And here are some caveats and thoughts overlapping: #cognitivecomputing #semanticweb
The attention on challenges in scaling system that can benefit from large amount of background/domain knowledge is important. And I also agree that neat, well crafted, traditionally built software systems would not be efficient. But, I can't get myself to agree with some statements such as these: "The non-messy way to develop would be to create one big knowledge model, as with the semantic web, and have a neat way to query it." That is an inaccurate way to characterise semantic web. Semantic Web approach is not all about building one consistent, comprehensive ontology to build one solution. DBPedia or Freebase cover a lot of domains, and linked open data consists of a large number of independently created, at times indifferent quality, at times inconsistent knowledge clusters. Black box approaches (aka neural networks) is not the only way to scale. White box approaches are possible and have benefits (they can explain the reasoning and outcomes, and often that is critical).
This statement is completely consistent with the semantic web approach (it seems to be implied that it is not): "All has to be combined, from natural language processing, machine learning, knowledge representation. " Semantic (web) computing (and its use of background/domain knowledge) routinely enhances NLP and ML. The "instead of (semantic web)" implied in this paragraph is misplaced. This is just as misplaced as an assertion in 2005 that ontologies and semantic technologies can't help in search- well we know in 2013, Google figured out using "knowledge graph" (a knowledge first hand built
An important hypothesis this article makes is: "software development is outpaced by the intelligence of the systems." If the system is becoming more intelligent, while we are benefiting from more powerful algorithms (e.g., recent progress in scaling deep learning and its applications), equally or even more important is out ability to use a system that has a variety of methods/algorithms/reasoning and then our ability to pick the right ones to solve a problem. It seems that is what Watson is about-- ability to synthesize a broader variety of computational methods to solve a problem-- just as what our brain seems to be able to with its top brain and bottom brain working together.