Q&A with Bob Rosenschein

Q&A with Bob Rosenschein

Q&A with Bob Rosenschein

Bob Rosenschein is a Pioneer of Internet entrepreneurial solutions, Founder, Chairman and CEO of

Q: There are many answers out there, but we all know that to get to the right answer, we need a very intelligent question. So my first question to you is, what kind of intelligence were you relying on when you were founding

“It went through several shifts – we call it a pivot. First, we started out with very simple factual encyclopedic information, but the pivot we did that made the most sense was to user-generated questions and answers. So we had a situation where people would put in any question – we can talk about the different kinds of questions – and other people could not only answer the question, but edited each other. We were the Wikipedia of questions and answers. The other thing we learned how to do was basically judge the quality of the questions and the quality of the answers through the user-generated process. There are many kinds of questions. There are simple factual questions – two plus two, how many pounds in a kilogram, what’s the weather in Tel Aviv tomorrow. These are easier to answer. The harder ones require more judgment or multi-stage or more complex information. And we found the most interesting ones were the ones that blended opinion, background, depth, and length. So we had to develop a system, we called it Remus, which basically measured, automatically and through user input, signals of both what we call ADQ, which was answers document quality, and AAK, which was answers author karma. And we came up with an SEO improver, basically, which was able to get the best questions and the best answers the most exposure in the system.”

[maxbutton id=”1″ ]

Q: Users actually rated both the questions and the answers?

“There were many ways or many signals to this machine. Basically, we got a lot of traffic from Google. Google in those days believed very much in keeping you on their page for as short as possible. We also had a very special relationship with Google, where for about five years we had a special link in the corner of every Google English results page with a reference to So we got a lot of traffic from them. But what they did is spider and they indexed our pages very successfully. To give you a very simple example: we would get a new automotive question in the order of every 26 seconds on average. Someone would ask, “Where is the air filter in a Mazda 6?” – a very specific question. That question was asked, someone else may answer it, Google indexed it, and anyone in the world who wanted to know the answer to that question would find it. It would be a factual question but a very specific one, and we found that the more questions we had in the system, the more answers we had; the more answers – the more traffic. So it was a virtuous cycle which would give us opportunity. Some of the other questions we had were personal, about belief, relationships. People would ask questions they would never ask their mother. I’m not even gonna go into it – “Can you get pregnant from…,” it doesn’t matter. I’ll tell you the saddest question ever asked of our system. We had 17 million questions and answers in our database. The saddest question was someone one day asking, “Why do boys tell you they love you when they don’t?” Anyway, many of the questions were much more factual or much more scientific – politics, history, chemistry – you name it, religion even, but you should realize that not all questions have simple factual answers. Some are opinion, some are expository, and we’ve benefited from all of those because basically we were capturing data, textual data primarily, that would be used to build the engine.”

Q: Actually crowdsourcing intelligence. So if people would have asked their mother or if they were too embarrassed, rabbi or a doctor, now they would ask

“Just like Wikipedia, we could not guarantee the accuracy of the answers. In fact, those signals were reinforced by traffic, reinforced by edits, and reinforced by scores. There were a lot of different ways. Our goal was to get the highest quality answers, but it’s a difficult area and I would say that we used some machine learning techniques to measure the question quality and the answer quality, but at that stage we weren’t generating answers intelligently.”

Q: You could tell by the way someone phrases a question, types in a question, what aspect of the answer or what version of the answer, that you have in your inventory, they were looking for?

“No. At that point we were just reading. Sometimes we would have many different versions. So if you had a question such as “When did WWII begin?” that could be phrased in many ways – WWII, Second World War – and there were, in fact, many different aspects. Someone said: “September 1, 1939”; someone else said: “Well, with this invasion.” There are many different ways to look at something like that. The goal is getting information. By the way, nowadays Google does a fantastic job of giving a first cut at many of the answers, so not only can you ask what is the population of Athens – that’s easy – they’re actually answering questions like: “I wanted to know yesterday how to turn off shuffle on an album in my Apple Music, and they gave me a step-by-step thing.” So they’re doing much more. But systems like Quora and others, certainly Stack Overflow in the technical area, are fantastic at gathering that information today. And when we sold we were well aware of where we felt the industry was going.”

Q: So today when people have a question they go on Google, and some people even feel like they have a personal relationship with Google – it can complete their question. So if I’m a hypochondriac, Google always completes my questions with a disease or some catastrophe. What do you feel is the added value of the human-generated Q&A-based system like over machine learning?

“It’s symbiotic because Google is looking for information out there on the web. They also supply some of their own very well, but they’re looking for information. Other sites are presenting data and text and Google is very good at connecting it. People go directly to Quora and places like and certainly Stack Overflow nowadays. Nowadays I might ask a question of Siri or Alexa, but the world is developing to where people are just looking for very specific knowledge on demand.”

Q: What do you think the future of asking questions to get the right answers is? Where do you think we can go?

“I think there will always be a need, of course, for getting quick information at your fingertips, when you want it, where you want it, how you want it, etc. So it may not be textual and maybe a little bit more word-guessing in advance. Amazon is coming up with a system now where they’re going to send you what they think you want and if you didn’t want it you put it in the box and send it back for free. So there are all kinds of different models for anticipating requirements in use, but It’s a very exciting time in the industry and we’ll see where it goes from here.”

[maxbutton id=”1″ ]