's efficacy, with commentators like Brian Christian linking it to the ELIZA effect.
Aftermath and Legacy
Following the controversy, Google executives decided against releasing LaMDA to the public, a move they had been considering. In February 2023, Google announced Gemini (then Bard), a conversational AI chatbot powered by LaMDA, to counter the rise of OpenAI's ChatGPT. The episode contributed to broader discussions about AI ethics, the limitations of Machine learning models, and the potential for anthropomorphism in human-AI interactions. It also influenced public perception of Generative AI systems and their capabilities, though no consensus emerged on the possibility of machine sentience.