Searching Outside the Box

You’ve probably used one of these today

For most of us, the search box is the entry-point to the web and a vital conduit for information access and retrieval.

But can you remember the last time it had a noticeable upgrade? For all of the growing hype around AI and advances in machine learning, we don’t often hear how these technologies are going to impact on search. It would be fair to wonder if there is indeed anywhere for search to go, whether we’ve reached ‘peak’ search capability.

I believe that web search is going to change, and in a big way. But it won’t look or feel like the familiar search box, and neither will it be a voice interface or chatbot. To understand where search is headed, it helps to understand where we are just now.

The Modern Search Engine

Google has defined web search over the past two decades. Their UI is universally familiar and perfectly optimised. A search field for typing natural language, a button that begins the search, results presented as a list of 10 blue links, each with a summary. We’ve seen new additions such as targeted verticals for images and shopping, the knowledge box for information on entities from Google’s knowledge graph, and of course, sponsored listings. But otherwise, not much has dramatically changed over the past 10 years or so.

The major changes have occurred under the hood. Google diminished PageRank’s contribution long ago and there has been a huge amount of machine learning research applied to making search better such as learning to rank, user behaviour models, personalisation, and, of course, deep learning. All of this in pursuit of better search rankings.

Extensive resources have been poured into perfecting the list of search results, ensuring that those 10 links are the best they can be. Each advance slightly improves well-worn metric scores such as nDCG and MAP, measured on standardised data sets such as those published by TREC and Google’s vast search logs. Yet on the surface, these marginal gains are scarcely noticeable.

I believe that we’ve taken the search results page to its limit. For its intended purpose of being a destination whereupon you trade words for the starting point of an online journey, the current UX and technology works extremely well. It’s familiar, dependable and now a core part of the infrastructure of the web, as vital as the Back button, the URL bar and the hyperlink.

Is Voice the Future of Search?

Many believe that the successor to search already exists and it’s called Alexa. For me, voice assistants are simply a solution to a new problem caused by the fact that the way many people now access the web is changing. Last year, mobile access to the internet surpassed desktop usage for the first time. People are accessing information from their sofas, in restaurants and while they’re on the move. And with this their requirements for search have also changed.

Searching using a mobile is not a great experience. The highly successful desktop search conventions have been shoe-horned onto a device for which they aren’t well designed. It’s difficult to type keywords using your fingers on a screen with one hand, then deal with the slowdown on auto-suggest caused by intermittent data connections only to land on a results page that is still a wall of text with small links.

Therefore, it is only natural that we are now exploring interfaces incorporating speaking and text messaging as modes of interaction. The growing list of voice and chatbot assistants attests to the progress that’s been made, and therefore we arrive at spin-off devices such as Amazon’s Alexa that are enabled by these new hands-free interfaces. Gone are the lists of search results, replaced instead with spoken responses or chatbot comments. Gone are the ad hoc 2 to 3 word queries, replaced instead by questions and commands.

Yet, despite the enormous strides in speech recognition that power this new interface, our usage is still limited to the sort of short, simple queries we would use in a search engine. We aren’t having conversations with our devices yet and we won’t be for some time. This creates as many problems as it solves: verbalising commands slows down the information finding process and alters your train of thought, it’s awkward to use in public or at a desk, and any kind of in-depth information finding becomes tiring and confusing.

Voice is an essential UX upgrade catering to an emerging medium for information access, but it’s still search within the box. So if voice isn’t web search’s next great leap forward, then where else can it go? I believe that there are two interrelated areas for improvement that will be critical to revolutionising web search.

Context

What is context and what does it have to do with search? Context is everything that exists outside of the words you put into the search box. It’s the location you’re standing in, the time of day and where you just ate lunch. It’s the 5 searches you’ve made in the last hour, the email you read and the conversation you’re currently having on Slack. It exists in the spaces between the words you use to describe your intent.

