Skip to content

Value Networks

Value Networks extend the Governance Engine framework with mechanisms for value-based decision making and alignment.

Value Objects are the voting subjects in abstract voting systems. Instead of voting between a list of representatives, you can vote on a list of values. Each value is described in detail in its value object.

There can be different types of value objects depending on needs and contexts.

Exact legal description representing this value, including references to existing law, contracts, and even specific court cases as examples.

LLMs and other text-sensitive AI can be trained towards these values by providing training data and testing data in the form of texts or other resources, as part of the value object definition. See AI Alignment for the broader vision.

Not everyone has expertise in every field. You should be able to distribute your vote.

Confidence-based / weighted voting:

  • Everyone votes on a set of values (defined through value objects)
  • When they vote, they assign a weight — a number of votes or percentage dedicated to an option
  • This allows people to express both their preference and their confidence

Apply negative votes to subtract other votes away. Combined with weighted voting, this enables expressing strong opposition.

Considerations:

  • Could be normalized to all positive in effect, but this might allow pushing out of small groups
  • Should probably only be applied to value votes, not representative elections

Values determined through decentralized voting can be abstractly represented as objects and used to guide AI training and behavior. These decisions can be formalized using Governance Engines.