In 2006, Scott A. Jones co-founded ChaCha.com, a search engine that uses human guides to answer queries. Now they’re expanding to the mobile space, allowing users to call or text their nuanced questions in plain English, and receive succinct answers in minutes. But with the amount of data on the Web growing exponentially, are human beings really the best tools for trolling it all? FastCompany.com sat down with Jones, also the co-founder of Gracenote and a voicemail pioneer, to talk about the future of mobile search, the irreplaceable nature of human intelligence, and why he doesn’t fear Google.
The amount of data online is growing, but more and more people are trying to access it on small-screen mobile devices. How will that alter the way search is performed?
Well, in order for search to be effective on phones, we’re going to have to dig through more data, but the results will also need to get more and more specific. Other companies like Google, Yahoo, and Ask are struggling to do that — to come up with deep, rich, succinct content from their searches. They’re having problems, because it’s not like algorithmic searches are always going to give you a relevant result as the first, second, or third thing on the page.
What’s so bad about having the user do her own filtering?
I think that in many situations, having human mediation is key. Especially when the mediation is from people who know something about whatever you’re asking – sports, entertainment, medicine – because they can sort through search results at their desktop, and get you back succinct answers very quickly at moments when you need them most. Those are what I call the “peak demand” moments. Right now, when you need something on the run, you don’t want to mess with your smart phone’s browser. So what do you do? You call or text your friends, so they can point you in the right direction. That behavior is already there, so it’s my gut feeling that because of the way technology is trending, humans need to be in this loop for a while.
So Cha-Cha is actually better suited for mobile phones than for computers?
We knew that it would be better for phones from the beginning, but we knew we had to build out our backend first. Twenty years ago I was working on voicemail technology, and I had science fiction ideas about the future of phones: that, for example, there would be a small phone in my ear that could access The Library of Congress – this was before the Internet — and also have access to practical information, like the best restaurant in town, and even which waitress to ask for while there. That’s the idea behind ChaCha.
What about searching outside of the “peak demand” moment? Why do we still need human mediation in those searches?
Before giving a speech at the National Academy of Sciences a few years ago, I was on Google, Yahoo and Wikipedia, searching for information on what new technology was coming out, trying to prepare what I would say. I called my friends at Sequoia, my CTO, my other friends, and they’d always give me answers that sent me back to the Web. That was the “aha!” moment: What if I could do that across all topics? That would be a hugely powerful thing. I was still thinking of this in terms of mobile phones, but we started at the desktop version first to figure out the taxonomy of managing our guides. In fact, most of our 25 patents are about managing the human army of guides, not about spiders or indexing.
Do you really have thousands of people out there waiting to respond to my search query?
We’ve had 40,000 guides — maybe 50,000 — come through our system, and they come from all walks of life. Some are college students, some are work-at-home moms, retired librarians, people who like Jeopardy, people who like working on Wikipedia. Some people do it for the money, others do it to help the online community, and some do it to learn more about a topic they’re interested in. To become a guide, there are some basic requirements, and they have to take a test online. They login online to take the tests, which take a few hours, and within a day they can be online and active.
You’ve announced that soon these guides will be able to accept not only SMS queries, but voice queries as well. In a world where most voice-activated systems are mediocre at best, is voice-powered search really viable?
There are actually multiple layers of problems you’d have to solve to make voice-powered search work: the first is speech recognition. Even the most advanced speech recognition software in the world requires correction of every fourth or fifth word. Sure, you can do some special tricks, give the system a limited vocabulary — but if you try to do a generalized search it just doesn’t work. The second layer of problems involves natural language recognition. Even if your system can understand the words you’re saying, it won’t necessarily get the syntax; now the question becomes, “What’s actually being said?” The final piece of the problem is the algorithmic issue: We need to get to a place where we’re coming up with half a page of results instead of two million. The algorithmic problem is far from being solved, and the big guys don’t really know how to do it any better than they’re doing it now.
Of course, the easiest way to solve problems one and two is to take them out of the equation completely — and we do that by using people. We have a bunch of different types of guides; if you call in to ChaCha, you get sent to a transcriber, who transcribes what you’ve asked and sends it to an ambassador guide. That ambassador triages that query, checks if the question has been asked before, and if it hasn’t, sends it off to a specialist, generalist, or expert. The ambassador tees up the question that way, as if it’s a game of volleyball — pitching this query around, and trying to get the best answer for you. And it all happens in a couple of minutes.
So the ambassador can access a database of previously asked questions? Do you hope that eventually the top questions will fill your database and make your guides’ responses faster?
As the databases of questions grow, those databases will definitely assist our guides with finding answers more rapidly. On the backside, we have a mechanism that enables expert guides to go through various answers to one question and decide which are best — those “best” answers will appear higher in their results. That’s actually better than simple crowd wisdom because the best answer is picked by an expert.
Sounds like you must like the idea behind Mahalo, then?
There are lots of ways to apply people to search, and Mahalo is going for the short tail of queries – the most common ones. That’s a good direction, if Google lets them get there. Who knows, maybe Mahalo is a place that we ought to be sending people to with some of our reference links. Maybe there’s a partnership brewing. I view most of the other guys as potential partners.
Does that mean you don’t fear Google?
They do some things really well. Take Google Mobile: It’s good for simple questions like sports and weather; even if it can’t give you much depth in its answers, it’s very fast. I actually think there’s an opportunity to partner with, or be on the shoulders of, the other big search engines. The founders of Google fundamentally believe in using algorithms over people, so we don’t really compete with them ideologically.
How will ChaCha draw revenue? Can you do targeted advertising via SMS and not annoy your users?
I don’t know that our users would see targeted SMS advertising as benign, and right now I’m experimenting with different payloads for delivering ads. I’m not happy with them yet. There are several places to put that extra ad message – in a separate text message, at the end of your answer, or on the page you access with the reference link you get at the end of your text. We will choose the method that is least offensive.
Our best idea for a solution is to send out ads so targeted and local that they’ll be viewed as a positive part of the search result. To do that, we intend to leverage our human knowledge at choosing ads as well.