Information Overload

We should retire the idea that goes by the name "information overload." It is no longer useful.

The Internet scholar Clay Shirky puts it well: "There's no such thing as information overload. There's only filter failure." If your filters are bad there is always too much to attend to, and never enough time. These aren't trends powered by technology. They are conditions of life.

Filters in a digital world work not by removing what is filtered out; they simply don't select for it. The unselected material is still there, ready to be let through by someone else's filter. Intelligent filters, which is what we need, come in three kinds:

  • A smart person who takes in a lot and tells you what you need to know. The ancient term for this is "editor." The front page of the New York Times still works this way.
     
  • An algorithm that sifts through the choices other smart people have made, ranks them, and presents you with the top results. That's how Google works— more or less.
     
  • A machine learning system that over time gets to know your interests and priorities and filters the world for you in a smarter and smarter way. Amazon uses systems like that.

Here's the best definition of information that I know of: information is a measure of uncertainty reduced. It's deceptively simple. In order to have information, you need two things: an uncertainty that matters to us (we're having a picnic tomorrow, will it rain?) and something that resolves it (weather report.) But some reports create the uncertainty that is later to be solved.

Suppose we learn from news reports that the National Security Agency "broke" encryption on the Internet. That's information! It reduces uncertainty about how far the U.S. government was willing to go. (All the way.) But the same report increases uncertainty about whether there will continue to be a single Internet, setting us up for more information when that larger picture becomes clearer. So information is a measure of uncertainty reduced, but also of uncertainty created. Which is probably what we mean when we say: "well, that raises more questions than it answers."

Filter failure occurs not from too much information but from too much incoming "stuff" that neither reduces existing uncertainty nor raises questions that count for us. The likely answer is to combine the three types of filtering: smart people who do it for us, smart crowds and their choices, smart systems that learn by interacting with us as individuals. It's at this point that someone usually shouts out: what about serendipity? It's a fair point. We need filters that listen to our demands, but also let through what we have no way to demand because we don't know about it yet. Filters fail when they know us too well and when they don't know us well enough.