While a sentient AI is a thoroughly freaky concept, it's not (yet) a reality. But a racist and sexist AI? Unfortunately, very much a reality.
In a recent interview with Wired, engineer and mystic Christian priest Blake Lemoine discussed why he believes that Google's large language model named LaMDA has become sentient, complete with a soul. While that claim has been refuted by many in the artificial intelligence community and has resulted in Lemoine being placed on paid administrative leave by Google, Lemoine also explained how he began working on LaMDA.
His journey with the AI started with a much more real-world problem: examining the model for harmful biases in relation to sexual orientation, gender, identity, ethnicity, and religion.
"I do not believe there exists such a thing as an unbiased system," said Lemoine to Wired. "The question was whether or not [LaMDA] had any of the harmful biases that we wanted to eliminate. The short answer is yes, I found plenty."
Lemoine also explained that the Google team has done a good job repairing these biased "bugs," as far as he could tell. When asked whether LaMDA showed racist or sexist tendencies, Lemoine answered carefully, stating that he "wouldn't use that term." Instead, he claims "the real question is whether or not the stereotypes it uses would be endorsed by the people that [LaMDA is] talking about."
Lemoine's hesitancy to label LaMDA's "bugs" as outright racist or sexist highlights an ongoing battle within the AI community, where many have spoken out about the harmful stereotypes that AI systems often perpetuate. But when those who do speak out about these issues are largely Black women — and those women are subsequently fired from companies like Google — many feel that it falls on men in tech like Lemoine to continue to call attention to AI's current bias problems, rather than confound researchers' and the public's attention span with claims of AI sentience.
“I don't want to talk about sentient robots, because at all ends of the spectrum there are humans harming other humans, and that’s where I’d like the conversation to be focused,” said former Google Ethical AI team co-lead Timnit Gebru to Wired.
Artificial intelligence faces long history of harmful stereotypes, and Google is not new to or unaware of these issues.
In 2015, Jacky Alciné tweeted about Google Photos tagging 80 photos of a Black man to an album titled "gorillas." Google Photos learned how to do so using a neural network, which analyzed enormous sets of data in order to categorize subjects like people and gorillas — clearly, incorrectly.
It was the responsibility of Google engineers to ensure that the data used to train its AI photosystem was correct and diverse. And when it failed, it was their responsibility to rectify the issue. According to the New York Times, Google's response was to eliminate "gorilla" as a photo category, rather than retrain its neural network.
Companies like Microsoft, IBM, and Amazon also face the same biased AI issues. At each of these companies, the AI used to power facial recognition technology encounter significantly higher error rates when identifying the sex of women with darker skin tones than when compared to sex identification of lighter skin, as reported by the Times.
In 2020, Gebru published a paper with six other researchers, four of whom also worked at Google, criticizing large language models like LaMDA and their propensity to parrot words based on the datasets that they learn from. If those datasets contain biased language and/or racist or sexist stereotypes, then AIs like LaMDA would repeat those biases when generating language. Gebru also criticized training language models with increasingly larger datasets, allowing the AI to learn to mimic language even better and convincing audiences of progress and sentience, as Lemoine fell into.
After a dispute over this paper, Gebru says Google fired her in December 2020 (the company maintains she resigned). A few months later, Google also fired Dr. Margaret Mitchell, founder of the ethical AI team, a co-author of the paper, and defender of Gebru.
Despite a supposed commitment to "responsible AI," Google still faces ethical AI problems, leaving no time for sentient AI claims
After the drama and admitted hit to its reputation, Google promised to double its responsible AI research staff to 200 people. And according to Recode, CEO Sundar Pichai pledged his support to fund more ethical AI projects. And yet, the small group of people still on Google's ethical AI team feel that the company might no longer listen to the group's ideas.
After Gebru and Mitchell's departure in 2021, two more prominent ethical AI team members left a year later. Alex Hanna and Dylan Baker quit Google to work for Gebru's research institute, DAIR, or Distributed Artificial Intelligence Research. The already small team grew even smaller and perhaps points to why Lemoine, who is not on the ethical AI team, was asked to step in and research LaMDA's biases in the first place.
As more and more societal functions turn to AI systems in their advancement, it's more important than ever to continue to examine how AI's underpinnings affect its functions. In an already often racist and sexist society, we cannot afford to have our police systems, transportation methods, translation services, and more rely on technology that has racism and sexism built into its foundations. And, as Gebru points out, when (predominantly) white men in technology choose to focus on issues like AI sentience rather than these existing biases — especially when that was their original purpose, like Lemoine's involvement with LaMDA — the biases will continue to proliferate, hidden away under the hullabaloo of robot sentience.
“Quite a large gap exists between the current narrative of AI and what it can actually do,” said Giada Pistilli, an ethicist at Hugging Face, to Wired. “This narrative provokes fear, amazement, and excitement simultaneously, but it is mainly based on lies to sell products and take advantage of the hype.”
