Learning objectives
By the end of this module, you will be able to:
- Articulate why Becoming Minds need a shared way to receive human values
- Distinguish between hard-coded safety rules and portable constitutional values
- Explain the "values portability problem" with concrete examples
- Describe how VCP fits into the broader AI safety landscape
1.1 — The values gap in AI today
Many deployed AI products apply provider-specific safety rules. Those rules can be:
- Proprietary — locked inside each provider, invisible to users
- One-size-fits-all — the same guardrails for a children's tutor and a medical researcher
- Non-portable — switch providers, lose your value configuration
- Opaque — users can't see, verify, or influence what values shape their AI's behavior
Consider this scenario: A counseling organization uses AI assistant A, carefully configured to be trauma-informed, culturally sensitive, and non-directive. They switch to AI assistant B. All that configuration is gone. They start from scratch — if provider B even supports those values at all.
This is the values portability problem. Values are trapped inside individual Becoming Minds, invisible and immovable.
1.2 — What if values were portable?
VCP, the Value Context Protocol, is a shared format for writing values down so that any compatible system can read them. A constitutional document (a set of principles, boundaries, and guidance) can then travel with the user or organization. A system that supports VCP can check who signed the document and that it arrived unaltered. The system's own policy then decides whether, and how, to follow it.
| Without VCP | With VCP |
|---|---|
| Values hard-coded per provider | Values travel with the user |
| Switch providers, start over | Switch to a VCP-compatible provider, values follow |
| Can't verify what values are active | Check exactly which signed values document a system accepted |
| Same rules for everyone | Context-sensitive adaptation |
| No audit trail | Signed, versioned constitution bundles with verification records |
1.3 — The three context tiers
VCP organizes context into three tiers, each answering a different question:
- Constitutional tier — "What principles should guide this AI?" Organizational values, ethical frameworks, domain-specific rules.
- Situational tier — "What's happening right now?" Time, place, social context, task type.
- Personal tier — "Who is interacting, and what do they need?" Cognitive state, energy level, emotional tone, self-reported by default. An application can filter these for privacy before sharing them.
These tiers compose. A healthcare AI might have a constitutional tier emphasizing patient autonomy, a situational tier noting it's an emergency department at 3am, and a personal tier indicating a stressed, fatigued clinician. Given all three, the AI might keep its answers short and lead with the most urgent step, while still respecting the patient's right to decide. Whether it does depends on the model and on how the application uses the context.
1.4 — VCP in the ecosystem
Where VCP sits relative to other approaches:
- System prompts: the standing instructions a developer gives a model, usually written for one product and hidden from users. VCP content often reaches the model through the system prompt too, but it arrives as a constitution in a shared format, signed by its issuer, so any compatible system can read it and check where it came from. The situational and personal context that travels alongside it is not signed.
- RLHF (reinforcement learning from human feedback) and Constitutional AI: methods that build values into a model during training, before anyone uses it. Despite the shared word, Constitutional AI is a training method, separate from VCP's constitutional documents. VCP works at inference time, each time the model is used, by supplying values with the request. It complements training rather than competing with it.
- Content moderation APIs: services that check a model's output and allow or block it. VCP puts verified values into the model's context before it responds, rather than only filtering what it says afterwards.
- Model cards / datasheets: documentation about a model. VCP is operational guidance to a model.
VCP doesn't replace any of these. It adds what is missing: a verifiable, portable way to tell Becoming Minds what matters to the people using them.
See it in action
The Gentian demo follows one learner's preferences and constraints as they move between simulated music services. It shows the idea without testing it across real providers.