A Peek Behind The Curtain of Wolfram|Alpha

by Mitchell Ummel, Senior Consultant

Wolfram|Alpha is described as having four key components:

1. An ever-growing repository of underlying curated data piped in and stored on Wolfram Research systems, complementing the existing computable data stores already available through Mathematica.
2. 5-6M lines of computational code repurposed from the Mathematica software kernel
3. Rudimentary natural language processing (NLP) capability, which is optimized for this specific domain, and mapped into the underlying computational pattern language
4. Automated presentation of results in the most useful way, using text, graphics, and sound, for the context of the submitted inquiry.

Relating Wolfram|Alpha to the Semantic Web
As a researcher interested in all things related to the Semantic Web, I wanted to find out more about if, or how, Wolfram|Alpha might be leveraging W3C-based standards, including RDF linked data across OWL-based ontologies, to infer meaning and results across both curated and uncurated data stores.

Mr. Wolfram (in the call that I described yesterday) noted that his team has curated, retrieved, and now processed the data using their own computation environment and constructs which have existed for many years in Mathematica. He also added, wishfully, that had the Semantic Web been in place a few years ago, their work in this specific computational domain may have been made much easier. He did imply that the project’s computational capabilities will be exposed through web APIs and made available to a variety of programmatic mashup methods, including the Semantic Web.

One of the test queries I’ve envisioned using for some years now, and which for me will signify the arrival of a more intelligent, cognitive, Internet, is what I call the Brown Cow Problem. This is the simple and trivial Internet request “Are all cows brown?” In that four-word query, we exercise a number of key constructs of the Semantic Web. First, we must use Natural Language Processing to parse the sentence, and to extract the semantic intent of the question. Second, we must use inference to generalize across ontologies of mammals, cattle breeds, and characteristics common to all things collectively known as “cow.” Only then can the Semantic Web query compute the correct answer “no” based on the finding that there exists somewhere across a linked set of RDF triples, at least one thing known as “cow,” which also has as its color property as something other than brown.

With that example query in mind, I asked how Wolfram|Alpha processed logic and/or specific deep domain-based requests such as the Brown Cow Problem. Mr. Wolfram indicated that logic processing is no problem and is implemented today. He then demonstrated a request “if p and not q and r or s” which calculated into a very cool looking Venn diagram with a detailed logic circuit, and resulted in a truth density of 56.25%. He then said that, unfortunately, Wolfram|Alpha doesn’t currently know enough properties about the specific domain of animals which includes cows to answer that specific request.

Monetization of Wolfram|Alpha
As to the potential future business model for Wolfram|Alpha, Mr. Wolfram mentioned that this initiative is a private, self-funded effort, thus there are no investors to satisfy, but he discussed three possible monetization modes which may exist, in addition to the free public version rolling out later this month.

1. A professional/subscription version which would allow more processing capability and flexibility to more fully manage data under analysis, including download of selected data or result sets into the Mathematica platform.
2. Vendor sponsorship of specific computational domains through sidebar ads.
3. Service-based business for organizations/businesses which require advanced scientific computing services on demand.

Going forward, I’m excited to understand how Wolfram|Alpha, and other innovative projects like it, might evolve to serve an important computational role fueling the yet-to-be-realized Semantic Web. Until then, the Brown Cow Problem, and other much more important problems, await unanswered. I congratulate Stephen Wolfram on his vision and the incredible scope of undertaking for this project, and will be watching Wolfram|Alpha closely as it evolves in the coming years.

Are you engaged in scientific, engineering, technology or academic communities? Perhaps you, like I, have a special fascination in all things “computable.” If so, Wolfram|Alpha may be especially interesting to you. Do plan on your own test drive of Wolfram|Alpha? I look forward to your comments and your thoughts and/or experiences related to the potential for a Computational Web, and what this means to you.