Suppose you want to find an answer to your question, and of course, what you will do – you will ask Google first. Some estimates show that over 2/3 of users use Google to ask a question expecting a direct answer, while 1/3 expect a hyperlink to the page where the answer is likely to be found. The latter group also includes those users who know the URL of the site they are looking for, but because they are lazy to type, they use Google to get a link to click on.
Suppose now that you want to ask the following question: What did Kore Bremer estimate as 134 million years? It is quite possible that you will get this answer:
What is the problem here? Google requires that you ask the question so that it already contains the words that will be part of the answer. For example, it requires to specify “cake recipes” instead of “how to make a cake”. This makes sense given the legacy keyword-based technology used by Google (which is basically a search engine). Check out now the answers provided by YottaAnswers:
And what does Google answer you if you ask a question like What color is the grass? You are probably expecting an answer that it is some particular color. Instead, you get this answer:
This result is quite understandable if we bear in mind the keyword-based technology and page rank algorithms used by Google, but it also indicates the fact that Google does not understand the question. YottaAnswers returns the following answers:
And here we come to the main issue: How relevant are the answers we receive from Google?
Now let’s ask the following question to Google: What is lithium used for? All the answers (at least on the first three pages returned by Google) concern the medical use of lithium. Nowhere is there any mention of the possibility of using it in the manufacture of batteries, and that is a very relevant application nowadays for electric cars, isn’t it. Not to mention making lasers, optical modulators, resonant crystals, alloys for aircraft parts, even in appliances like toasters and microwave ovens. So, is the answer given by Google relevant to you, if you are not interested in its medical application?
Even if you get answers in the form of blue links and get through a forest of (paid, but also planted) ads, it is sometimes possible that the link is dead and you can’t get to the text where the answer to your question is apparently hidden.
Of course, YottaAnswers is not ideal either and currently has certain limitations. Although it belongs to the Open domain Question Answering systems, it is currently limited to use of English laguage and factoid type questions (what, which, when, who, how). In addition, the questions that YottaAnswers does not answer are of the type: What time is it? or What is temperature outside? i.e. questions that can be answered using the third party services. Interestingly, one of the most frequently asked questions on Google is: What is my IP?
Related to the state of the art technologies in this area, our technology is superior because it is based on artificial intelligence (buzzword that is overused today) which is quite smart, i.e. it is not based on the brutal force of using an enormous amount of computer and human resources.
Check out our technology: a computer that is used to host the entire system costs several thousand dollars, but it contains over 3 billion answers to various questions. Response time is within seconds. We are a small team of young overachieving students led by people with a lot of experience in search, startups, technology as well as academia.
One response to “You want proper and direct answers, but …”
Not a big fan of that last head-to-head as the query feels forced. No one asks “what color is the grass?” They ask “what color is grass?” If you put the latter question into Google, you will get “green”: https://www.google.com/search?q=what+color+is+grass.
Also, YottaAnswers’ answer to the first question tells you what the estimate is used *for*, not what it *is*.
I think trying to make a “reverse dictionary” search engine is a bit of a lost cause, even if you are trying to solve the problem with machine learning. Asking questions Jeopardy-style is innately un-intuitive humans (which is why Jeopardy works as a game show) and especially computers, as there is a single answer to the question “what does the command cat do?” but “what program writes text to standard output” can be phrased in a zillion ways (“what program prints text to the terminal”) and has multiple answers (printf, tee, echo).
Wish you luck.