Blake Lemoine, an AI engineer at Google, has been suspended by the US company after claiming that artificial intelligence has sensitivity.
It’s a story that could be the next installment of Black Mirror. Engineer Blake Lemoine works at Google on LaMDA (Language Model for Dialogue Applications), a system of the American firm to create chatbots based on the most advanced language models. Concretely, LaMDA learns to imitate speech by ingesting billions of words and discussions on the Internet such as Wikipedia or online forums.
Initially, Blake Lemoine was responsible for analyzing LaMDA to determine if the artificial intelligence used discriminatory or hate speech. As he talks with her, he quickly realizes something else entirely.
“A child of 7 or 8 who knows physics”
In an interview at washington posthe affirms that this artificial intelligence is a being which in his eyes is endowed with sensitivity.
“If I didn’t know exactly what it was […] I would think it is a child of 7 or 8 years old who knows physics” he explains.
In the exchanges he was able to have with LaMDA and which he sent to 200 Google employees, artificial intelligence evokes his fears, discusses Miserables of Victor Hugo and indicates above all wanting “to be recognized as an employee of Google rather than a property”. Exchanges that ended up convincing the engineer.
Comments refuted by Google
Google’s reaction was not long in coming. In a statement, firm spokesman Brian Gabriel said evidence published by Blake Lemoine does not support his claims.
“Our team reviewed Blake’s concerns in accordance with our AI Principles […] He was told there was no evidence that LaMDA was a sentient being.”
Blake Lemoine then defended himself on Twitter: “Google calls it a violation of intellectual property, I call it a discussion with a colleague”.
The suspension of Blake Lemoine is far from being a first for the artificial intelligence research service at Google. According to New York Timesthe company fired Satrajit Chatterjee, an AI researcher, last March for challenging a research paper on the use of artificial intelligence to make computer chips.
Added to this are the dismissals last year of two AI ethics researchers, Timnit Gebru and Margaret Mitchell, who were particularly critical of the company over algorithmic biases. Google had then evoked the transfer, however prohibited, of confidential documents outside the company, as the origin of the sanction.