As we all know, binary math is the foundation of our computers. When combined with electricity, transistors, and I/O, it forms a bridge between physics and human logic. This bridge has allowed us to imbue nearly infinite logical capability into mass-produced machines, helping us solve our logic-friendly problems. So far that’s worked out pretty well.
But not all of our problems are logic-friendly. This new breed of binary, which I’ll refer to as predictive binary, deals fundamentally with logic-unfriendly problems; problems that are difficult to define, fluid, and perhaps without a concrete solution. It is for problems whose beginnings and ends are not known and perhaps unknowable. Distinctly human problems, in other words.
It happened or it did not
While the binary we’re so familiar with is the foundation of computing, predictive binary operates in software rather than silicon, above the level of the GUI but below the level of the network. It too is a bridge, but between humans and computers rather than logic and physics. Key to this bridge is a data structure that aligns very nicely with the natures of both human and computer.
The input into this data structure is event streams provided by a human through a user interface. Each event stream is homogeneous, containing events of the same type. Each event is an observed action and is as simple as “something happened”, or a binary 1. Before, between, and after each 1 are binary 0's.
This means a human is providing the raw binary input but also that anyone can do it since there's no binary logic or math involved. Like our machines can switch a laser on and off very fast to transmit in binary, a human can communicate in binary by, for example, simply pressing a button. Each button press becomes a 1 while the gaps between presses are 0's.
Predictions from simple data
How many more 0’s until the next 1? Each event stream becomes the basis for an estimate of how many more 0’s there will be until the next 1.
This ability to predict arises when:
- Each 1 represents a real event
- The stream is homogeneous
- Each binary digit represents a consistent interval of time
It looks like this:
...00001000001000000001000010000001(...?...)1 |--5--|---8----|-4--|--6---| (5 + 8 + 4 + 6) / 4 = 5.75
So the prediction is roughly 5.75 more 0's until the next 1, based on recent history.
Because each digit represents a consistent interval, we can also look at the prediction in terms of time. At a 1Hz frequency, the prediction above would indicate a peak probability of the next 1 to occur roughly 5.75 seconds from now. Of course, the event may happen immediately, may never happen, or may happen in an hour. It's a probabilistic expectation, not a guarantee.
Computers have an easy time of these calculations, as linear regressions and naive extrapolation (like above) both work. Like the building blocks of binary logic can be extended out to amazing complexity and capability, we see from machine learning that predictive ability is fertile ground.
These simple predictions are then integrated into and displayed by the same GUI that captured the original input, resulting in a feedback cycle of event input -> prediction output -> event input. This enables predictions, a fundamental of human cognition and psychology, to be used to reward and encourage further inputs.
We can compare classic binary to predictive binary like this:
Classic binary Predictive binary -------------- ----------------- Binary math -> Predictive algorithms Electricity -> Observed events Transistors -> Data structure I/O -> Graphical UI
Predictive binary has a structure that is quite different from classic binary, which as we know is divided into logical chunks like bytes, words, instructions, and at higher levels into cache lines, pages, and so on.
The structure of predictive binary of a different type. Its structure involves linking a collection of binary streams together into a type of graph, with each node in the graph representing a different binary stream. The structure of this graph is organized by the human so that the type of activity represented by any particular binary stream is more abstract (general) near the root of the graph and increasingly concrete (specific) toward the leaves. More details here.
The raw data of a binary stream isn't often used directly, rather its predictive quantity -- when is the next occurrence? -- is usually what matters.
Applying predictive binary to human problems
The solution to every human problem is rooted in behavior. Wishing, wanting, or waiting will not reliably solve problems. Only actions solve problems; appropriate and efficient action, ideally. Our behavior is the totality of our actions. Each of our actions is preceded by a choice (not necessarily conscious), and each choice is informed by the options we perceived to be available (again, not necessarily conscious).
Solving large-scale social problems requires changing collective behavior. Solving a single person's problems requires changes to that person's behavior. Why or how any problems came to be does not change how a problem is solved. Underlying complexity changes how difficult a problem is to solve, but not how it is solved.
Why do we change our behavior?
By looking for the solution to human problems, we find ourselves looking directly at behavior modification. And so we must ask, how and why do people change their behavior? You chose option A every day for the past month but today you chose option D. Why?
Well, there could be no reason at all; that's always a possibility. But when there is a reason, it's a response to new information. This new information might be a connection made by your brain from the older information in your memory. It might be something you heard from your family or friends, or from the media, or from a professional. The information might be a quantification of your past behavior, like a utility bill or a speeding ticket.
The thing about new information is that it changes our view of the future, a view which is a complex intermingling of predictions. Some information we receive is a prediction in itself, such as our doctor telling us we have 6 months to live. When we believe the source of the prediction is credible, our future choices tend to change, sometimes dramatically. When we hear of a more positive prediction, we tend to feel warm and fuzzy and double down on what we're doing.
Other information is more data-like, and when it is credible and clear like an electricity bill, we tend to make the predictions ourselves. Like with other credible predictions, we receive the signal that our behavior should change. A big electricity bill might lead us to let the house get a bit warmer in the summer and cooler in the winter, or being vigilant about turning off the heat/AC when out.
The domain of online marketing
Between information and behavior modification, we quickly find ourselves in the realm of marketing and sales. More specifically, targeted online advertising. Targeted advertising uses predictions about your future behavior (or the behavior of people like you) to put the right message in the right place at the best time so that you're most agreeable to it. The goal of advertisers is to improve the probability of you choosing option B (clicking the ad, buying the product) over option A (ignoring the ad).
The modern online experience is steeped in behavior modification. There's a huge amount of research and development behind this trend, and it's clear that it works. We know it works because the entire online ad industry is based on quantifying the impact of every ad campaign. Google's entire business is based on its ability to transform information about you into predictions and from predictions into truckloads cash from repeat advertisers.
The thing is that changing other people's behavior is difficult, especially when the information you have about them isn't very good. Good predictions are hard to make from information that is incomplete, old, poorly structured, and often plain wrong. Google and friends do well enough using advanced statistical and machine learning techniques on massive datasets but those tools are out of the reach of regular people.
Changing your own behavior -- such as to solve your problems -- is also difficult, but you have a distinct advantage. The information you have about yourself is as accurate and complete as you care for it to be. Perhaps most importantly, there's no conflict of interest: you want your behavior to change, and you know (roughly) how it has to change. It's becomes a matter of modifying your behavior for your own benefit rather than someone else's. That's quite a bit easier, and this is where predictive binary comes in.
Back to predictive binary
As I described above, predictive binary involves a series of homogeneous events. To achieve behavior modification, the events we record -- the 1's -- are each actions that we've actually taken, choices that we've made. Each stream, being homogeneous, then represents a type of activity. That means the next occurrence of each type of activity can be predicted within a useful range of accuracy. Then, the person whose behavior has been digitized this way gains the power of those predictions. The difference is that, unlike advertisers and other predators using prediction to gain competitive advantage, individuals can now capture that benefit instead.
The value of predictive binary pops out when we view it as the foundation of a set of tools that give each person the ability to modify their own behavior to align with the demands of their problems. This applies to any problem, no matter how deeply human it is. To better solve one's own problems is to better adapt and compete, and so get more of one wants and less of what one doesn't want. That's pretty good.
The overall result is a continuous, dynamic system from which we can modify our future behavior using high-credibility predictions based on our past behavior. There's no need to involve the Internet or any 3rd party, no more than classic binary does. Being self-contained, the system is as safe, private, and trustworthy as we choose it to be.