Re: Misconceptions about what knowledge representation truly is

Thanks for bringing this up.

According to Randall Davis, knowledge representation provides a language for describing the world and a computational model for consequences, albeit a fragmentary theory of reasoning. Another definition is that knowledge representation provides the ability to understand, reason and make informed decisions. In a similar vein: knowledge representation refers to encoding information about the world into formats that can be utilised to solve complex tasks. Another perspective is that knowledge representation refers to the way information is mentally represented using symbols or mental images. Cognitive Psychologists emphasise the importance of mental representations in problem solving and communication. However, these representations are just a convenient fiction when talking about the operation of the brain rather than dealing with the complex waves of neural activation across the brain.  

What about explainability versus adequacy?

Formal logic provides explanations in terms of mathematical proof from the stated axioms. However, this comes at the cost of adopting a distorted and oversimplified view of the world. As such, formal logic is rarely useful in real world contexts despite the aspirations of AI researchers over many decades.

Neural networks are highly effective for dealing with the complexity of everyday knowledge, but lack formal explainability due to the opaque statistical models derived from machine learning. It is more productive to ask for human understandable explanations for a line of reasoning. This replaces mathematical proof by rational argument and rhetoric, harkening back to Aristotle, and more recently, the Age of Enlightenment that brought us industrialisation.

So what are the benefits from explicit knowledge representation in the era of strong AI?

I think this relates to the mundane need to avoid misunderstanding both within and between businesses, along with the need for persistent records in support of taxation and legal actions. There is an opportunity to apply knowledge representation as a technical argot for de jure standards.


Dave Raggett <dsr@w3.org>