This article is from WeChat official account:Geek Park (ID: geekpark) , author: face face, original title: “Google AI then exposed black female employee turnover, which is a problem for all technology companies will face a” head Figure: vision China span >
At the moment when the AI industry is developing rapidly around the world, only such a large-scale serious discussion broke out within Google.
On February 20th, Mitchell, the founder of Google’s Ethical AI (AI ethics) team, posted a short statement on Twitter: “I am fired”, which is exactly the same as what happened to Timnit Gebru, another female researcher last year. , And these two women are the two former co-heads of Google’s Ethical AI department.
At the end of last year, Gebru tweeted that she was expelled from the company. After she was asked to withdraw the paper because of a conflict between the publication of a paper and the upper management of the company, she expressed in an internal email that she expressed her views on the company’s racial prejudice. Dissatisfied, was subsequently fired.
From an outside perspective, it seems difficult not to link Mitchell’s dismissal to the Gebru incident.
The chatter on Twitter
This Gebru is the co-head of Google’s AI ethics department and one of the very few black female researchers within Google (according to the company’s 2020 annual diversity report, 3.7% of Google employees are black), and she is also a Black Co-founder of In AI, dedicated to the equal rights of blacks in the field of algorithms.
The introduction of the paper reads: “We asked whether we have thought enough about the potential risks associated with developing these risks and the strategies to mitigate them.”
Google AI first invented the Transformer language model in 2017. In 2018, the BERT language model launched on this basis caused a sensation in the field of artificial intelligence. At the same time, it has been widely used in the training of Google search engine and has become Google’s economy. An important source of income.
This paper is a concern about Google’s current most profitable AI research, and this concern is not groundless, and the ethical bias involved in AI is not the first time it has appeared.
In 2015, Google Photos labeled two dark-skinned people as “gorillas.” Google promised to “take immediate steps” to prevent repeated mistakes. But then in order to solve this problem, Google directly removed the label of this term from the search results. Wired magazine tested more than 40,000 animal pictures on Google Photos and finally came to the conclusion: Google no longer labels any pictures as gorillas, chimpanzees, monkeys, even pictures of such animals. .
Last year, the most shocking AI bias was probably the PULSE algorithm that recognized Obama as a white male. This algorithm developed by Nvidia uses a technology called “upscaling” to process visual data and generate a new high-resolution image from the original low-resolution image.