Limitations of Knowledge Engines

Wolfram|Alpha was advertised as a revolutionary knowledge engine designed to eclipse Google. The result is quite mixed. While the power of symbolic computation certainly helped people solve mathematical problems, Wolfram|Alpha still provides highly eccentric answers to queries like iOS. Instead of giving a description of Apple's flagship OS, the query results in a obscure Greek island. The knowledge engine even ignores case sensitivity. The far-from-satisfactory performance begs lots of questions.

First is about knowledge model. Wolfram|Alpha works as a interpreter between human queries and knowledge databases. Given the sheer volume of human knowledge, it's natural that pure knowledge databases omit huge areas of public knowledge, and Wolfram|Alpha can only cover a very restricted domain. Compared with search engine's wide indexing of web pages, which of course contains falsehoods, Wolfram|Alpha is cleaner, but also very much less encyclopediac. How about depth? The tragedy is that Wolfram|Alpha doesn't read books and papers, and provides bangia as a answer to Yangian in mathematical physics.

Both Google and Wolfram|Alpha don't cover highly private knowledge like sophisticated semiconductor manufacturing process. But since public knowledge is often local and scattered through the web, Wolfram|Alpha failed to capture the majority of it. Pieces of knowledge are often separated in Stack Exchange, Wikipedia, and blogs, etc. Wolfram|Alpha only provides links to Wikipedia entries. Search engine like Google isn't bad after all. Not to say databases utilized by Wolfram|Alpha are often outdated.

Serious sciences are protected by law of large number. Beyond law of large number, black swans happen and can not be reliably predicted by computation based on existing data. The idea of computational knowledge engine sounds cool, but in reality it's not very general. Google collects up-to-date information through mechanisms like surveillance, and easily beats Wolfram|Alpha for timeliness.

Furthermore, vast amount of human inquiries are beyond the domain of knowledge. David Hume showed that normative morals can not be derived from positive doctrines. Aesthetics also can not be categorized as science. Although not without problems like political extremism, search engines generally accommodate wide varieties of views much better than monolithic pure knowledge databases. Wolfram|Alpha is of limited use to most people.

Will there be successful knowledge engines?


Popular posts from this blog

Maps and Yellow Pages