During the early days of AI, the field was heavily dominated by a symbolic approach to processing. In other words, it relied on processing information with symbols and structures, as well as rules to manipulate them. It wasn't until the 1980s when the field of AI took a different approach—connectionism. The most promising modeling technique of connectionism is neural networks; however, they are often met with two heavy criticisms:
- Neural networks accept inputs of a fixed size only, which won't be of much help in real life where inputs are of variable length.
- Neural networks are unable to bind values to specific locations within data structures that are heavily employed by the two information systems we know of—the human brain and computers. In simpler terms, in neural networks, we can't set specific weights into...
