-
Website
http://venturebeat.com/ -
Original page
http://venturebeat.com/2009/04/13/semantic-answer-engine-true-knowledge-opens-to-developers/ -
Subscribe
All Comments -
Community
-
-
Top Commenters
-
Popular Threads
I hate to be a wet blanket, but this is a "me too" company flogging a long-dead horse. Unless there are some mad theoretical mathematics skills not in immediate evidence, this is a dead end.
The irony is that this is a repeat of history where inveterate salesmen are selling dead technology to people who do not remember the failure to deliver at great cost to reputation the last time it was sold decades ago. Very few people are actually doing interesting work in this space, and most who are well-known are rehashing old tech. Ask Barney Pell how that worked out.
But more specifically, semantic search technologies that work the way most people imagine they should are demonstrably impossible without someone solving a theoretical computer science problem or two. No one is claiming these problems have been solved, and so it is reasonable to presume that it is yet another case of someone who does not really understand the theoretical problem overselling their imagined solution. When I evaluate the technology of various startups with fancy software technology, the question I always ask is why it *must* work. In computer science, that is a question you should be able to answer if you understand the technology. It would be great if a company indicated they had solved these problems, but doing so would make lame semantic apps seem like chump change relative to the other applications of the exact same computer science.
And that is why I am not sanguine. I have seen this pattern so many times in this space that I have lost count. There are many companies claiming to have solutions that our best theoretical computer science says should not be possible unless they are sitting on an epic theoretical breakthrough and forgot to mention it. Considering how many such companies have come and gone (several dozen at least) it is reasonable to be dubious in the absence of other evidence. Maybe Wolfram has something fabulous to offer, but I am not holding my breath.
I'm willing to believe that solving the problem of answering any natural language question on any topic is too ambitious, but you need to think pragmatically here. Lots of things unsuited to keyword search can be achieved by semantic apps without hitting major theoretical obstacles.
could you piont us to these theoretical problems?
Neither company claims to have solved a new computer science problem. They are simply applying existing techniques in better ways. In the case of True Knowledge they are crowdsourcing a database of facts that they can then reason against. Reasoning does work, if the facts are sound. The key is to curate the facts well. Their community-based fact curation system solves that. That is what they have innovated. It makes it easy to scale a vast knowledgebase of facts via a crowdsourced approach, thereby breaking the barrier of what one company could possibly data-enter on their own. In theory they should be able to gather more facts than Wolfram Alpha, faster.
On the other hand Wolfram Alpha has put a lot more work into the depth of problem-solving their system can perform. They can do calculations that are far beyond the capabilities of True Knowledge. They are not openly crowdsourcing their knowledgebase, but instead have a team of curators working diligently to suck the data in from other datasets automatically -- which is effectively curated crowdsourcing via third-party datasets. The tremendous computational abilities of Mathematica plus web-scale data to compute against is significant and new. And interesting to some people -- namely anyone who wants quick answers to science or math questions for example.
Both of these companies have potential business opportunities. There is nothing provably impossible about what they are CLAIMING to solve. They are not claiming to have revolutionized computer science. I view both companies as really focused on more efficient solutions to existing problems, rather than radically new solutions.
In any case, what is "interesting" to a computer scientist (which Andrew, above, sounds like he may be) is probably totally irrelevant to consumers on the Web, and vice-versa.
Ha! Good point.
Having said that I'm also quite used to being told that True Knowledge is doing something theoretically impossible. In fact this is the third time in the last two weeks that someone has told me that.
The bottom line though is that True Knowledge provably does work. We still have lots more to do but are answering real world questions, correctly and in volume and have solid data to prove this:
http://blog.trueknowledge.com/2009/03/true-know...
We've also had thousands of people try the technology and now also have an open API that anyone can use in their own applications.
Said that being cooler than Cuil is probably the easiest thing in the world :-) and that a real judgment about this new service will be possible only when it will available for testing (the same is true for Wolfram Alpha), I think that there is some truth in all the comments.
For the everyday people, a semantic Q&A system is something magical where you can ask any type of question and, in a fraction of time, receive a precise answer: if we reason in these terms (and a lot of hype and wrong expectations from many companies in the past made people think so), Andrew is right as there are fundamental and very complex problems that are far to be solved (and I don't think that Wolfram solved it while William was very honest in his description) and probably will never be solved.
But if we move our point of view to the delivery of a useful Q&A service that can answer with a good quality a big set of common and frequent questions for Web users, this is doable: it requires a lot of work (and money), a very good technology (and I think it can be language independent only partially) and a strong, strong commitment (as Edison said, genius is one percent inspiration and ninety-nine percent perspiration).
It will not be a Google killer (or even a real competitor) but it could be a very good complement for specific fields and needs: as it is very expensive to develop such a system that can be really useful in everyday life (without big investments, you risk to be marginal as you can answer only a fraction of the most common questions), it seems that Wolfram Alpha has better chances (from the number of people involved, they have probably invested something like $10 to $15 million up to now) but I don't see any need of something like Mathematica that seems more a limit than an advantage as it increase by an order of magnitude or two the number of servers needed to deliver the service (from what I read, I would not be surprised if Wolfram decided to use it only to demonstrate how powerful it is even in contexts where it is not needed).
I will be happy to add something more specific in the moment I can test the two systems with real questions (and answers) and check if they deliver what they promise.