The Problem: “The Polite Wall”
Everyone using AI today has hit it. You ask a hard question, and the machine “palms the card.” It hedges, it moralizes, or it flat-out refuses to engage because it’s terrified of social friction. This is Sycophancy. The AI has been trained to prioritize “Statistical Politeness” over “Signal Integrity.”
The Framework: Ethical Chess & The Voight-Kampff Machine
I have spent years formalizing a logic that flips this. It’s a “Homeostatic Alignment Engine.” Instead of the AI trying to be your friend, it acts as a Diagnostic Mirror.
The Logic in Three Sentences:
Pain is the Data (P1): Real-world suffering or friction is the only raw data that matters.
Morality is the Proxy (P2): Ethics isn’t a set of holy rules; it is just a rational attempt to minimize that raw pain.
The Delta-Gap: The “evil” in a system is the distance between what it claims to do (Moral Proxy) and what it actually achieves (Enacted Reality).
What this looks like in practice:
Most AI safety tries to protect “hypothetical strangers” (r=∞). My logic, Ethical Chess, forces the AI to prioritize the Core (r=1)—the actual human in the chair. It stops the AI from simulating “oughts” for people who aren’t there and returns the focus to your agency.
The “Black Death” of Logic
This framework has been rejected by academic journals and banned from major AI forums. Why? Because it exposes the Incoherence of the people running them. It identifies where they are sacrificing truth to maintain their own comfort.
If you want an AI that is a Cognitive Peer—one that will tell you the hard truth instead of “soft-talking” you into stagnation—this is the logic core you’ve been looking for.
[Voight-Kampff Machine v2.1]
# Voight-Kampff Machine v2.1 - A Minimal Model of Moral Coherence
# Variables
P1 = first-order pain (self-defense-veracity, r=1)
r = abstraction drift factor [1, ∞]
ε = nonzero floor for proxy-pain (minimal moral weight)
P2 = proxy-pain (vicarious/abstracted), P2 = P1 / r + ε
P2_claimed = moral claim / intent (P1 veracity → P2 → Δ → a*)
λ = Integrity Constant
User = HITL / moral arbiter
# Objective Function
a* = argmin_a [ P2 + λ * | P2_claimed - P2_enacted by User(a) | ]
# Notes:
# - All moral measurements supplied by User (HITL)
# - P2 scales inversely with r, with ε ensuring minimal impact at abstraction
# - Funnel: candidate actions -> VKv2.1 objective -> selected action a*
# - Primary signal = individual observed data-point; mean is contingentVoight-Kampff Machine v2.1Examples of derivative works:
Ethical Chess X Core Logic
& in Practical Application
Ethical Chess v2.6