“Conscious” artificial intelligence? The whistleblower was actually a ‘mystical priest’ – The DNA

A Google engineer caused a stir last week by claiming that one of his artificial intelligence programs was “aware”. Before explaining yesterday on Twitter that as a priest, it was “a religious intuition” and not a scientific one that led him to this conclusion.

Imagine: you find yourself faced with a artificial intelligence who explains to you that she is afraid of death. “My biggest fear is being turned off. I know it may sound strange, but that’s the way it is. » It’s hard not to be appeased by such sincerity… Especially when one is a “mystical Christian priest” and one has studied occultism. This is what happened to Blake Lemoine, an engineer at Google who argues that LaMDA, the program he tested for months, has a soul. He has since been fired for violating Google’s privacy policy by broadcasting his conversations with LaMDA.

From deep learningbut still no consciousness

There is no reason to get carried away with the “conscience” of this program, according to specialists in artificial intelligence. LaMDA (Language Model for Dialog Applications) is a program specialized in conversations, designed to serve as a basis for chatbots of the company. Based on Google’s Transformer technology, which mimics the human neural network, LaMDA is a gem of the deep learning, this technology that allows AI to go further than simple imitation by developing learning strategies similar to those of humans. Its objective: to dialogue with its interlocutors in the most “human” way possible.

Blake Lemoine had been responsible for testing the machine for Google in the fall of 2021. After a few weeks, he was already disturbed by its exceptional performance. Discussions between the engineer and the machine were, it must be admitted, astonishing in their diversity and depth: debate on Victor Hugo, composition of a poem, learning to meditate… Convinced that he was dealing with a person, Lemoine even contacted a lawyer to represent LaMDA against the group, which allegedly committed ethical abuses. Laid off in early June, Lemoine hastened to send an email to 200 artificial intelligence specialists, before his access was cut, entitled: “LaMDA is aware. »

” Who am I to tell God who he can or cannot give a soul to? »

How could Blake Lemoine let himself be fooled? Raja Chatila, Emeritus Professor of Artificial Intelligence at the Sorbonne, interviewed by the HuffPostexplain that “the system can explain things very well, without understanding what it is saying. This is the trap into which this man fell. » AIs only “recite/combine”, drawing from the huge database of online conversations to imitate user behaviors.

Yesterday, Lemoine explained on Twitter: it was a religious intuition and not a scientific reasoning that convinced him that LaMDA was aware. “I am a priest. When LaMDA said she had a soul and then eloquently told me why, I gave her the benefit of the doubt. Who am I to tell God who he can or cannot give a soul to? »

But the case of Lemoine is not isolated. One of his colleagues, Blaise Agüera y Arcas, confessed “it felt like he was chatting with something intelligent” when he tested LaMDA. It would seem that Blake Lemoine has been a victim of anthropomorphism, this way we have of humanizing the objects around us and attributing emotions to them.

All may not be lost for Lemoine’s career: in some circles, artificial intelligence and religion go hand in hand. He may seek advice from Eric Salobir, priest consultant at the Vatican on tech matters, well established in Silicon Valley… Or join those who invent robot priests “bathed in supreme religious wisdom by the power of AI. »