We have a “Values Elicitation Chatbot”, a version of GPT-4, that interviews people and makes these “Values Cards”. It makes sure it finds real values that are at stake in their choices, double-checking to ensure it's not incepting values in people's minds, or finding a value that just sounds good.
The values card format has teeth — it’s a representation of values that contains the nuances of how a value balances concerns and guides attention when it's used in choice, plus it’s a format people can understand and use to recognize the values they have or to connect with others’ values. The cards can be duplicated, so we can tell when two people have the same value. They're human readable, but can also be used to evaluate of guide machine behavior.
Values cards can be aggregated up into a “Moral Graph”, which captures the values of a community, including disagreements and trade offs. It also maps the “edge of wisdom” of a community — which values the community thinks are wisest to decide by, in which contexts.
Moral graphs can be used to shape the behavior of LLMs to make them wise or moral according to the community, through a fine-tuning process. Moral graphs can also be part of a voting or collective decision-making process.
Finally, the values of a community form a “map of the sacred” which can be used to make resource-allocation decisions, or to generate ideas for interactions between community members that support them to live by their values. This kind of values-based coordination can replace markets and bureaucracies for allocating labor and suggesting interactions. The map of the sacred can also be used to evaluate the welfare of a community, by whether people are well-supported in living by the values they collectively and individually have.
The Meaning Alignment Institute exists to pioneer, evaluate, improve, and drive adoption of this values-based tech stack, so as to bring to life the AI-models and social structures that will best support people in living meaningful lives. We believe that social technologies that operate with human-compatible values, reasons, and morality (but at larger scales) are the way to address many current and near-future social ills.
<aside> 🔥 Values Explicit Tech can be useful in all these places