Hacker News Reader

Comment by walterbell | Hacker News Reader

Would the Semantic Web (e.g. vertical knowledge graphs) qualify as AI?

We need business models which support partially-open graphs, e.g. object IDs and some metadata are open, some metadata is closed via API. Open metadata can be cached offline and standardizes models within a vertical market, driving demand for paid metadata. API revenue grows as the open graph grows.

Semi-open knowledge graphs reduce the cost of adversarial algorithms. Better to have many competing skynets than one big skynet. The open part of the graph lowers the cost of consumption. Multiple, closed annotations on the graph compete to support AI use cases, and can optionally become open as value moves to higher-order representations.

View on HN

Would the Semantic Web (e.g. vertical knowledge graphs) qualify as AI?

Anything resembling OpenCog or Freebase isn't AI—it's just a database with clever retrieval through traversal hierarchies.

The world does need better ways to analyze data sets, but conflating information retrieval with self-directed AI is misapplying the term "intelligence."

In practice, "AI" is used as an umbrella term that definitely includes that kind of stuff. I can't speak for YC, but IMO Semantic Web work should probably fall under the AI umbrella in this case.

(Edit: in the interest of full-disclosure, my startup is doing a lot of work with Semantic Web technologies, particularly in terms of applying them to Enterprise knowledge management)

I think there's two viewpoints. There's "outside in" when the media calls things like Siri "AI" then there's the scifi AI perspective where we want talking starships.

Siri++ doesn't get us closer to sarcastic GSVs. If saying a structured connected wisdom of the crowds folksonomy is actually AI helps people get funding/attention/growth, that's what people will do.

There's also "middle out," but that's a talk for another day.

FWIW, when you talk about the Semantic Web, that also entails reasoning and reasoners, and there's no requirement that those reasoners be simple deductive logic based reasoners. Work being done to make a more useful reasoner could certainly be "real" AI work. I'd also personally say that "knowledge representation" is part of AI, and KR problems are also being attacked by the SW community. Those, among other reasons, are why I say that one should classify SemWeb "stuff" as AI.

But you're right to say that there are different definitions, and it does depend on who you talk to. The media definitely take a more expansive view of AI than probably most of us, but what can ya do? shrug

A third and important viewpoint is academia. In that realm, you'll often hear people give the definition: "AI is search," which would encompass reasoning algorithms, knowledge bases and ontologies like the Semantic Web. Deep Blue and Watson come to mind as the best examples of AI as reasoning through search.

It's not information retrieval if there's no information to be retrieved in a vertical market. With the right incentive structures and business model, new vertical data sets can be generated, which enables new applications of existing algorithms.

Hopefully the term AI still includes cybernetics (e.g. Biological Computer Lab), and is not limited to currently fashionable neural nets. In cybernetics, the vertical observer is as important as the observed "data".

I have a natural language question answering system build on top of information repositories.

Understanding the intent of questions is an AI topic for sure.

Reply on news.ycombinator.com