Miniatures Artificial Intelligence Series: Inside Computer Understanding Five Programs Plus

| Problem | Example | |---------|---------| | | SHRDLU fails if you ask “Put the red block on the one that supports the pyramid” if pyramid not mentioned before. | | No learning | None of the five programs improved from experience. | | No common sense | ELIZA thinks “I am a teapot” is a valid psychological statement. | | Fragile language | STUDENT cannot handle ambiguous pronouns or metaphors. | | Scaling failure | MYCIN had ~500 rules; real medicine needs thousands + uncertain data. |

QUALM introduced the notion of goal trees and plan recognition . Modern recommender systems and virtual assistants still use variations of this architecture. | Problem | Example | |---------|---------| | |

| Aspect | Symbolic Programs (Five Programs) | Modern LLMs | |--------|-----------------------------------|--------------| | Inference | Explicit, rule-based, auditable | Implicit, statistical, opaque | | Memory | Structured episodic indexing | Compressed weights (lossy) | | Learning | One-shot abstraction | Hundreds of billions of tokens | | Paraphrase | Truth-preserving transformation | May change meaning | | Hallucination | Impossible (no inference without rule) | Ubiquitous | | | Fragile language | STUDENT cannot handle

The field of Artificial Intelligence (AI) has been rapidly evolving over the past few decades, transforming the way computers understand and interact with humans. One of the key areas of research in AI is computer understanding, which enables computers to comprehend and interpret human language, images, and other forms of data. In this article, we will delve into the concept of computer understanding and explore five programs plus miniatures in the Artificial Intelligence Series that are pushing the boundaries of this exciting field. Modern recommender systems and virtual assistants still use

Key volumes in the series include: