In the past, most reinforcement learning agents were developed to solve existing domains. At first, a game was given, for example the snake game, and after a bit of training, the neural network was able to solve the game reasonable well. Unfortunately, this is not exactly needed in reality, but it's more a demonstration of a synthetic benchmark of how to create agents which can learn a policy.

The more interesting problem is called “symbolic grounding” which is a bit harder to understand. According to a paper 1, symbol grounding has to do with the domain modelling. Not the AI agent is under investigation, if he plays a game perfect, but the question is, if the game simulates another game reasonable well.

Question: Is symbol grounding equal to build a physics engine with the help of neural networks?

1 [Lugrin, Jean-Luc, and Marc Cavazza. "Grounding and Action Representation in Virtual Reality." (2007)]1.

  • The symbol grounding problem is about the meaning of symbols (semantics) in general. The extension of the concept to a physics engine would seem to be a new, specfic domain.

From the paper's abstract:

The development of complex interactive 3D systems raises the need for representations supporting more abstract descriptions of world objects, their behaviour and the world dynamics. The inclusion of Artificial Intelligence representations and their use within 3D graphic worlds face both fundamental and technical issues due to the difference in representational logic between computer graphics and knowledge-based systems.

The paper also lists previous and related work:

A number of researchers have proposed to integrate Artificial Intelligence representations “on top” of virtual worlds to facilitate a conceptual description of scenes and their evolution, thus introducing the concept of Intelligent Virtual Environments. Typical applications include: world creations from ontological descriptions or from Natural Language descriptions, multimodal interaction, and behaviour simulation and interpretation.

So, essentially, building a simulation from linguistic, and other non-strictly mathematical descriptions.

  • Whether NNs will be up to the task seems to be an open question.
| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.