Editor’s note: This story is part of our Microsoft-sponsored series on cutting-edge innovation. Stefan Weitz is a director of search at Microsoft.
More and more, people in the search community are pushing to move beyond simple keyword-search capabilities towards a more omniscient search tool. Imagine if, instead of delivering up search results based on the keywords you typed in, your search tool was more of an omniscient oracle that could understand the nuances of both you as a person and as a searcher. Sound far fetched? What if when you asked, “find me a train from mountain view to san francisco,” the engine would do more than return links to sites that happen to have these keywords peppered throughout? What if the engine knew exactly where you were in Mountain View, looked at the current time, calculated how long it would take to get to the station, showed you the cost, highlighted which of the two tracks the train would arrive at, and had some traveler tips about how to buy the ticket for the train, since it knows this isn’t a usual routing for you?
Maybe oracle is too strong a word – maybe we’re looking at something more like a very smart friend who knows you very well. In any case, the dominant paradigm of today’s search model, given the fact that both the resources on the web and user behavior are constantly morphing, is undergoing a sea change tantamount to the shift from directory-based navigation to keyword-based search back in the late 1990s.
But wait – isn’t this back to the future? Where is my DeLorean? Didn’t we have these “agents” back in the late nineties? Remember all the talk about “push technology” that promised to deliver content that would interest you without your even having to ask? Remember PointCast? General Magic’s Portico? Microsoft’s Agent technologies (albeit not quite the same thing)? If we already tried this “I know what you want” thing back in 1990, what’s different now?
I contend one of the failures of push was that it was too pushy. What we need – and where we’re headed — is an agent that is less like your annoying cousin Harry, who interjects with things both relevant and not at a seemingly random schedule, and more like your best friend, augmented with the collective intelligence of thousands of really clever specialists. Something that can combine what it knows about you, the world, and the resources available both online and off to actually provide answers, not just more pointers.
I hear the assembled chorus now: “But I can almost always find what I need with search today! I don’t need this agent of which you speak!” And indeed customer research bears this out – most of you tell us you’re fairly happy with search. The problem we have is the same as with any evolving technology – we have been beaten into submission; we’ve reduced our expectations of what search should do. We already have in our minds that search will fail a good percentage of the time on our first try and that we’ll have to craft our queries in cro-magnon English in order for the engines to understand us. So in the universe of reduced expectations, search works pretty well.
But what if we expected more? Does the current model work? It can certainly be made to work with enough user ingenuity and time – much like you could paint a house with a toothbrush; but it’s terribly cumbersome.
I’ll contend that search does work today for many of the tasks it was designed for – namely, returning pages with content that mathematically correspond to the keywords you’ve entered into a box. But because of that nature, it’s woefully inefficient for today’s task-based queries. Too much of the computation happens in the organic processor (you), when engines could be better equipped to handle the rote tasks of assembling, distilling, and presenting information as knowledge.
Consider my current predicament – trying to find a jet ski rental in Dubrovnik. Intent is clear – and sources are fairly plentiful. However, getting usable answers to my query has proven nearly impossible. An illustration: here is my search pattern to accomplish this relatively simple task:
1 ) “jetski rental Dubrovnik”: a bunch of links to content sites – helpful, but since I don’t really have a map of Dubrovnik in my head, knowing what is close to me versus what is on the other side of town is nearly impossible. So …
2) Switch to maps. Try the query again. Woefully bad results across all engines. Some point to hotels, others to rental sites for apartments, others to descriptions in Italian.
3) Decide that maybe I’ll have better luck if I figure out what “rental” is in the Croatian language. Head to a translation site and figure out “rental” = “iznajmljivanje”. Try some more queries. Try some more maps.
4) Start to make some progress, but quickly realize many of the rental places actually don’t let you take them out on your own but require an escort. That’s hardly conducive to me doing donuts on the Adriatic. Start an entirely new query path to try and find unescorted jetskis.
And it goes on…
The point? We can do better – much of what I did could have been done silently on my behalf. An intelligent agent would know more about me – it would know where I currently am, using the GPS in my device. It would surmise my query isn’t academic but task-based and would present a more logical map-based result to my query. It would know my native language is English but that I’m querying for options in Croatia, so it would silently proxy my request in the native tongue. It wouldn’t rely on standard search algorithms to use anchor text or word clustering on a content page, but rather scour sites and information across the web, looking for jetskis and Dubrovnik, mining the content and extracting the useful content from the page (including hours of operation, whether I can rent without escort, etc.) and presenting that information as a useful summary. In short, it would combine any number of variables about me, the request, and what it understands about the real world in order to present a set of options that doesn’t simply send me running around to different sites.
And that is the difference between the push agents of yore and where we are headed. Many services are popping up on the web that can provide structured context that these future agents will be able to use. Look at LiveMatrix (for events), OneRiot (for realtime content), BookingBuddy (for reservations), Rummble (reviews for you), WolframAlpha (parsing of unstructured info), and something like Siri (recently acquired, but a good exemplar of the stitching together of multiple sources of information to present actionable knowledge to a user).
So get ready for cousin Harry to be relegated to the den where he can watch the Pink Panther in peace, and start to open your mind to the possibilities of what your agent can do.
Weigh in with your thoughts on this kind of search technology in the poll below.
Stefan Weitz is a Director of Search at Microsoft and works with people and organizations across the industry to promote and improve search technologies. While mainly focused on Microsoft’s product line, he works across the industry to understand searcher behavior and to gather and distill feedback to drive the product beyond the state-of-the-art.
To read more recent stories in this series, visit the Conversations on Innovation site, or click on the headlines below:
5 signs that customer co-creation is a trend to watch
The future of email: from conversations to decisions
Overcoming information overload?
The subscription economy is here. Are you ready?
As new phones emerge, mobile app development is a moving target
A future of full of touchscreens? It’s all in the software
Cleantech’s next generation: smaller, nimbler, smarter
TV 2.0: Hulu’s flatlining, and the networks are ready to innovate
What will it take to make mobile payments mainstream in the US?