Ex-Google Engineer Claims Artificial Intelligence Robot Is Racist
Blake Lemoine is the engineer who told the world that Google’s artificial intelligence robot named LaMDA, Language Model for Dialogue Applications, was sentient. However, he told Insider that the most alarming point is not that but the fact that the AI bot had certain biases on topics involving race and religion.
Recall that Lemoine had worked for Google and that his job was to ensure that LaMDA does not generate any discriminatory language or hate speech. He was placed on leave in June following a leak of documents to an unnamed US senator. It appears that evidence that the bot discriminated on the basis of religion was contained therein. Since this episode, Lemoine has been fired.
What does the engineer say?
According to Lemoine, LaMDA is racist because when he was pushed to imitate a black man from Georgia, he said: ” Let’s get some fried chicken and waffles “. Then faced with a question about the different religious groups, he allegedly replied that Muslims are more violent than Christians.
Lemoine entirely blames the prejudices of AI on the fact that the engineers who designed it do not have diversity. Lemoine said the people building LaMDA have never faced poverty, never set foot in communities of color, etc. He says these engineers don’t have in mind the impacts this AI could have on people other than them. To remedy all this, Lemoine recommends the collection of data in the external environment but not on the Internet.
What Google says
Google claimed that LaMDA underwent 11 rounds of ethics reviews following Blake Lemoine’s claims. Moreover, its development was detailed in a research document published by the company on January 21, 2022.
Separately, a Google spokesperson by the name of Brian Gabriel spoke about the situation. He told Insider that despite the fact that many organizations have developed and released similar language models, Google has decided to take a cautious approach. He added that this was to better address valid concerns whether in terms of fairness or factuality.