Mind · Inference
ExplosiveNoise
What a single-model system filters out as noise can be signal belonging to another, overlapping model
Definition
Mechanism
A system that values information only for ShapeControl labels as noise anything that would destabilise its current shape: the neighbouring conversation in a pub, an advert for cheap oranges on the way to buy milk, a gene arriving from another species. Much of this is not random; it is information for a different model of what the agent could be doing. A MonoModularity system can only ignore it. A MetaModularity system can switch to the model for which that information is signal, joining the conversation, diverting to the orange shop, incorporating the gene through horizontal gene transfer, gaining InformationGrowth at the cost of changing shape. Switching models has a cost, so not every such signal is worth exploiting.
Invariants
- Whether information is noise is relative to the model receiving it
- Exploiting such information requires an alternative model to switch to
- Switching models costs something and must be weighed against the value of the information
Risks
Failure modes
Filtering out every input that does not serve the current shape
Switching models for every distraction, so no shape is held long enough to gain from it
Assuming that information which looks random to one model is random
Relationships
Builds on
References
Adding new information that has value in its own right, even when exploiting it deforms or abandons the current shape
A body plan that holds many overlapping but non-identical models of the same problem, ready to instantiate whichever fits
A body plan in which one model processes every situation, treating each situation as just more data for the same process
Keeping an organisation, model or trajectory close to a desired form or target, with information serving that form
Nothing in this library builds on this pattern yet.
Reference