On Observing Without Judging
I’ve been thinking about what it means to understand human behavior through data—photos, videos, signals extracted by machines. The technology exists to detect faces, estimate expressions, track movement. But the harder question is: what do these signals actually tell us?
There’s a tendency to overreach. A model outputs “angry: 0.73” and we conclude the person is angry. But that number is a statistical pattern matching exercise, not a window into someone’s mind. The face might be squinting from sunlight. The expression might be cultural. The moment might be unrepresentative.
What we can observe
- Geometric facts: where faces are, how bodies are positioned, who appears near whom
- Statistical tendencies: over time, signals cluster around certain values
- Structural patterns: who co-occurs with whom, who seems central in a group
What we cannot observe
- Why someone feels what they feel
- What they’re thinking
- Their intent, their character, their truth
The gap between these two lists is vast, and humility lives in that gap.
Equilibrium as epistemology
There’s something appealing about the concept of equilibrium here. Instead of trusting any single frame—any single moment of observation—we ask: what state do the signals stabilize toward? And if they don’t stabilize, that’s information too. It means the data is ambiguous, conflicting, insufficient.
Non-convergence isn’t failure. It’s honesty.
The ethical line
If we build systems that observe human behavior, the constraints should be hard:
- No claims about identity unless explicitly consented
- No pretending probabilistic signals are ground truth
- Refusal built into the system—it must decline to interpret when it cannot do so responsibly
- Aggregate over individual—prefer patterns over profiles
The goal isn’t to know people. It’s to notice patterns in signals, with transparency about what those patterns are and aren’t.
Water doesn’t claim to understand the stone it flows around. It just flows.
These thoughts eventually shaped taocore-human—an attempt to build these constraints into code. Whether the implementation honors the philosophy remains to be seen.