The Washington Post wrote earlier this week on Google’s increasing use of knowledge boxes in its searches – the inset boxes at the top of search results that attempt to shortcut the search process by displaying the actual factoid of interest on top of the traditional infinite page of hyperlinks. As users increasingly access the web through mobile and voice, the goal of such systems is to get the user an answer such as “how many ounces in a pound” or “who is the president of Estonia” as quickly as possible. Whereas search engines of the psat simply returned a pile of links for a user to wade through, the goal of knowledge boxes is to provide the actual response the user is looking for by leveraging advances in natural language processing to have machines actually understand the user’s question.
According to the Post these factoids are now displayed for almost a third of Google’s 100 billion monthly searches, meaning they are playing an ever-increasing role in mediating our access to the world’s information. The rise of bots across the communicative continuum from workplace tools like Slack to social communication like Facebook means machine interpretation of the world’s knowledge will increasingly supplant the historical concept of the keyword search.
Anyone who has spent a bit of time searching the web or perusing social media is likely intimately familiar with the sheer magnitude of false and misleading information deluging the online world from the classic claims of a flat earth and a faked moon landing, to dangerous medical misinformation, to false rumors that lead to real ethnic violence and insidious claims like Holocaust denial. Though some of these may be easy to spot, the volume of entries on rumor fighting sites like Snopes.com offer a sad lesson in the gullibility of the online community.
In many cases, however, it may be unclear to a non-expert user what the generally accepted answer is. For example, someone researching a medical condition online might be confronted with a myriad of conflicting recommendations across dozens of seemingly authoritative looking websites, some of which might recommend life threatening home remedies. As I have seen firsthand over the years, even within the hallowed halls of academia, individuals with their doctorates (and sometimes two doctorates) who are full professors with decades of experience in their narrow field all-too-frequently fail the basic task of researching something outside of their field and recognizing what is fact and what is fiction.
The Post article cites the head of research of the Wikimedia Foundation as lamenting the rise of knowledge boxes, that the trend “undermines people’s ability to verify information and, ultimately, to develop well-informed opinions. And that is something I think we really need to study and process as a society.” Yet, such an ideal assumes that people are inherently good at fact checking and reasoning about complex and sometimes highly technical information outside their field of expertise and resolving a myriad of conflicting claims from a wealth of different sources. Instead, I frequently find that even those in academia struggle with the simple task of translating a question into a search result and understanding which sites are more likely to produce a reasonable answer.
Thus, as the Post points out, Google has partnered with the venerable Mayo Clinic to ensure its responses to medical queries correspond to the best available understanding of the medical community. More intriguingly, Google has also tied searches for food ingredients to their nutritional facts provided by the USDA (search for “cheddar cheese” for example to see it in action). Imagine if a search for the nearest restaurant serving a particularly unhealthy menu item yielded not only directions to the restaurant but also a summary of just how unhealthy the item was and the potential health impacts of consuming it on a regular basis. Might this be the next logical step in our health and fitness technology saturated world?
Thus, one could argue that the rise of such knowledge panels (and the corresponding summarized answer derived from them that is provided through voice searches) have the profound potential to cut through the fog of misinformation online. Imagine if Google automatically displayed entries from Snopes for searches of popular rumors in the same way it displays traffic alerts when returning driving directions.
Of course, when others make decisions for us, differences in opinion can lead to conflict. This is especially the case when technology companies try to determine concepts like “authoritativeness” at a global scale, as Facebook learned earlier this week. Even in the case of medical searches, a practitioner of a religion that discourages the use of modern medicine or an adherent to certain brands of alternative medicine might strongly disagree with the consensus of the Western medical community as to the best treatment regime for a given condition.
One of the early promises of “social search” services was that they would replace the universalness of traditional search engines, which historically provided the same set of results to every searcher worldwide, with answers custom tailored by your social connections. The premise was that someone searching for the “best” Thai restaurant in Washington might agree more with recommendations by his/her close friends than the consensus view of the Internet at large or the combined scores of the city’s most-read restaurant critics. Instead, search engines today are increasingly personalized, using a variety of technologies from deep learning to their vast stores of historical data to individualize each search result, replicating some of the benefit of "knowing" a user and meaning the same user will see different results over time as the algorithms learn his/her interests.
This leads to the interesting question of whether instead of striving for a single “correct” answer to every search, we might instead aim for providing users with an “informed summary” of available information. In this case searches that have a single authoritative answer, such as “what time does the metro close tonight” might yield a traditional response, while a question for which no generally accepted answer is available, for which there is genuine disagreement in the professional community or where there simply is no objective answer (such as the “best” restaurant in a given city) might offer the primary viewpoints and their respective sources for the user to decide. At the very least, the source of a given piece of information should be stated as prominently as possible and sourcing should become an integral part of information response, much as the website of each search result has become a standardized part of search display, to encourage users to think critically about the information they consume.
Or, given the level of personalization of modern search, might it be possible for search engines and intelligent agents of the future to learn a user’s interests and tailor such responses automatically, knowing for example that a user prefers deep dish pizza, wishes to spend less than $40 and wants their meal within 30 minutes and using that information to transparently determine the “best” pizza restaurant in the same way a social search service might leverage the collective domain knowledge of a user’s social network to learn the same factoids? In such a case the system might display each of the characteristics it believes about the user and has used to filter the results, so the user is aware of the assumptions being made and can alter them (such as to adjust the amount spent when celebrating a friend’s birthday to allow for a more formal setting).
In the end, one thing is clear and that is that the simple keyword searches and infinitely scrolling pages of results that have come to define the computer era for the past half century are giving way to a new future in which machines increasingly filter our information for us, making ever more decisions on our behalves in what we should know about the world. The enormous power and potential of these new intelligent systems offers incredible promise for changing how we consume and interact with the world’s information, but also raise many questions about how we will design these systems into the future. Moreover, as computers increasingly curate our information consumption, the wealth of our societal history that has yet to be digitized will rapidly fall to the wayside with only that which has been digitized available to our automated overlords. Welcome to our computer mediated information future.
