In The Anti-Google I introduced the idea that what ad networks like Google are doing - transforming vast amounts of information into revenue by predicting your behavior - can be achieved with much less effort. Not only with less effort, but with a vastly superior value proposition for individuals.

Here's the key.

Consider that your daily behavior consists primarily of actions that extend threads established long ago. Occasionally, new actions establish additional threads. Each day brings different threads to the surface where they are duly extended by a further action.

Some threads operate on a weekly frequency like laundry day, others daily like showering, others annually, like spring cleaning. Some seasonal, some hourly, none perfectly regular. Threads can be high-level (containing many others), low-level, or anywhere in between. For a physical analogy, think of each thread as a unique wave.

Each of our interpersonal relationships implies a thread of action. As does each of our interests, goals, responsibilities, and possessions. Perhaps without being fully aware of it, we tend to manage thousands of these threads, each with their own complexity and frequency, each related in some way to all others.

It's the simple, low-level threads that provide the way in. I'll use an accessible thread like brushing teeth. Say that most days you brush once or twice, some days three times, occasionally not at all. Each occurence of brushing teeth represents an action on that thread. If the time of those actions is recorded, that produces a time series.

Example

The time series:

20, 25, 30, 45

Calculate the average interval between the most recent few actions (3 in this case), producing A.

A = ((25 - 20) + (30 - 25) + (45 - 30)) / 3
A = (5 + 5 + 15) / 3
A = 25 / 3
----------
A = 8.33

Calculate the time since the most recent action of that thread, called T.

NOW = 60, RECENT = 45
T = NOW - RECENT
T = 60 - 45
-----------
T = 15

Finally, the score S.

S = T / A.
S = 15 / 8.33
----------
S = 1.80

Now the wave is quantified. Very roughly, when S gets close to or above 1.0 the action is likely to happen again soon. That's enough to get the ball rolling.

Altogether a very simple method for predicting future behavior from past behavior. Too simple to find in a reputable textbook, no doubt, yet very quickly computed. It comes into its own when built into a very specific implementation that is then pushed in a particular direction.

For an application of this algorithm, check out Automated Group Activities.

The Anti-Google: Part II lays out the high-level structure and direction under which this algorithm thrives.