Governor Newsom,
In your veto of SB 1047, you rightfully demanded that California's approach to AI regulation be based on "empirical trajectory analysis" rather than fear. You asked for a framework that prioritizes science and evidence over broad, precautionary bans.
We agree. The era of debating the possibilities of AI safety is over. It is time to implement the realities.
At The Algorithmic Consistency Initiative (ACI), we reject the false binary between "stifling innovation" and "accepting risk." We propose the "Third Way": a rigorous, engineering-first approach that transforms AI safety from a philosophical art into a hard science.
We Are Not Guessing. We Have the Playbook.
While the industry struggles with the "Black Box" problem of unpredictability, ACI has developed the operational Playbook for Constitutional AI. This is not a theoretical policy paper; it is an implementation guide for converting abstract civil rights laws into executable code.
Our framework provides the specific mechanisms required to operationalize the NIST AI Risk Management Framework:
1. GOVERN via Constitution
We replace vague "human feedback" (which leads to sycophancy and bias) with explicit Normative Constraints. We draft the "Constitution"—the prioritized set of rules—before a single line of code is trained.
2. MEASURE via Error Rates
We do not rely on "vibes." We calculate the Constitutional Error Rate (CER), measuring exactly how often a model fails to adhere to its governance clauses.
3. MANAGE via Auditable Traces
We abolish the "Black Box." Our architecture forces the AI to produce a Reasoning Trace—a step-by-step, human-readable log explaining why it made a decision, citing the specific Constitutional clause it followed.
From "Foreseeable Defects" to "Engineered Trust"
Current industry standards rely on Behavioral Mimicry—training models to mimic human raters. Legally, this creates a "Digital Mirror" that reflects our worst biases and constitutes a foreseeable design defect.
ACI offers the remedy: Constitutional AI. By moving from "trying to be safe" to "proving compliance," we unlock a future where California's AI sector is not only the most powerful in the world but the most legally defensible and trustworthy.
The Offer
We are done talking about what should happen. We are ready to show you how it happens.
ACI invites your administration to review our Constitutional AI Playbook. We stand ready to partner with the state to pilot these standards, proving that you can have innovation and safety—if you have the courage to engineer it.
Sincerely,
Alberto Rocha, Angel Luevano, & Argentina Davalos-Luevano
Co-Founders, The Algorithmic Consistency Initiative (ACI)
Engineering Civil Rights into the Digital Age
AlgorithmicConsistency.org
