We just passed a major milestone at Cogo Labs. The Lab recently exceeded 10 petabytes of collected and stored data on consumer web activity. This is a 100% increase from the amount of data we had access to when we started Vestigo three years ago.

A few weeks ago we ran XPLR on over 400,000 unique websites that were showing early signs of viral consumer traffic. We have encouraging ongoing conversations from the batch of early stage FinTechs the algorithm identified. The continued growth of Cogo Labs’ data assets is critical to our investment success.

How to conceptualize 10 petabytes? One petabyte is 500 billion pages of standard printed materials. If you converted 10 petabytes of data into the Robert Fagles translation of The Iliad, you could produce over 7 billion copies. If you stacked these copies on top of one another, you could just about reach the moon. It is a staggering amount of data to query and learn from, an ocean of information to find signal in. It is our edge.

With that volume of information in mind, this issue focuses on the data science making its secrets known to us, a key focus for our companies at Vestigo, and the amazing applications they’ve developed so far.

—Mark & Dave

Turning Back the Clock: Symbolic Processing or Machine Learning

Aside from unlimited cold brew, one of the great perks of my job is the opportunity to make Dave tell me stories about the “old days” of machine intelligence. At that time, (believe it or not) MIT students were not encouraged (financially or academically) to study machine learning (ML). Marvin Minsky was running MIT’s artificial intelligence lab and researchers were focused on a top down approach to called “Symbolic Processing.” The fact that you are looking it up on Wikipedia (as I did), illustrates how successful that technology has been.

Less aesthetically enticing, ML required a vast amount of computing power and since it was out of favor, Dave had free rein to use the connection machine after 10pm. Walking the well trod path of other MIT probabilists like Claude Shannon and Ed Thorpe, Dave achieved technical breakthroughs in the middle of the night in an MIT “computer” room.

It was clear to Dave that computing power was reaching a point that enabled ML applications to dominate in a business that could scale relatively narrow human tasks. The actual task or industry was less important than having a well-defined problem. His team could therefore create software for digitizing handwriting, reading mammograms, and demand forecasting at Walmart before ultimately building the first predictive algorithms for Amazon.

It is in this field of so-called “narrow AI” that methods from data science continue to generate the greatest commercial value. This is the same approach we used to develop XPLR, which enhances our ability to find early-stage FinTechs. The majority of our portfolio companies use supervised learning in one form or another; Dave’s perspective is invaluable in both due-diligence and aiding our founders in their efforts to leverage their own proprietary data.

AI shows no sign of relinquishing its position as the most investable technology in FinTech today. As a true pioneer of commercial AI, having Dave’s network and expertise helps us intelligently invest in its continued proliferation.

—Frazer Anderson, Investment Analyst