We think we know AI dangers because we know software bugs are bad & fixable. This is wrong! AI system flaws are fundamentally different from traditional software issues. Applying old "bug fix" logic to AI is misleading and stops us from seeing the real risks. https://boydkane.com/essays/boss