You’re right that this piece doesn’t touch on the actual arguments for danger, it’s moreso an introduction to a possible reframing of how we talk about AI safety. Seeing and engineering around danger is essential, and I’d argue that’s exactly what better aimed framing would support. That discussion around arguments for danger is worth having separately.
The Native American comparison is where I’d push back, since Native Americans didn’t have agency in a way that labs and builders do. AI is something that humans are actively building, with ongoing decisions being made about architecture, training data, deployment contexts, and governance. The danger isn’t arriving from outside; it’s being constructed from the inside.
You’re right that this piece doesn’t touch on the actual arguments for danger, it’s moreso an introduction to a possible reframing of how we talk about AI safety. Seeing and engineering around danger is essential, and I’d argue that’s exactly what better aimed framing would support. That discussion around arguments for danger is worth having separately.
The Native American comparison is where I’d push back, since Native Americans didn’t have agency in a way that labs and builders do. AI is something that humans are actively building, with ongoing decisions being made about architecture, training data, deployment contexts, and governance. The danger isn’t arriving from outside; it’s being constructed from the inside.