The greatest difficulty in optimising the modern search engine is that there is only so much understanding you can derive from the 2.4 words that the average user enters into the search box. This is one of the reasons why Google use cookies to track your behaviour across the web to learn your preferences and habits (and to advertise to you, of course). Each of us has a Google feature vector describing our hobbies, our demographics and our purchase history. When you search for something ambiguous like “eagles”, your preference for classic rock may net you different results than your ornithologist neighbour.

This is a form of context in aggregate that has had success in traditional search, but it’s limited in that it can only tell you so much about why a search was made at that moment in time. Context evolves. Context encapsulates every action you took before you searched and it influences everything that’s likely to happen next.

The huge problem just now is that you are the arbiter of your context. As you do things online, you have to personally remember everything you’re discovering, connect the dots as you go along and figure out what to search for next. You retain all of your own context, which makes the meaning of your search query seem obvious to you but obscure to the tools you’re using. Web search cannot advance until this problem is solved.

Tasks

Everything you do online is a task, both at work and at home, even your visits to Facebook. From the broader perspective of your ongoing task, a web search is simply a discrete unit of work that sits alongside other units such as reading, comprehension, data collection, creativity, learning, decision-making and action-taking.

Context is the language of tasks. Using current technology, the completion of a task involves a process of mental context management; finding and remembering information, piecing it together and making conclusions. Tasks materialise and become clearer as context is collected, and once a critical mass has been reached then actions become enabled.

The limitation of the modern search engine is that your task starts with your search query and ends with a click. Features such as query suggestion do acknowledge that you may wish to return to the search results page, but ultimately, once you’ve clicked one of those blue links, you’re on your own.

The Future of Search

Web search’s progress has stalled because the modern search engine doesn’t understand you or what you’re doing. It’s great at one-shot queries and questions but it can’t help you beyond that, not until it can deal with the task you’re trying to complete, and the context in which that task exists. For search to move forward, it needs to expand beyond simply being a transaction where you trade words for a list of search results. It’s no longer about giving you 10 million different starting points, but rather, helping you get to the end of an information journey.

It is a dialogue between you and your search system, comprised of the actions you’re taking and the actions your search system is making alongside you. As you do things online, your search tool should be a cooperative ally understanding the content of what you’re looking at, gradually piecing it together to build up a contextual picture for you.

The search system of the future won’t be a search box but will instead be the platform in which you do things. It will no longer really be seen as ‘web search’, this aspect of it will be relegated to the background and offered as one among many of its services, which will include other technologies such as voice UI and email generation. In fact, these services are likely to be interchangeable and offered by specialist 3rd parties. This platform could be described as a work companion or apprentice, working alongside you to help you do things and always learning how to be better.

Right now, it’s difficult to imagine the functionality of this new search paradigm given that the technology or its UX doesn’t currently exist. However, considering that the primary tool for information access just now is the web browser, I can imagine it will at first take the form of a sort-of smart AI-enabled browser. One that doesn’t just dumbly render HTML, but rather understands webpages and how they fit together, thus building up the essential user context. Task assistance will come in the form of webpage augmentations and automated synchronisation with other applications, one-click buttons that take your context and do meaningful things with it. Search will be one automated service among many.

However, the search engine won’t disappear. It’s still perfectly designed for being a starting point destination, or else a last resort for those tricky one-off tasks your smart browser can’t handle yet. These are the same reasons we use search engines now and this won’t change. Instead, the tools of tomorrow are going to understand what we’re doing outside of the search box, and more importantly, why.

A Comment on Automation

Consider a future where this search platform does exist, where everyone has a personal context “cloud” that plugs into all of their services and helps them with everything. The more the platform is used, the better it gets at predicting that user’s particular tasks and offering them assistance. Consider what it can begin to learn about how to do tasks in general, how to deduce new solutions having seen similar tasks performed millions of times already. Almost like a new kind of PageRank, where the aggregation of connected human actions (akin to user-curated hyperlinks) helps teach systems how to do work.

Knowledge automation is the endgame for search, and task and contextual understanding is the next step there, perhaps even the key to unlocking it.