Editor's note: Nadav Gur is the founder and CEO of Desti, a virtual personal assistant for travel incubated by SRI. Previously, he was founder and CEO of Worldmate, the first mobile travel app.
"Any fool can know. The point is to understand."
–Albert Einstein
Since the launch of Siri on the iPhone 4S last year, the media has been abuzz with the potential implications of what's next – from Google's Eric Schmidt commenting that Siri poses a great threat to Google, to countless articles by VCs and thought leaders.
Has artificial intelligence finally come of age? And is it ready for broader applications in industries ranging from travel to finance? Are we destined to grapple with fast-following Siri clone after Siri clone, or will the category evolve?
Siri excels at setting reminders (and a little less so at ordering Scottish lunches). But is she ultimately more than a better front-end for basic smartphone functions?
This post is about changing how we use computers to manage knowledge, and not just information.
Knowledge = Information + Meaning
You may "know" that it's 100 degrees out there today, but unless you also know that 100 degrees equates to meaning that it's hot, that piece of information is pretty useless.
To understand the concept of "100 degrees", one has to know the meaning of 100 degrees and the concept of hot. That sophistication is called domain knowledge. Domain knowledge allows one to give meaning to information. As a result, the value one gets from knowledge helps make better decisions. The cashmere sweater stays home today.
With knowledge comes understanding, which helps users make better choices. Better choices mean fewer mistakes, and an increase in productivity.
Better Search Through Understanding
This focus on meaning is the foundation of semantics and semantic search (an oft-abused term), which ultimately means searching for concepts, not keywords.
Over the last 10+ years we've all been conditioned to expect search engines to basically match keywords to documents. We then receive a list of a gazillion links of pages that include some permutation of our keywords. It is then the user's job to manually sift through all of these listings in the hope that one of them is a good match.
It goes without saying that this is a lot of work. A semantic search engine, on the other hand, would attempt to understand what we're looking for, and then retrieve the best results, whether or not the specific words we used are mentioned or not. The real promise of this approach is that by understanding our intent, we will get more relevant and more accurate results.
Part 1 is to understand what the user is asking for, and part 2 is to understand what is discussed on a particular page. Siri has made strong headway into literally understanding you (voice to text) but more importantly about deriving meaning from what a user has just said.
The Holy Grail is to take this ability to understand what people asked for and to also understand what's written across the billions of pages comprising the Internet.
Instead of "organizing the world's information" you'd be organizing the world's knowledge.
Understanding what people are asking or saying and converting it into usable meaning is a huge first step towards making heads or tails out of the billions of pieces of information scattered across the web.
The traditional way of interfacing with information, whether with keywords or by refining searches with sliders, menus, or widgets just doesn't cut it anymore. To truly unleash the power of search is to use natural language, in voice to text or even text to text. With this new paradigm of starting a search we will be able to better unlock the value that is hidden in unstructured data and provide infinitely better results.
This is why the natural language front end of Siri, a forerunner in the category, is such an existential threat to Google (or at least their existing keyword-based search).
Google understands this better than anyone else, and their recent announcements of the forthcoming Google Semantic Search is a direct response.
The Rise of Vertical VPA's
The immediate next step is to help systems develop domain expertise, and this is no easy task. Even for the brightest people domain knowledge is something that can take decades to achieve, and even then people can't be a true expert on everything.
Such is at the core of advancing us beyond the search paradigm of today's digital information to an age of Virtual Personal Assistants.
SRI, who of course developed Siri from decades-long research on the CALO project, understands that for VPAs to be truly useful, they need to be focused on a specific domain of knowledge.
SRI actually has developed that full core technology and begun systematically applying it to specific veriticals. It was recently unveiled as the foundation underlying BBVA's Lola, a virtual personal assistant for banking. My company Desti was likewise incubated at SRI, and is built on top of the same stack.
We're creating a trusted advisor who can take a user's travel intent and provide directly relevant answers and actions, all tailored specifically to that individual. According to Google, the average user visits 22 sites prior to booking travel. Planning a trip can be unnecessarily frustrating and time consuming.
What if planning your travel was as easy as interacting with a friend from the area you plan to visit, one whose on-the-ground recommendations cut through the clutter with aplomb, and who knows what you enjoy?
SRI's overall VPA play is our well-reasoned effort to apply the Siri concept in a way that enables it to be adapted to specific domains or verticals, encapsulating the specific language, knowledge and reasoning that are unique to each.
Instead of teaching an assistant to be all things to all people, we are teaching one virtual assistant to be a bank teller (Lola), another one to play educational games (Kuato), and yet another to be a travel agent.
This is an extraordinarily difficult challenge, but one that is finally becoming a reality. The proof will be in the pudding of course — in how well we instill domain expertise in each new venture.
Summary
We want a VPA that understands us as individuals — that leverages our past for our present. A VPA has deep domain knowledge, and can finally move us beyond a big list of blue links. Software really can provide us with real and expert assistance when and where we need it most.
What we're seeing is a paradigm shift on multiple levels, one that will play out more profoundly over the coming year than it did the last. We're on the cusp of an entirely new thing:
- Managing info => Managing knowledge
- Matching keywords => Matching concepts
- Serving "searches" => Serving "actionable intents"
At the end of the day, what people really want is to be…understood. That starts with a conversation, and it gets better when that conversation begets a real relationship — one based on mutual respect, a shared language, and a certain intimacy.
No hay comentarios:
Publicar un comentario