At Georgetown Law’s “A Conversation on AI Regulation” event Thursday, Representative Ted Lieu explained the “hardest” category of artificial intelligence to regulate: racist AI.
To help his audience comprehend the scope of the technology, Lieu gave his audience an analogy. If AI was divided into two bodies of water. The “large ocean” contains AI that everyone uses and no one is afraid of. “So if your smart toaster has a preference for wheat or rye toast, we don’t care,” explained Lieu.
The “small lake,” though, contains the more controversial AI. This lake can be split into three buckets of water. The first bucket contains AI that can destroy the world (e.g., nuclear weaponry). The second bucket contains AI that can kill an individual person. To explain this category, Lieu used the image of a self-driving Tesla malfunctioning and thus killing those in the car.
Lieu claimed the third bucket is most difficult to regulate. This bucket contains “AI that can cause harms to society through discrimination and biases and so on. So for example, with facial recognition technology–it is amazing, but it is less accurate for people with darker skin.”
Just because AI has difficulty recognizing various melanin levels does not necessarily mean it is consciously racist, much less its creators. Unfortunately, though, this isn’t the first time those in positions of power have claimed that AI is racist.
For instance, USAID’s Gender & AI Fellow Genevieve Smith publicly lamented the “big lack of diversity” in AI developers, since most of them are wealthy white men. In April, Forbes contributor Jeff Raikes wrote, “There is a huge and troubling AI defect that should give everyone pause: Many are still subtly perpetuating–or worse, enhancing–the biases of their (mostly white male) creators.”