Prologue

The Stevenson Test states that nature does not allow for appeals. Civilizations are the product of world machines humanity has built to generate systems for appeals.

Venkatesh Rao, Aneesh Sathe and SIGPSY (the Special Interest Group in Psychohistory), with the World Machines and Prime Radiant projects, have been mapping out humanity's recent efforts and discovered New Nature: an AI-powered phenomenon where pre-commodification re-wilds the world. [a] In this jungle, a small sliver of light fell onto my industry, left behind on the forest floor as tech-based businesses reached for the sky.

Along with agriculture, construction has concerned itself with providing the physical backing needed to drive the creation of increasingly complex machines. Unlike agriculture, which provides uncertain outputs that are easy for outsiders to measure, construction has been plagued since the days of Hammurabi by its need to provide certain outcomes that are hard for the non-builder to evaluate. The Babylonians resolved the issue with the simplest solution: strict liability for the builder. The homeowner could appeal for the builder to be liable for all damages; the builder could not appeal.

While our society's lack of appetite for murdering the sons of general contractors in eye-for-an-eye legal remedies is welcomed by construction workers and their families, moving away from the Mesopotamian solution to the principal-agent problem has required a far more complex framework.

These days the tension is distributed through multiple protocols: training, certification, inspection, licensing, engineering, claims, and contracts. This system for resolving appeals and counterappeals has developed over thousands of years and, in common law jurisdictions, resists New Nature through a single legal fiction: the reasonable person. [b]

The Gramsci monster in our story, the one that spawns in the time between machines, is the reasonable bot.

The current tech stack allows AI bots to design buildings, write specifications, read specifications, and bid on requirements. They may not be able to do this well, but the tech keeps improving. When it reaches parity with human estimators, the challenge will come not from calculations but from liability. How could the productivity gains from the automation of the administrative portions of construction procurement be realized if humans must always review the work of bots?

For long-time followers of the Protocol Institute, the answer may have been obvious: standardization. For those who, like me, needed to take the long route, a mini-Prime Radiant project focused on Bot Procurement may be needed to arrive at the same answer: Procurement Machines.

AI-first procurement will need two tools: [c]

• a shift away from the primacy of human-centric documentation to standardized instructions generated for machine readability. (Yes, for those in the know, it would mean getting rid of PDFs — no one really liked looking at paper sheets on a screen anyhow.)
• a standardized bot for reading the specifications.

With these tools, engineers and architects could still certify the canonical copy, the court could use the standardized bot as a surrogate for a reasonable bot, and bidders would be able to lower their estimating costs. [d]