Wiser! Essay: When Timnit Gebru raised serious questions about Google's use of AI, she lost her job. One year later and Gebru has launched an AI Ethics Research team.
BackStory: This time last year, Timnit Gebru tweeted that she had been “immediately fired” from her role as co-lead of Google’s AI ethics team.
Exactly one year later, Gebru has announced her new venture: the Distributed Artificial Intelligence Research Institute (DAIR). This global institute will focus on calling out the harms of artificial intelligence, especially it affects the most vulnerable groups. Through both research, and eventually consulting, Gebru hopes to push the tech industry toward a more ethical approach to AI deployment.
To recap, in December 2020, tensions between Gebru and the Google leadership team came to a head after a paper co-authored by Gebru raised some thorny questions for Google. The paper challenged Google (and Big Tech) about whether AI language models may be too big.
In simple terms, what this means is that these artificial intelligence systems are being trained on staggeringly huge amounts of data. They are now extraordinarily good, under the right conditions, at producing what looks like convincing, meaningful results. However, Gebru is also known for her research in AI biases where she found facial recognition to be less accurate at identifying women and people of colour, i.e., finding that AI systems can be discriminatory.
TakeAway: The point is that Gebru was asking the question as to whether Google, and all tech companies for that matter, were doing enough to reduce potential risks.
The paper, entitled, "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”, essentially questioned inherent biases in large language models. For Google’s AI team, this focused criticism on their AI language model called BERT, which was launched in 2018. The AI algorithms in BERT were found to improve 10% of searches by their ability to understand language context. As a result, BERT was incorporated into the Google search engine.
Now, to understand the significance of a 10% improvement in search results is to understand the role of advertising and its relationship to the Search segment of Google’s business. Google's revenue in 2020 was $181 billion, of which $146 billion came from advertising revenues. Google dominates the global Search market. In the US alone, more than 4 out of every 5 searches on a desktop are made with Google. Whilst it is over 9 out of every 10 on a mobile (largely thanks to Google's ownership of Android).
With DAIR, Gebru has said she wants to use her past experiences in Big Tech to create an environment where researchers can openly communicate their findings of AI’s harms, and where research subjects are acknowledged or paid.
Since the tech world skews white and male, and AI research skews Western, DAIR will recruit staff from global communities that are underrepresented and underserved in the tech world—people who can help build beneficial AI applications that may not be created otherwise.
Never Miss an Issue!
Sign up for the free membership of the Wiser! Newsletter and receive your copy of the weekly newsletter every Friday.
Premium Members get additional content with deep-dive INSIGHTS on the 1st and 3rd Tuesday of every month.
We read the paper that forced Timnit out of Google. Source: Technology Review
Open-sourcing BERT, State-of-the-Art Pre-training for Natural Language Processing. Source: Google
BERT increases 10% of searches. Source: The Verge
Google Stats and Facts. Source: Statista
The Wiser! Newsletter is only made possible because of the generous subscriptions and donations from readers like you. The easiest way to make a donation is to Buy Me A Coffee. Each virtual coffee costs €2 and you can buy as many or as few as you like. Try it out here...
Sharing is caring...