Maybe try Bayesian approach—where you have several ontologies with different probability weights? In that case, you could work with each ontology as if it is real, but take its predictions with high discount, and also use the predictions to update relative weights of the ontologies.
Different ontologies are useful in different contexts. If you model airplanes with Newton’s laws you don’t get a benefit by applying some probability to them and another probability to the formula of general relativity.
I’m not sure what you mean with “which ontologies are true”. If you take the example of Valentine’s ki, that ontology allowed Valentine for a long time to do things that he couldn’t do without the ontology. Putting probabilities on ki being the true ontology misses the point.
If he assigned 50 per cent probability to the ki ontology, he still could do everything which is required by ki-ontology, but a) divide expected gains on 2 b) updated probability of the ki ontology depending if it works or not, and also based on explanation power of other ontologies for the same set of the evidence.
Maybe try Bayesian approach—where you have several ontologies with different probability weights? In that case, you could work with each ontology as if it is real, but take its predictions with high discount, and also use the predictions to update relative weights of the ontologies.
Different ontologies are useful in different contexts. If you model airplanes with Newton’s laws you don’t get a benefit by applying some probability to them and another probability to the formula of general relativity.
Surely if you know which ontologies are true and in which context you currently are.
For example, one could give 50 per cent to the probability that he lives in the real world and 50 per cent that he lives in a computer simulation.
I’m not sure what you mean with “which ontologies are true”. If you take the example of Valentine’s ki, that ontology allowed Valentine for a long time to do things that he couldn’t do without the ontology. Putting probabilities on ki being the true ontology misses the point.
If he assigned 50 per cent probability to the ki ontology, he still could do everything which is required by ki-ontology, but a) divide expected gains on 2 b) updated probability of the ki ontology depending if it works or not, and also based on explanation power of other ontologies for the same set of the evidence.