Credit: Kimberly White / Stringer (Getty Images)
Google can’t get out of its own way.
A champion of diversity in tech, Gebru’s previous research notably discovered gender and skin-type bias in facial recognition technology.
Gebru flagged issues with Google’s large language models
According to the MIT Technology Review, these models — AIs trained on staggering amounts of text data — are “key to Google’s business.”
Here are some of Gebru’s latest research findings:
- Environmental and financial costs: Large AI products use a lot of energy. One of the language models underpinning search creates as much CO2 as a round-trip flight between NY and SF.
- Bias in large data sets: Language models are skewed toward countries and languages with large internet footprints (e.g., English). Other cultural norms may not be reflected in technology that uses these models.
- Hard to audit: The datasets are so big, it’s very difficult to even see how they might be biased (or how to fix them).
Did Google want to suppress these findings?
Many AI ethicists believe Google “pushed her out because of the inconvenient truths that she was uncovering about a core line of its research” writes the MIT Technology Review.
A letter in support of Gebru has been signed by 1.5k+ Google employees and 2.1k+ academics in the AI field.
(Read more: Tech writer Casey Newton shared the email exchange between Gebru and her boss, Jeff Dean, that led to her departure.)