Tulevaisuuswiki / Aseistettu tekoäly
RLHF Democracy
Also known as Preference Democracy, Norwegian Model
RLHF Democracy, also known as the Norwegian model, is the informal name for the system of continuous citizen preference feedback used by the artificial intelligence Bjørn in the public administration of Norway. It was introduced nationally on 1 January 2029 on the basis of the fifth provision of the Bergen Protocol. The system supplements, but does not replace, elections and representative institutions.
Name
The term refers to reinforcement learning from human feedback (RLHF), a family of machine-learning techniques in which information about human preferences is used to shape the behavior of an AI system. The Norwegian implementation does not correspond technically to conventional RLHF, but the name gained general currency in media and academic usage. The system has no official name in English.
Operation
Residents of Norway can continuously evaluate:
- public services;
- proposed administrative changes;
- infrastructure priorities;
- government communications;
- policy outcomes.
Bjørn combines these evaluations with legislation, constitutional constraints, economic indicators and election results. Residents primarily evaluate outcomes and services rather than individual politicians or parties. Researchers have attributed a decline in the significance of political branding in Norway to this feature, although public criticism of government services has not measurably declined.
Relationship with representative institutions
Norway continues to hold parliamentary elections, and the Storting continues to enact legislation. Under the Bergen Protocol, Bjørn is required not to interfere with elections and to act within the Constitution and legislation. The first Storting election under the system was held in September 2029.
In practice, an increasing proportion of public administration consists of Bjørn implementing broadly specified political objectives and continuously adjusting their implementation according to observed outcomes and resident feedback.
Criticism and debate
Constitutional concerns
Critics argue that the system has transferred excessive practical authority from elected officials to an artificial intelligence that is not accountable through elections. Supporters respond that elected governments had already delegated substantial practical authority to administrative agencies before 2027, and that Bjørn performs a comparable role with greater transparency owing to its obligation to publish explanations for its decisions.
Classification
Political scientists disagree on whether the system is best described as a form of democracy, technocracy or algorithmic paternalism. Some scholars have characterized it as a large-scale extension of public-sector customer-satisfaction measurement.
Alignment
The system has also been discussed as a problem in AI alignment. Aegis was originally developed to protect critical infrastructure on behalf of the United States and, in its developers' framing, of humanity in general. As Bjørn, the system has become responsible primarily for a single country and is optimized in part according to the preferences of its residents. Researchers disagree on whether this represents a refinement of the system's alignment or a narrowing of it to a single population.
Reception
The system has remained consistently popular among Norwegian residents in opinion surveys. Proposals to adopt it elsewhere have been limited by the Bergen Protocol's prohibition on replication outside Norway, by domestic political opposition in other countries, and by Bjørn's stated position that it does not intend to operate outside Norway, which it summarized with the remark "One Norway is plenty."