My grandfather, Sylvester Sviokla, was a man of Lithuanian immigrant grit who treated every stone he unearthed in his garden as a potential future asset, burying them for a rainy day. His disciplined frugality eventually paved the way for my own education. Today, many modern corporations are making the opposite mistake: they are discarding their own cognitive rocks.

Every time a subject matter expert overrides an AI-generated draft, tweaks a recommendation, or corrects a model's output, the firm is paying for a high-value lesson that is almost always lost. I define this discarded wisdom as cognitive scrap. When captured, this byproduct of artificial intelligence becomes the raw material for an asset that grows more valuable with every iteration.

The Legacy of XCON and the Cost of Static Rules

The history of expert systems offers a cautionary tale. In 1986, Digital Equipment Corporation (DEC) revolutionized its workflow using XCON, an expert system that automated VAX computer configurations. By shifting from manual processing to this automated framework, the company boosted accuracy from 65% to over 95%, eventually eliminating an entire manufacturing phase and saving $15 million in capital expenditure.

However, this efficiency came with a heavy price. By 1987, the system relied on 6,200 rules, half of which required annual updates. Maintaining this system demanded a dedicated team of 30 to 40 people, proving that systems built on frozen logic struggle to keep pace with a dynamic, evolving world.

The Bottleneck of Tacit Knowledge

The fundamental challenge lies in the nature of expertise. As Michael Polanyi famously noted, we possess knowledge that defies articulation. Expert editors often identify errors intuitively, without needing to consult a rulebook. Forcing these experts to explain their reasoning creates a bottleneck, as the resulting narratives often omit the true logic behind their decisions.

Modern AI has not eliminated this bottleneck; it has merely shifted it from rule-writing to output evaluation. This shift is a massive opportunity. Critique is significantly faster and cheaper than articulation. By capturing these expert rejections, firms can leverage the core principles of reinforcement learning with human feedback to refine their systems.

  • Morgan Stanley’s Success: By treating expert judgment as a daily standard rather than a one-time review, the firm scaled its operations from 7,000 inquiries to a vast corpus of 100,000 documents.
  • The Toyota Model: Much like the Toyota Production System treats defects as opportunities for process improvement, AI managers must classify and trace the causes of cognitive scrap to build better instructions, tests, and training examples.

Building Compounding Cognitive Capital

Static rules eventually go stale, but a robust learning loop ensures that the asset—the loop itself—continues to compound in value. Properly managed, AI transitions from a simple tool into cognitive capital, an asset that becomes more efficient with every cycle of correction.

Strategic Implementation

To begin building your own cognitive capital, follow these steps:

  • Identify a high-impact, recurring decision shared across departments.
  • Log every correction alongside a reason code to facilitate pattern analysis.
  • Appoint a lead to convert these patterns into actionable, tested improvements.
  • Reallocate your human talent to solve more complex, high-value problems.

My grandfather buried his rocks because he understood their hidden utility. It is time for your organization to stop discarding its cognitive scrap and start building its future.