Single agents can generalize better to unseen tasks because they are not constrained by predefined heuristic-based workflows.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Smith & Boyd (1991) criteria are irrelevant across the vast majority of the space of possible agents.claim0.787
  Dismissal of earlier criteria as too narrow.
• A naive agent equipped with reduced priors from an experienced agent performs perfectly with maximum confidence from the first trial.finding0.778
  Demonstration that model-level priors (not parameter-level knowledge) suffice for immediate transfer
• Agents perceive by minimizing variational free energy to ensure model consistency with past observations and act by minimizing expected free energy to make future sensations consistent with preferences.claim0.778
  Formalization of perception-action cycle integrating inference and decision-making.
• Active inference agents can learn their own reward function (prior preferences) by interacting with the environment, bypassing the need for an explicit reward signal.claim0.768
  Abstract and §3, preference learning section.
• The simple fact that intelligent agents are collectives and do not continue unchanged fulfills a necessary requirement for an open-ended evolution of intelligence.claim0.767
  Connects collective intelligence to evolutionary potential.
• Open-ended evolution of intelligence is possible because agents are collectives without fixed essencehypothesis0.765
  Follows from observation that intelligent systems lack context-transcendent core; their maxima are not contingent on permanent character.
• What an agent does not do—silence, deferral, waiting—is a positive structural feature, not absence.claim0.762
• There are fewer representations competent for N tasks than M<N tasks, so training more general models should yield fewer possible solutionshypothesis0.762
  Selective pressure toward convergence via task generality