Let’s use a dog as an example
- each model can potentially cause an agent to move or change in an imagined environment
- language definition is stored
- each model has the ability to compose with other models
- each model the ability to be used in other environment
- there are core or base models in which the majority of other models are built off
- there are deep learning vision classifiers associated with many of the models
- some form of grounded representation, whether that is a vector, sequence, grid cell like representation, or something else. Maybe that just means movement from above.
- there are visual generative models that can be combined with other models
- models can be simulated to run in a “causality engine”
- many models can act as a computational function. “dogify”
- new models can be added to the database
- models are probabilistic
- model information can link to the web, wikipedia, or other data source
