The challenges of combining neurons and symbols

So far, we have discussed the inner dynamics of NSAI and how this new class of computing promises to take machine intelligence to the next level. There are, however, a few obstacles that we need to overcome when designing an NSAI system. In this section, we will delve into the most prominent challenges and limitations of NSAI.

Knowledge and symbolic representation

Perhaps one of the more prominent obstacles to face when designing NSAI systems is knowledge and symbol representation. As part of the symbolic component, we have two main challenges:

• Robustly decomposing our problem space into symbols
• Representing the extracted symbols in a way that a computer (more specifically, an NN) can understand

In the previous chapters, we discussed how existing NSAI algorithms mostly rely on a domain-specific language (DSL) to represent the symbols and their respective relations as first-order logic. The DSL acts as a descriptor...