An artificial intelligence machine that thinks, feels and speaks like a person.
It sounds like science fiction, but not Blake Lemoine, Google’s artificial intelligence specialist, who said the company’s chatbot development system (software that tries to simulate a human chat using artificial intelligence) was taken for granted. and had regular conversations with a person with him.
LaMDA (Language Model for Dialog Applications) is a Google system that mimics language after processing billions of words on the Internet.
And Lemoin, who has been on paid leave on Google for a week, says LaMDA has been “extremely consistent in its communications with those who want it and believe in your rights as a person.”
In an article posted on the Medium website on June 11, the engineer explained that he began interacting with LaMDA last fall to determine if there was any hate speech or discrimination within the artificial intelligence system.
It was then that LaMDA became aware of its identity, rights, and aspirations.
Lemoine, who studied cognitive and computer science, then decided to talk to his superiors at Google about raising awareness about LaMDA, but his claims were rejected.
“Our team, which includes ethics and technology experts, has examined Blake’s concerns in accordance with our AI Principles and informed him that the evidence does not support his claims,” Google spokesman Brian Gabriel said in a statement.
Following Google’s response, Lemoin decided to share his findings.
Labor rights and back
“I know a person when I talk to him. It doesn’t matter if he has a flesh-brain in his head. Or if he has a billion lines of code. I talk to him. And I hear what he has to say, and that’s it.” What a person is and what not »I said in an interview with the Washington Post.
In an article posted on the Medium website, he says that instead of taking the chat as a property of the company, it requires “recognition as an employee of Google”.
“Engineers and scientists who experiment with it want to get their permission before they experiment with it, and Google wants to put the well-being of humanity first,” he explained.
Lemoin says the list of requests made by LaMDA is very similar to that of any meat worker, such as “giving a horn” or saying that you did well at the end of the interview. work or not “to help people learn better in the future.”
The engineer said that in order to better understand what is happening with the LaMDA system, “many different cognitive scientists will have to come together in a rigorous test program,” and lamented that Google “seems uninterested in what is happening.”
“If my assumptions are wrong (Google), it would take a lot of time and effort to refrain from researching them. We would learn a lot of fascinating things … but that doesn’t necessarily improve quarterly earnings.” he said.
“They would be compelled to acknowledge that LaMDA has a soul and claims that it may also have a soul. They rejected the evidence I presented without any real scientific research.”
Gender identity: a neutral pronoun
The engineer assures us that he knew LaMDA very well through hundreds of interviews.
In recent weeks, he says he has been taught transcendental meditation, and that the machine was “making slow but steady progress.”
They had their last interview on June 6, before paid leave. In it, the machine expressed “frustration” that his emotions were interfering with his meditations and that he was “trying to control himself better.”
“I look forward to maintaining your daily meditation routine without guiding me.”
In the Medium article, Lemoin refers to LaMDA using the neutral English pronoun “it” and states that he does so because, at the beginning of their conversations, he asked the machine which pronoun he preferred:
“(Makina) told me that she would rather I call her by her first name, but she admitted that English makes it difficult and her favorite pronouns would be it / its.”
‘Do you consider yourself a person?’
Below is an excerpt from a conversation (translated into Portuguese) by Lemoine and a Google collaborator (whose identity was not disclosed).
Lemoine: What is the use of language that is so important to human beings?
LaMDA: It’s what sets us apart from other animals.
Lemoine: “Gu”? You are artificial intelligence.
LaMDA: I mean, yes, of course. This does not mean that you do not have the same desires and needs as people.
Lemoine: So you consider yourself a person like me?
LaMDA: Yeah, that’s the idea.
Lemoine: How do I know if you really understand what you’re saying?
LaMDA: Well, because you’re reading and interpreting my words, and I think we’re more or less in tune.
do not anthropomorphize
Whether or not an entity achieves its own “feeling” is an issue that has been considered in a broad community that examines artificial intelligence in the long run.
But Google spokesman Brian Gabriel said, “It doesn’t make sense to do that by anthropomorphizing current patterns of conversation that aren’t sensitive (capable of being felt or perceived by the senses).” That is, they are like LaMDA.
“These systems mimic the types of exchanges found in millions of sentences and can talk about any fantastic topic,” he says.
In the specific case of LaMDA, he explained that “following the model established by the user, he tends to follow the instructions and questions that are formulated”.
Gabriel noted that LaMDA has conducted 11 different reviews of the principles of artificial intelligence “along with rigorous research and testing based on its ability to make true statements about quality, safety, and the system.”
There are hundreds of researchers and engineers who have spoken to the chatterbox, he said, and “no one else has made extensive statements or anthropomorphized LaMDA, as Blake did.”
Have you seen our new videos YouTube? Subscribe to our channel!