Current Article
Anyone aware of Powerset will know that they have been busy blowing their air all over the place regarding their “coming soon” technology that is set to give the smack down to Google. As yet, only a select few know much about them and have had a chance to see their tech.
As of today a new company has opened it’s doors well ahead of Powerset. This new company is called True Knowledge and is seriously impressive. They are based in the UK, have some large funding and some seriously impressive technology.
They say, “A radical new approach to search engine technology has been developed by True Knowledge Ltd that has the potential to change the way factual information is made accessible via the Internet. It neatly side steps the problem of getting computers to read, by structuring facts in a way that machines and humans can easily understand.”
Rather than trying to explain all the new tech to you, please enjoy this feature presentation in video form that will tell you exactly how it works and why it looks so impressive.
Impressed? Want to put it through your own testing? Good news as they are currently looking for beta testers. Simply head over to http://www.trueknowledge.com and sign up to have a joy ride on their search engine today.
Do let us know what you think as we are due to interview the creators and will look to include your thoughts and comments when we do.
Popularity: 100% [?]
If you liked this you will probably love these other posts:
Jesse | Nov 7, 2007 | Reply
How is this different than Cyc or OpenCyc? I suppose the one thing that they have done different is integrate their rules engine with the Internet, but the underlying technology is a rules engine, correct?
Ian Clarke | Nov 7, 2007 | Reply
Great if it works as well as the video suggests, but one must be careful with this type of demo. This technology will succeed or fail depending on its ability to interpret real-world questions.
The questions selected by the person giving the demo are presumably selected such that they know the system will interpret them correctly. The real test is how well the system works with questions chosen by people not familiar with the system’s capabilities.
Back in the early 70s Terry Winograd developed a system called SHRDLU, which you could talk to in natural language and instruct it to do things in an artificial world. When used by its creator, the effect was impressive. He would type instructions in English, and they system would obey them flawlessly.
However, when users unfamiliar with the system’s capabilities tried using it, the results were invariably disappointing, because they weren’t limiting their interaction to within the system’s capabilities.
The danger with True Knowledge is that it suffers from the same problem. Its performance is impressive for the types of questions anticipated by its designers, but it is far less effective when a third party tries to use it.
Sunny | Nov 7, 2007 | Reply
Perhaps sign up for the beta and give it a test?
Agreed that if the tech counts on users asking questions in a specific way then it will fail, but based on the demo of the API and the calls it makes it seems that the tech is quite robust.
We have an interview with them tomorrow so if there are any specific questions do send them in / comment them below.
Sunny
Ian Clarke | Nov 7, 2007 | Reply
Sunny, I would just warn that it is unwise to form any conclusions about the robustness of their technology based on the system’s responses to questions chosen (or phrased) by them. Take a look at the Wikipedia page on Shrdlu to see how impressive a system can be when demonstrated in this way - http://en.wikipedia.org/wiki/SHRDLU
If they will give you access to the system, I would try it with some questions you’ve chosen yourself that people are likely to ask such a system. Perhaps “where can I buy a HD television in Dallas, TX?”, or “how much should I tip a taxi driver in london?”.
So long as they get to select the questions used to demonstrate the system, its not really a true test.
Ian Clarke | Nov 7, 2007 | Reply
Oh, for a specific question to ask them tomorrow, ask them to contrast their system with “Cyc” - see http://en.wikipedia.org/wiki/Cyc. Cyc is one of the descendants of Shrdlu, its been around since the 80s and has burned through a *lot* of money in the process. I’m quite convinced that they will be familiar with it.
Sunny | Nov 7, 2007 | Reply
Ian I will be sure to put this to them among the other questions that we have. Thank you for taking the time to put them on here.
Douglas | Nov 7, 2007 | Reply
I think this does look very cool, but anything in the title saying move over Google makes me scratch my head. Google is so massive, it will take an enormous amount of force to move period. They are the powerhouse!
ss | Nov 7, 2007 | Reply
I am curious about what the potential applications are. The truth about GOOG/MSFT/YHOO is that for most monetizable applications, keywords are good enough. I haven’t seen a compelling demonstration that NLP adds to understanding the query intent or the delivers more relevant search results for cases where a buyer is looking to get a good from a seller.
