The quest for knowledge never ends. That’s a good thing for the Q&A AI platform True Knowledge. The Semantic Web Blog recently had an opportunity to catch up with its founder and CEO William Tunstall-Pedoe to discuss the semantic-powered platform he’s been developing for the past couple of years.

▏ On why the world needs ‘the first Internet-scale platform for directly answering the worlds questions’”:
° The idea, William Tunstall-Pedoe says, came out of his observations about Internet search: That the search engines don’t understand what you ask, and as a consequence can’t produce the answers to what you ask – just a statistical list of documents that may/may not be what you are looking for. “Computers don’t understand natural language. Documents are magical things humans can read and understand but the best a computer can do is scan it for keywords and patterns. So the starting point is how to represent knowledge in a way that computers can understand and process it. How do you build a system where the world’s knowledge can be stored in structured form, where computers can reason with it and directly answer questions. The ultimate dream for that is an Internet search experience by mobile or web where you get back exactly what you are looking for, not a list of links, but direct perfect answers to every question.”

▏ On what sets True Knowledge apart, including from services such as WolframAlpha:
° Tunstall-Pedoe calls his service’s approach unique. “There are only really two that do end to end questions and answers. A lot of semantic search companiess are about better processing of natural language—they’re still taking documents as their knowledge source but applying more sophisticated algorithms to processing them. Until someone solves natural language understanding those systems will still look and feel like search engines — they may produce more relevant results sometimes, but still they can’t perfectly understand what is in the documents and perfectly match to a user’s query.”

There is some similarity in the user experience of True Knowledge and WolframAlpha, he notes, in that both systems are about getting users’ input and producing a direct response. “But there are all sorts of differences in terms of how it’s done and the emphasis.” WolframAlpha is built on Mathematica, is stronger in scientific areas, and puts a lot of emphasis on display, he says. A fan of WolframAlpha, Tunstall-Pedoe says the differences between the two systems include the fact that True Knowledge system is designed to integrate more into web search, to produces smaller results, rather than pages of graphs and charts, and that it hails from the starting point of “being knowledge-driven, about a universal representation of knowledge. … Also think we have more focus on parsing questions and things. Another difference is user-generated content. On our platform users can add and correct content while WolframAlpha is read only. They have the approach of their own curated knowledge powering the system.”

▏ On TrueKnowledge’s knowlege:
° Two years ago TrueKnowledge was just a prototype on Tunstall-Pedoe’s laptop with just a few hundred facts. “We’ve scaled up orders of magnitude to get to where we are.” The knowledge in the system primarily comes from sources such as Wikipedia and Freebase, with fewer than 200,00 user-generated facts among the more than 200 million in the system. “So less than one fact in 1,000 comes from users. However, those facts are disproportionately useful….Something like 15 percent of the questions we answer use at least one user-generated fact in the inference chain to produce an answer.”

Users’ input has the edge in the “skewness of knowledge” department, he says. “A big database you import has knowledge people may not be interested, in but if users are adding a fact it’s heavily skewed to what users [generally] are interested in.”

Tunstall-Pedoe makes the interesting point that, as impressive as the real number of facts in the system is, that’s a poor measure of true usefulness. “Some are thousands of times more useful than others. So it’s an interesting statistic but not particularly useful in terms of gauging how far you are or where you need to be.”

How do you gauge that? A better measure, he thinks, is the percent of time that users’ questions get answers. “It’s how useful you are, how frequently you produce useful results to users. We see benchmarks and see what users’ type and count how many times we produce a good answer. We see that growing substantially with the amount of knowledge within the platform.”

▏ On conflict resolution among facts in the system:
° “One of the advantages of our semantic technology is we know when facts conflict semantically. And the system has a process in it which switches off facts that are semantically incompatible with others,” he says. Take an example where of a conflict between birthplace facts, with one fact indicating someone was born in Chicago and another indicating Hawaii. Its semantic technology knows that you can’t be born in two places and more to the point that Chicago is not a city in Hawaii, and True Knowledge will bow to the source with the better reputation and track record. Users trying to input a fact with an eccentric view, so to speak, not backed up by evidence are unlikely to see that fact survive in the system.

“It’s non-trivial,” Tunstall-Pedoe says. “It’s quite possible to be born in both Honolulu and Hawaii but not Chicago and Hawaii. It requires a general inference system to know that.” That said, the number of facts generally in dispute is very, very small he says.

As facts accurately change on sources such as Wikipedia, those updates are automatically reflected in TrueKnowledge

▏ On the importance of its API to the business:
° The API platform—which is used by Siri in its Virtual Personal Assistant—is fundamental to the business, he says. “The platform powers everything we do. So we are absolutely interested in partnering with people to make use of our technology. Given the platform’s suitability for Q&A and search, he says any organization hoping to improve either could take advantage of it.

“We are fundamentally a structured knowledge technology, so if a corner of search involves processing vast numbers of documents, than that is outside our scope.” He says. “But in terms of knowledge areas there is no area we can’t address. It covers all knowledge areas. We have a 20,000-class ontology that covers all you can talk about, so it’s a pretty board set.”

▏ On the progress of its Local Product Search feature:
° True Knowledge is fundamentally a horizontal technology but it can be used for verticals just by fleshing out a corner of the knowledge base, he says. “So we did it around products, places and where to buy products in places, and the inference system just takes care of it,” he says. “It’s about the intersection of products, retailers and places, and absolutely our technology is suited for mobile for a number of reasons. Being location-aware and services like that are suited to mobile. If you have a modern smart phone, it has a GPS location it can trandmit. So True Knowledge query can be, ‘Tell me where to buy this in a place near the GPS location where I am currently standing.’” For example, TrueKnowledge knows that a mattress is a bed, so you can shop for it in a bed store, and there is one 50 yards from where you are.

Currently it has licensed U.K. retailer data, but it hopes to address the U.S. at some point.

SemTechBiz is Less Than 2 Weeks Away

The Semantic Tech & Business Conference (SemTechBiz) is coming to San Francisco on June 3-7! Join us for case studies, innovative panels, tutorials, and keynotes that will provide you with practical advice, hands-on guidance, and breakthrough approaches to solving business problems with semantic technology. Passes go up $200 at the door. Sign up now and save !