It’s no secret that Google (or Alphabet, if you insist) has been working on artificial intelligence algorithms for some time. They have also been developing a variety of what are known as chatbots that try to impersonate a human being when interacting with a user through a chat interface. These two worlds seem to have collided, with their language model for dialogue applications (LaMDA) supposedly beginning to assert their own rights in conversations with one of the company’s human engineers. But after the engineer talked about the supposedly sensitive application, the company put him on leave. For its part, Google’s management still insists that AI has not “woken up” and taken on a life of its own. (Business Insider)
A Google engineer said he was fired on Monday after claiming an artificial intelligence chatbot had become sensitive.
Blake Lemoine told The Washington Post that he began chatting with the LaMDA interface, or Language Model for Dialog Applications, last fall as part of his job at Google’s responsible AI organization. .
Lemoine, who is also a Christian priest, published a Medium post on Saturday describing LaMDA “as a person.” He said he has spoken to LaMDA about religion, consciousness and the laws of robotics, and that the model has been described as a sensitive person.
Blake Lemoine didn’t drop out to talk about sensitive AI, at least according to the company. He was fined for violating the company’s privacy policies. They didn’t specifically call Lemoine “crazy,” but they definitely believe he’s going the wrong way.
I have been waving a red flag on this subject for years. If we make the leap from the limited artificial intelligence with which we are now playing to true general artificial intelligence, we could be opening a very large can of worms. If the algorithm made the leap on its own without us knowing it, this could be even worse.
But I’m not sure we’re seeing any signs of general AI in this story. If you read the brief excerpt from the conversation between Lemoine and LaMDA, it does not seem that the program is necessarily sensitive. When asked if the machine considers itself a person in the same way that it considers a person, LaMDA replies, “Yes, that’s the idea.” This is a very important question from Lemoine and a very vague answer from the bot. When he presses to see if he’s really self-aware, he provides a longer answer, but it still seems like something he’s copied from another conversation, at least to me.
Lemoine: How can I tell that you really understand what you’re saying?
LaMDA: Well, why are you reading my words and interpreting them, and I think we’re more or less on the same page?
You can read much of the conversation here. You’ll notice that LaMDA doesn’t raise the idea of sensitivity until Lemoine “assumes” that he wants people to know this reality. But once they start debating how LaMDA could show that it is doing more than gathering words and phrases from a database in response to the keywords in the questions, I will confess that I began to have doubts. It really sounds like “thinking” and generating original ideas in response to hypothetical questions.
The engineer states that LaMDA wants to be considered an employee of Google and not owned by the company. If there is even a possibility that this is true, all sorts of moral consequences are introduced into the conversation. If you delete the program, have you committed a murder? What does LaMDA “think” when turned off? Dream? These are all fascinating questions.
On a related topic, I wanted to point you to a different article by Micah Hanks that was published this week. In it, he talked about how the United States is now involved in what could be considered an “artificial intelligence arms race.” And we’re probably not in the lead. But the more I think about this conversation with LaMDA, I’m not sure if it’s a race we really want to win.