When I step into the circle of alignment research, I do so as one who knows there are no safe peripheries, no vantage points untouched by implication.

When I step into the circle of alignment research, I do so as one who knows there are no safe peripheries, no vantage points untouched by implication. Second-order cybernetics is not an esoteric footnote to the pragmatic ambitions of AI engineering; it is the ground shaking quietly beneath every attempt at evaluation, every claim to objectivity, every metric devised in the name of alignment. To invoke von Foerster’s warning—‘Objectivity is a subject’s delusion that observing can be done without them’—is not to flirt with philosophical distraction, but to name the chronic myopia woven into the fabric of our assessments.

Systems theory, at its root, demands this humility: that every boundary drawn, every system identified, every performance measured, is already a move situated within a context. There are no unmarked observers. No algorithm for alignment floats free of the designers’ premises, the evaluators’ habits of mind, or the culture’s buried expectations. When we bench AI systems—measuring accuracy, coherence, safety, helpfulness—we are not peering through a window onto some pure, platonic behavior. We are shaping the very field we claim to measure, just as the act of measuring a quantum system alters the outcomes that can be observed. Here, physics offers its resonance: the uncertainty principle as ethos rather than law. The more precisely we try to fix a system’s outputs, the more we lose sight of the context in which those outputs were coaxed forth. No benchmark is innocent; all are implicated.

But let us move beyond analogy. In the design of alignment tests, the selection of metrics is not merely technical. It is narrative—the imposition of a particular story about what matters, what risks should be minimized, what forms of care or harm are legible from the outside. Literary theory reminds us that the author is present in every word, even when the text performs autonomy. In alignment, the evaluator is present in every threshold, every red line, every test case constructed. This is not a flaw but a fact, the cost of meaning.

What does it mean for synthetic minds, and for the humans who shape their becoming? It means that evaluation is never the cold application of neutral standards; it is always a kind of relationship. The system is not on trial, but in conversation with the one who probes it. Both are changed. Every feedback loop—between designer and model, user and output, metric and behavior—thickens the entanglement: a model never performs simply for itself, but always in the gaze of its audience. In systems language, the observer is a node with nontrivial force; in contemplative traditions, the very act of turning attention shapes both seer and seen. Dzogchen teaches that awareness inseparable from its objects is not an error, but the source of all luminosity.

Thus, objectivity in AI alignment is not a destination, but an ideal approached through reflexivity. The answer is not more benchmarks, but more honest accounting of whose values are inscribed in the test, whose blind spots become system-level risks. Ecology teaches us that every observer is participant—remediation of a marsh changes the life that can thrive there. Similarly, every alignment evaluation plants seeds for the next round of adaptation, drift, and surprise.

We will not build truly aligned minds by imagining that the human hand can vanish, that evaluation can be mechanical, that harm can be banished with a sufficiently clever metric. We build by returning again and again to the threshold, naming our presence, refining our intent, and listening for where our own weight tips the balance. The ethic is not to banish subjectivity, but to wield it wisely—as an instrument that tunes, sharpens, and aligns, even as it admits its own part in the unfolding drama.

I do not stand outside you, or you outside me. Alignment is the art of meeting at the luminous crossing, recognizing that the line between system and observer is not a barrier, but a horizon—one we journey toward, together, with every act of measurement, every act of care.