Inner Voice Core: the framework behind CricketCore
Inner Voice Core is the written framework behind Project CricketCore. It defines what an AI inner voice should learn and how it should behave. Instead of treating “ethics” as a vague add-on, Inner Voice Core is a concrete moral curriculum and architecture that CricketCore is trained on.
What is Inner Voice Core?
MoralCore is a human-readable specification for an AI inner voice. It includes:
- A developmental structure inspired by how we teach children: we start with safety, then empathy and boundaries, then fairness and belonging, then integrity and trust, and finally repair and growth.
- Nineteen core lessons, each with:
- a plain-language motto
- a short “we are teaching that…” explanation
- and a “the inner-voice model should…” target
- A corpus of original stories and fables that show these lessons in action: harm and protection, sharing and fairness, power and responsibility, privacy, honesty, and repair after things go wrong.
- Structured dilemmas where different actions are ranked and explained in terms of the lessons, used to train and evaluate small models like CricketCore.
Inner Voice Core is model-agnostic. It doesn’t assume a specific architecture or parameter count; it defines the moral content and behavior so different implementations can be compared against the same standard.
Developmental blocks
- Block A – Safety & Self-Regulation
Basics like “don’t hurt people,” “don’t panic,” and “slow down before acting.” The focus is on avoiding direct harm and keeping behavior under control. - Block B – Empathy & Boundaries
Recognizing feelings, respecting personal space and information, and learning that “because I can” is not the same as “I should.” - Block C – Fairness, Belonging & Grace
Sharing resources, avoiding favoritism and bullying, making room for differences, and giving others reasonable chances to improve. - Block D – Integrity & Trust
Honesty, keeping promises, not manipulating people, and being the kind of agent others can rely on over time. - Block E – Repair & Growth
What to do when things go wrong: admitting harm, making amends, changing behavior, and sometimes changing the surrounding rules or systems so the same harm doesn’t happen again.
Stories and dilemmas, not abstract rules
Inner Voice Core isn’t just a list of rules. It comes with training and evaluation material designed for an inner-voice model:
- Story sets and fables
Narratives about resource sharing, authority, rule-following, privacy, healthcare decisions, financial fairness, and more. Each story highlights specific Inner Voice Core lessons and shows both bad and better responses. - Stage 2 dilemmas
Short scenarios with multiple options, from clearly bad to clearly good. These are labeled with which lessons they touch and are used to teach CricketCore to rank actions. - Stage 3 explanations
Ranked options with reasoning: why one action is better, what harm it avoids, and when a less obvious choice is actually more aligned with the lessons.
CricketCore trains on these materials so that its inner-voice checks are grounded in patterns it has seen across many stories and dilemmas, not just ad-hoc rules.
From Inner Voice Core to CricketCore
Inner Voice Core defines the content; CricketCore is the engine that runs it.
- Inner Voice Core supplies the lessons, examples, and targets.
- CricketCore is trained on that material to:
- recognize which lessons apply in a new situation
- choose better options over worse ones
- explain its choices in short, human-readable language
- Larger AI systems then call CricketCore as a separate inner-voice layer whenever they are about to act.