Synchronized Swimming and Lessons for AI in Search

26 / August 2016,

Synchronized Swimming and Lessons for AI in Search

The 2016 Olympics have concluded. The Closing Ceremonies mark in some ways a return to normalcy for many of us. While many events both inside and outside of the competitions generate continued buzz for a few weeks, for the most part our world of sports returns to the professionals we watch around the year. Still, some events maintain a place in our consciousness that does not fit the relative anonymity of the competitions outside of Olympic years.

Synchronized swimming has a way of locking into our minds, much in the way that the background functionality of artificial intelligence has worked its way into the public consciousness, and into the thoughts and nightmares of SEO content providers. Both feel magical and miraculous at first blush. When you demystify the processes, though, you learn that mechanics and processes make both work.

The Invisible Machinations

Part of the magic that we see in synchronized swimming comes from the apparent grace and ease with which head and arm movements occur above the water. We see what appears to be ballet-like movement from individuals, performed in unison by an entire team of swimmers. They smile, wave, and dive with what appears to be effortless movement across the pool. Under the surface, though, legs and arms churn with power and speed. The unseen effort pushes the swimmers up and across in ways most of the spectators never get to see.

Artificial intelligence works in much the same way. We see search results that seem to intelligently analyze search strings and provide sensible results that do not depend exclusively on keyword phrases, and imagine a thought process underway that eliminates the benefit of SEO planning. The hype that accompanies the technology certainly feeds this impression. Even the nomenclature of “artificial intelligence” and “machine learning” pushes the concept of a computerized being thinking through the information with which it is provided, and then producing results in a creative, dynamic way.

The technology involved in AI is impressive, and processes information quickly and effectively. Like the swimmer’s unseen legs, though, the processes underlying AI’s performance in processing search queries proceed not through miraculous discoveries, but rather through high-speed processing of data gathered from a high quantity of inputs. RankBrain and its ilk quantify data, run probability analyses across existing searches and functions, and produce results that fit the numbers. Impressive, to be sure, but mechanically explicable.

Learning and Growth


The artificial intelligence that search engines apply, much like the work that swimmers do, builds from a foundation presented. For synchronized swimmers, it comes from a base of swimming technique and strength. Time and dedication go in to every movement, as swimmers learn to use their lower bodies to propel and support their upper bodies. The dance techniques and the ability to move together in a group require more work, one step at a time. The process of learning and growing into the work they do takes time, talent, and effort—the three classic ingredients required of any recipe for human success.

AI, too, might feel initially like a giant leap forward for technology, and in some ways it is. The movement to get there, though, has been

incremental, moving from basic programming to data analysis, and up to the point that processing speed and sophisticated programming have combined to produce the tools we have today. The information that AI produces comes from the quality and the quantity of the data that enters, run through a code-based protocol that helps provide increasingly accurate search results for individual queries.

The technology is exciting. Just as Deep Blue processed and ran chess scenarios with which even the best chess player in the world could not keep up, so current programs run searches in more impressive ways than humans could do. The programs will continue to become more sophisticated, in the progeny of RankBrain, Watson, Siri, and the like, but they still represent technology built on research, experimentation, and coding, processed effectively and quickly. Processes have built on each other to create the present tools we use.

Difference Between One Part and Many

The greatest effect that synchronized swimming produces comes not from the ability of one swimmer to do incredible things, but from a team of swimmers learning to do them together. Coordination of efforts that results in movements that duplicate and complement each other lifts an entire performance, both artistically and technically. A great performance combines showmanship and technique to produce something more impressive than either element would allow by itself.

AI, too, functions well because it combines different elements. The search string analyses that it performs puts words in context, and identifies meanings based not only on the words and phrases, but how and where they occur. This allows a tool like RankBrain to determine that mercury in a thermometer is fundamentally different from Mercury holding astronauts inside it, or that a search for “Barack Obama’s wife” does not seek information about Barack Obama himself. The speeds at which the technology processes and analyzes the data, combined with the sophistication of the analysis itself, gives the magic to what AI can do.

But we can go a step further as well. AI analyzes search strings to determine meaning, and this helps a search achieve the results the internet user wants to reach. But RankBrain is not itself the algorithm Google uses. Rather, it is part of the overall process that allows search results to emerge. Keywords matter, and the mass of inputs that RankBrain analyzes necessarily lead to the output it gives. In other words, the system process the data available to it; it cannot function outside of that data, even if the volume with which it works creates enough input to make it feel that way. A wholly new search, should such a thing exist, would not allow AI to process meaning outside of the connections it can identify to similar iterations.

Just as a group of swimmers cannot succeed outside of the practiced dynamic, so AI cannot work outside of the programming and inputs it has received. It remains a magnificent programming achievement, but it remains dependent on everything else around it.