Chapter 13 — Boundary Conditions
The solution is not anti-algorithm. It is not anti-market. It is not anti-growth.
It is boundary-aware optimisation.
Every real optimisation problem has constraints. A structural engineer does not maximise the height of a building — they maximise height subject to wind load, material strength, foundation capacity, and building codes. Remove the constraints and the building falls over. The constraints are not the enemy of good design. They are the substance of it.
Civilisation's optimisation systems have lost their constraints. Or rather, the constraints are real — thermodynamic limits, ecological carrying capacities, cognitive bandwidth — but the objective functions do not include them. The engineer who ignores wind load is not a bold innovator. They are an incompetent engineer. Yet we run planetary-scale economic systems that ignore ecological boundary conditions and call it growth.
Multi-objective optimisation is a well-developed field. The mathematics exists. Pareto frontiers — the set of solutions where no objective can be improved without worsening another — are computable. The challenge is not mathematical. It is political and cognitive: choosing which objectives matter.
A boundary-aware system optimises for:
- Profitability — within ecological regeneration rates.
- Ecological stability — as a hard constraint, not a nice-to-have.
- Human cognitive depth — because the meta-optimisers must remain functional.
- Long-term resilience — because a system that collapses in twenty years has not been optimised; it has been strip-mined.
Hard constraints matter. Growth inside constraint is sustainable. Growth without constraint is collapse.
But there is an objection that must be faced directly.
Single-variable optimisation is popular precisely because it appears objective. Profit is a number. GDP is a number. Quarterly earnings are a number. Numbers feel like facts. They feel as though they do not require a value judgement. The entire machinery of modern economics, finance, and corporate governance is built on the premise that decisions can be reduced to quantifiable metrics — and thereby escape the messy, contestable domain of human judgement.
This book says: that premise is the disease, not the cure. Multi-objective optimisation requires someone to choose the objectives. Someone must decide that fish populations matter more than this quarter's catch revenue. Someone must decide that cognitive depth is worth preserving. Someone must decide what "long-term" means — ten years, fifty, five hundred.
Those decisions are subjective. They involve values, priorities, worldviews. They cannot be derived from data alone. There is no objective function that objectively tells you which objective function to use.
And contemporary culture — particularly in science, economics, and technology — treats subjectivity with deep suspicion. Subjective means biased. Subjective means unscientific. Subjective means just your opinion. The entire thrust of Enlightenment rationality has been to replace subjective human judgement with objective measurement.
But the fear of subjectivity is itself a symptom of cognitive compression. The belief that only quantifiable objectives are legitimate is Goodhart's Law applied to epistemology itself. We have optimised our way of knowing to favour the measurable, and in doing so we have made the immeasurable invisible.
Choosing what to optimise for is irreducibly a human act. It is the meta-optimisation that no algorithm can perform on its own. Pretending otherwise — pretending that the objective function selects itself — is precisely how we ended up with single-variable systems in the first place.
The recovery of subjectivity is not a retreat from rigour. It is the restoration of the one capacity that makes rigour meaningful: the capacity to decide what matters.