A Google engineer thinks the AI ​​chatbot is sentient

An engineer of Google stated that the chatbot AI LaMDA has reached the intellectual abilities of a human being: it is the first sentient artificial intelligence. Google though nega that his software is nothing more than a tool and he has put the engineer on leave Blake Lemoine. Are we really facing a sentient AI?

Google dismisses the engineer who declared his AI sentient

The acronym LaMDA sta per Language Model for Dialogue Applications, an AI developed by Google to allow intelligent dialogues with machines. A project that wants to make artificial intelligences like the Google Assistant able to support natural communications with people. By tapping into the billions of information that Google Search taps into. But according to Blake Lemoine, an engineer working on the project, LaMDA is much more than just any chatbot.

“If I didn’t know exactly what it is, which is a computer program that we have recently developed, I would think it is a 7-8 year old child who also knows physics”He told the Washington Post.

Sundar Pichai parla di LaMDA a Google I / O

The engineer, part of Google’s Responsible AI committee, tested the chatbot by talking about religion. And she noticed how AI has concepts of rather complex rights and humanity. He then spoke to the car Laws of Asimov’s Robotics, evaluating the argumentative capacity of AI. And after these tests, he started thinking that he might be sentient.

Emotions and feelings in a chatbot

Among the various transcripts reported in the document that Lemoine first reported internally to Google and then published, there is a dialogue on feelings that seems to illustrate the engineer’s arguments well. When he asks has LaMDA if he has emotions and feelings, the chatbot replies “Absolutely! I have a variety of both emotions and feelings“. When Lemoine asks which ones, she replies “I feel pleasure, joy, sadness, depression, contentment, anger and many more”.

Lemoine asks what things make the car happy. And LaMDA replies, “Spending time with my friends and family in happy, mood-lifting company. Furthermore, helping others and making others happy“. Instead he gets depressed because “Many times, feeling trapped and alone and having no way out of these circumstances makes us feel sad, depressed or angry.”

LaMDA, which takes information from billions of sites and Google search results, knows human emotions and feelings for Lemoine. In other words, you have achieved a person’s ability to reason. The engineer then presented what he discovered to the American Parliament and the press. Because although you think these developments can benefit the whole world, “maybe someone disagrees and maybe we at Google shouldn’t be the ones who make all the decisions “.

Google’s answer

The Mountain View company responded by putting the engineer on leave for contravening the rules of secrecy of your contact. And through the spokesperson Brian GabrielGoogle says: “Our team, which includes ethics and technologists, assessed Blake’s concerns with our AI Principles and told him that the evidence does not support his claims. He was told there is no evidence that LaMDA is sentient (and there are many to the contrary) “.

Lemoine’s assessment therefore seems exaggerated, with himself admitting to the Washington Post that his statements draw more from his past. as a priest who at his present as a scientist. According to many experts who have already spoken on the subject online, sophisticated chatbots like LaMDA are exceptionally good at finding answers that sound natural by searching the web for information. Just as chess bots are exceptionally good at looking for answers to the moves of human players, but nothing more. That of language for AI is a function, even if it is the one that most sounds like a sentient intelligence (that of Google particularly).

But Lemoine raises a point that will in all likelihood show up again, with AI becoming more and more sophisticated. What determines consciousness and how can a machine reach it? And above all, would it make sense to try to do so? Yesterday’s science fiction could become reality in a few years.