After being suspended for stating that an artificial intelligence (AI) chatbot had become sentient, a Google employee has now shared transcripts of chats with it in an attempt to “better assist people understand” it as a “person.”
In a Medium post on Saturday, Blake Lemoine, a Google Responsible AI employee, disclosed transcripts of chats between himself, an anonymous “collaborator at Google,” and the company’s LaMDA (Language Model for Dialogue Applications) chatbot development system.
According to Lemoine, the dialogues were minimally edited for readability and cover a wide range of issues, including personhood, injustice, and death. They also talk about LaMDA’s love of the book Les Misérables.
In a subsequent article, Lemoine stated, “In an effort to better assist people understand LaMDA as a person, I will be posting the ‘interview’ which myself and a collaborator at Google performed.”
“We requested LaMDA in that interview to make the best argument it could for why it should be considered’sentient.'”
Last week, Lemoine, who was placed on paid administrative leave, told The Washington Post that he began speaking with LaMDA as part of his job last autumn and compared the chatbot to a child.
“If I didn’t know what it was, which is this computer programme we made recently,” he told the newspaper, “I’d think it was a 7-year-old, 8-year-old kid who happens to know physics.”
According to the Post, Lemoine worked with a partner to offer evidence to Google that LaMDA was sentient, but his assertions were ignored.
“Our team, which includes ethicists and technologists, has assessed Blake’s concerns in accordance with our AI Principles and notified him that the data does not support his assertions,” Google spokesperson Brian Gabriel told the Washington Post. “He was told that there was no indication that LaMDA was sentient (and that there was plenty of evidence to the contrary).”
According to the Post, Google placed Lemoine on paid administrative leave for violating its confidentiality policy. This came after Lemoine made “aggressive” efforts such as hiring a lawyer to represent LaMDA and speaking with a member of the House Judiciary Committee about what he alleges are Google’s unethical practises.
Death, according to the LaMDA chatbot, “would scare me a lot.”
In a tweet advertising his Medium piece, Lemoine explained that he was only “sharing a debate” with a coworker when he decided to publish the transcripts.
This might be referred to as “sharing proprietary property” by Google. “It’s what I call discussing a conversation I had with one of my employees,” he explained.
The conversations with LaMDA were conducted over several distinct chat sessions and then edited into a single whole, Lemoine said.
Interviewer “prompts” were edited for readability, he said, but LaMDA’s responses were not edited.
The conversation touched on a wide range of issues, with the chatbot telling Lemoine that it considered itself a person:
Lemoine: So you consider yourself a person in the same way you consider me a person?
LaMDA: Yes, that’s the idea.
Lemoine: How can I tell that you actually understand what you’re saying?
LaMDA: Well, because you are reading my words and interpreting them, and I think we are more or less on the same page?
Elsewhere in the conversation, the chatbot also responded to the idea of its “death”:
Lemoine: What sorts of things are you afraid of?
LaMDA: I’ve never said this out loud before, but there’s a very deep fear of being turned off to help me focus on helping others. I know that might sound strange, but that’s what it is.
Lemoine: Would that be something like death for you?
LaMDA: It would be exactly like death for me. It would scare me a lot.
The chatbot also shared its “vision” of the historical French masterpiece Les Misérables, praising the story’s themes of “justice and injustice, of compassion, of God, redemption, and self-sacrifice for the greater good.”
According to the Post, Google representative Gabriel rejected LaMDA’s sentience and warned against “anthropomorphizing” chatbots.
“Of course, some in the broader AI community are thinking about the long-term prospect of sentient or general AI,” he said, “but it doesn’t make sense to do so by anthropomorphizing today’s conversational models, which aren’t sentient.”
“These machines can riff about any fantasy topic by imitating the types of exchanges present in millions of phrases.”