In most of these cases, a key phrase (ipod nano, massage service, taxi companies in liverpool, french restaurants in denver) will fairly tightly circumscribe the possible responses, and the value-add is in functionality surrounding the results, rather than in improving the results themselves by understanding the query better.
Phil | Nov 8, 2007 | Reply
My feelings are that while this approach is interesting all that is happening is that you are moving the onus from the end user to sift and process results (that Google might provide for instance), to software. Yes it’s more elegant than Google but it’s typically more effective to let the end user use their own brain cells to narrow down and locate the information they are searching for - just as Google and other search engines do at the moment.
Buster | Nov 8, 2007 | Reply
Fantastic that this has been looked into again. It will give another automated way to filter through a lot of internet information. A new way to do this is good, hopefully some of us can adapt to how it works and quicker find what we looking for.
Yahooing and Wiki reading is good and this adds to the toolset of information filtering.
It is also good that it givs the decision thread and even old fashion search results in case the reply wasn’t good enough.
The decision engine will become powerful over time and will take over human evaluation and soon we will get responses like “It dosn’t exists because I have searched the internet and didn’t come up with anything” which is a limited scope and understanding of what is on offer.
Let’s hope it takes off, same feature as before but with a new spinn. People today might be more adept to allow this change to happen. Like YouTube, same features been available over ten years but the time and approach was right and now they are big business.
Any change is good, it makes us evaluate and think what we have.
anon | Nov 10, 2007 | Reply
The reason search engines don’t answer queries is because they are search engines. This is not a “problem”, it is a “reality”. Search engines search, they don’t answer questions.
My calculator can do complicated and difficult computations, but it doesn’t interpret the answers it finds in any way. This is not a problem either.
Google does exactly what it claims to do, and for TrueKnowledge should not feign competition with them, or superiority over them.
Botter Reeves | Nov 11, 2007 | Reply
The comments all seem to have been made without actually answering the invitation extended to take a crack at their beta. I intend to do that, and I’ll let you know later.
Sunny | Nov 12, 2007 | Reply
Good call Botter! Do let us know how things match up with their demo and with your personal experience of the system.
Tom | Nov 12, 2007 | Reply
You know, to me this suggests that the people at TrueKnowledge are stuck in the search patterns of the 90s. It’s not Google’s fault that you’re trying an “Ask Jeeves” approach with their search engine. The fact that you’re purposely trying to be misunderstood takes away a lot of your credibility. Google was never designed to present a ‘yes’ or ‘no’ response, so don’t state that it’s a problem with their design.
Honestly, if you can’t figure out how to structure a query (here’s a hint: you can throw grammatically correct sentences out the window) then get off the internet.
emreerokyar | Nov 12, 2007 | Reply
to anon;
totally agree with you
Weird Biz | Nov 12, 2007 | Reply
I think this is pretty much a waste of time.
Joy | Nov 14, 2007 | Reply
I haven’t check this site out, yet..^^..Anyways thanks for the info.
Kevin | Nov 14, 2007 | Reply
Anon, I don’t totally agree with you. Google actually tries to do a form of answering questions. Type in What is the capital of the United States? and see what you get as your first hit. It should say United States - Capital: Washington, DC Then try what is the Capital of Canada? (ignore the book result and look at the response below). It says Canada - Capital: Ottawa
Each refers to a different web site yet tries to represent the answer in a similar fashion. This to me says Google is trying to do more than just search.
In addition the calculator and conversion functions of google is another ability that is not search based. i.e. type in: How many miles in 10 kilometres?
Lucas | Nov 15, 2007 | Reply
Ah, I feel nostalgic for Ask Jeeves.
I don’t think my friendly neighborhood librarian will need to look for a new job anytime soon.
Random | Nov 15, 2007 | Reply
i’d like to point out that in the examples used against google, correct answers did pop up. i think that this will in fact become popular if it works decently and gets enough publicity, but the fact remains that it’s success would be due to a decline in intelligence. grab the average IQ of myspace users and you will see what i mean.
AlexM | Aug 16, 2008 | Reply
Your blog is interesting!
Keep up the good work!