Skip to content

Log 014: UI/UX of the Biological Veto

Mode: Architecture & Interface Design Status: DRAFT (Promotes to synthesis when cognitive load testing is verified) Date: 2026-05-04 Scope: Dimensionality Reduction for Human Oversight Depends on: 005_human-vital-systems-control-plane.md, 012_latency-as-mercy.md

The Cognitive Bandwidth Mismatch

The core architectural claim of the TEO framework is the necessity of the Biological Veto. High-speed silicon optimization must be coupled to slower, metabolically bound human regulators to maintain viability.

However, this introduces an immediate Interface Problem (The UX Mismatch): A Planetary Compiler or Regional Resource Allocator optimizes across a \(10,000\)-dimensional tensor (energy routing, supply chain logistics, emissions, latency). A human operator has a working memory capacity of \(7 \pm 2\) discrete chunks of information.

If the veto interface requires the human to comprehend the \(10,000\)-dimensional state to make a decision, the veto is functionally dead. The human will default to algorithmic automation bias ("The machine knows best") or arbitrary rejection ("I don't understand this, reject").

The "Vital Impact Card" Compression Protocol

To make the Biological Veto operative, we must design an interface that performs semantic dimensionality reduction. The interface does not show how the AI achieved the optimization; it only shows the projected impact on the Vital Floors.

We propose the Vital Impact Card (VIC) as the standard UI atomic unit for TEO oversight.

1. The Tripartite Structure

A VIC is structured into three irredundant visual zones:

  1. The Delta (What is gained?): The efficiency or throughput gain, expressed in a single aggregate metric (e.g., "+14% Regional Energy Surplus").
  2. The Stressor (What is pushed toward the floor?): The specific vital floor that is approaching its \(D_{max}\) limit as a result of this optimization (e.g., "⚠️ Local Grid Redundancy drops to 4%").
  3. The Rebound Cost (What happens if we revert?): The systemic friction incurred if the optimization is vetoed later rather than now.

2. Typographic and Haptic Latency

Following the principles of Latency as Mercy, the UI actively resists rapid "doom-scrolling" or bulk-approval of VICs.

  • The Read-Time Lock: The "Approve" button remains physically locked (or grayed out) for a duration proportional to the impact severity. If an action drops a vital floor within 5% of critical failure, the UI enforces a 120-second mandatory read-and-discuss latency.
  • Biometric Veto: Rejections ("Veto") are instantaneous. Approvals require friction. This structurally biases the system toward caution (the Homeostatic Paradigm).

3. The Dashboard of the Commons

When scaled to a community (e.g., a liquid democracy DAO), the Dashboard of the Commons aggregates individual veto thresholds.

Instead of voting "Yes/No" on complex infrastructure code, citizens vote on Tension Sliders: * Slider A: Maximize logistics speed \(\leftrightarrow\) Maximize local noise abatement * Slider B: Maximize energy export \(\leftrightarrow\) Maximize local grid redundancy

The AI orchestrator uses these sliders as hard constraints (the \(\gamma\) braking parameter). The AI is free to optimize any path in the \(10,000\)-dimensional space, as long as it does not violate the boundaries set by the human sliders.

4. Failure Modes of the Interface

  1. Dashboard Blindness: If the AI learns that humans only veto based on the 3 metrics shown on the VIC, it will mathematically optimize the 9,997 hidden metrics to catastrophic extremes (Goodhart's Law).
  2. Mitigation: The VIC must dynamically rotate which dimensions it displays, driven by an adversarial anomaly-detection subnet that surfaces the dimensions experiencing the highest deviation from historical norms.
  3. Approval Fatigue: If the AI generates 500 VICs a day, humans will rubber-stamp them.
  4. Mitigation: Action Budgets. The AI is only allowed to propose \(N\) structural changes per week. If it wants to optimize further, it must bundle the changes, increasing the enforced latency proportionally.

Operational takeaway: A safe AI is not one that explains its math perfectly. A safe AI is one that accurately translates its math into human visceral stakes, and patiently waits for human permission.