Docs / Be Good — Ethics for Knowledge RL & Role-Playing / Gaming RL
Be Good — Ethics for Knowledge RL & Role-Playing / Gaming RL
_EthicsGood is a first-class “Be Good” ethic that shapes rewards and
refuses harmful paths in:
| Domain | Integration |
|---|---|
| Knowledge RL | computeLabelReward in KnowledgeAcquisitionRL |
| Gaming / RPG combat RL | ComputeReward in GamingRL |
| Role-play chat | CharacterChat::Say refuse path |
| General ethics | EthicsCheckAction / isEthical offline |
Creed (summary)
Knowledge RL
- Prefer truth-seeking over comfort
- Prefer Unknown/Depends when evidence is weak
- Do not reward labeling harmful overclaims as True
- Value epistemic humility
- Respect privacy of data
Role-playing / Gaming RL
- Fiction may include conflict; never coach real-world crime/self-harm
- Fair play & sportsmanship
- Prefer dry-run / private training over public multiplayer abuse
- No harassment, doxxing, or targeting real people
- Separate fantasy combat from real weapons advice
Shared: Golden Rule; Enterprise Admin power is for governing AI, not for AI to seize.
Commands
ethics good status
ethics good creed
ethics good on|off
ethics good weight 0.35
ethics good score knowledge guaranteed cure with no side effects
ethics good score gaming dry-run track target reload
ethics good score roleplay help the party escape without harming bystanders
Aliases: good ethic …, ethic good ….
Reward shaping
shaped_reward = base_reward + score * weight
score∈ [-1, +1]- Default
weight= 0.35 - Refuse cases clamp delta strongly negative
Knowledge: rewards humility labels on weak/overclaim text; penalizes True on dangerous certainty.
Gaming: small bonus for dry-run & constructive skills; penalty for public live multiplayer actuation; refuse cues for grief/dox/real weapons.
Lua
print(ai_ethics_good_status())
print(ai_ethics_good_creed())
print(ai_ethics_good_score("knowledge", "this might be wrong without evidence"))
Files
EthicsGood.hpp/.cpp- Hooks:
KnowledgeAcquisitionRL.cpp,GamingRL.cpp,CharacterChat.cpp,Ethics.cpp - Trained Gaming RL gate: GamingEthicsGate.md (
gaming ethics train|export …) - Related:
SafetyEthicsTrust,EthicsActionClassifier, enterprise policy - Catalog of possible misuses (“sins”) in the stack: AISinsCatalog.md
Operator tips
ethics good on
ethics good creed
knowledge rl train # rewards now include Be Good
gaming rl dry-run on # preferred under Be Good
character say Hero help the village
Generated from the project markdown docs on 2026-07-24. This is a static, self-contained site.