A bunch of buttons to visually predict your own behavior
So you can get more of what you want
I’d like to introduce a system of buttons that actually empowers the average person. It’s a system that I believe can grow to contribute significant value to anyone’s life and even become a radical alternative to the usual types of personal computers we are surrounded by.
In the unlikely event that you’ve been following along through parts 1 and 2, you might recall that I first introduced a data structure called Graph of Timelines which is capable of representing observed human behavior to any degree of detail or complexity. I followed that up with an algorithm called Contextual Prediction that uses the data from the Graph of Timelines to efficiently make contextually relevant predictions of your own future behavior.
I’ve promised a graphical user interface to tie everything together. This animation shows how it arises from the ideas expressed in parts 1 and 2.
I’ve simply taken the graph of timelines, where each node is both a context and a type of activity, and made each node a button. As we know, buttons provide a kind of binary input: pressed and not pressed. And as we also know, nearly anything can be represented or computed via binary encoding. That’s how computers work. That’s how the telegraphs in ancient times worked.
The all-powerful button
The beginning of this GUI is a single button.
As I described in Part 2, Contextual Prediction, the reason to press this button is because you took action. Not just any action, but the type of action you have associated with this specific button. The type of action can be anything at all, no matter how large or small. The scope could be as large as “I won a land war in Asia” or as small as “I blinked”. Usefulness is somewhere in the middle, naturally.
Pressing the button causes the current time to be recorded, generating a timeline. That timeline becomes the basis for a continual prediction of when the next press might occur. It’s not possible to predict exactly when or if that next press will happen, but even a simple extrapolation is useful in the context of human behavior.
The following animation shows how the timeline develops and the prediction arises from the timeline. The button is pressed and so a timeline is generated that moves off to the left. From that timeline comes a projection of when the next press might be. In this animation, the darkest circles represent higher probability and the lighter circles lower probability.
As you can see, the prediction changes each time a new event is added to the timeline.
While a single button pressed at a ridiculously high frequency could represent pretty much anything, that’s not practical for humans. We need a lot more buttons to keep track of our view of the thousands of different types of activity and the tens of thousands of actions we take throughout the day. Ideally these buttons are spatially organized because we’re quite good at spatiality and visual pattern recognition.
So here’s a bunch of buttons, organized spatially. An interactive cluster, if you will:
Notice that you can shift the focus around, which is necessary for contextual prediction.
Each circle is, at the same time:
- A behavioral context
- A cognitive context or focus
- A type of activity
- A button which, when pressed, records an instance of its type of activity.
One of they key properties of this navigable bunch of buttons is that of mirrored focus. Mirrored focus means that when you enter a behavioral or cognitive context — you’re doing household chores, you’re focused on your code, or you’re making sales calls — it’s easy and natural to maintain a parallel focus within this cluster. By mirroring the two focuses — the one in your life and mind and the one in the cluster — the options presented to you from the cluster of buttons are strongly biased toward immediate personal relevance.
Adding in predictions
As per Part 2 of this series, Contextual Prediction, the predictions from each timeline are sorted by immediacy and used to filter out activities that are unlikely to reoccur any time soon.
The threshold of likelihood can be set anywhere between 0 and 1 and freely adjusted to suit the immediate need. In the example below, the threshold is set to 0.15 but can be shifted via a scroll-wheel, finger swipe, or another method. The gray activities are unlikely to happen soon because they occurred relatively recently.
After filtering, what remains visible are the buttons/activities likely to reoccur. Here’s an example of how this filtering is visualized by the cluster layout:
As you can see, visual filtering like this, whether user-driven or automatic, is a natural extension of the fundamental properties of the cluster layout.
This cluster layout works so well because it mirrors the underlying structure of human behavior. Each of our actions occurs within a hierarchy of context, beginning broadly (e.g. having fun) and narrowing down to the specifics of what you’re doing (e.g. beach volleyball doubles). The structure of the activities within the cluster reflects the contextuality of behavior: general, encompassing activities near the root, increasingly specific activities toward the edges. I go into more detail in Part 1, Graph of Timelines.
Keep in mind that you’re pressing these buttons when you act, which means that all predictions are about your future behavior. The entire system is all about you. Your behavior is already highly predictable, as much as 93% of it, even though it may not seem that way. By arranging your activities into this particular cluster layout, and using the predictions as a filter, clear efficiencies emerge:
- Fewer button presses are needed to capture information about your behavior.
- Valuable information arises from extremely simple inputs.
- Each person is inherently intimately familiar with their own gradually-grown structure.
- Focus more on what is likely to come next.
Solving a common, vexing problem
There’s a common problem that this cluster of buttons is well suited to help solve, even in its early form. This problem is the continual decisions of what to have for dinner, where to go for lunch, what to do this weekend, what to do with the kids right now, what to do when bored, and many more. Continual decisions like these vex millions of people and yet there is no standard tool to turn to for help.
Repetitive decisions can be exhausting, making it easy to fall into the trap of least resistance, also known as comfort and familiarity. And yet these types of decisions are critically important because of their long term effects. If you choose well on a regular basis you’ll likely be rewarded with an easier and more successful life. If you take the easy short-term choice too often, you’ll likely find your life becoming harder and full of risk. I’m sure you’d prefer an easy and successful life, as I would, yet the emotional exhaustion of day to day life can lead any of us down the wrong path.
The Daily Dinner Dilemma
Let’s start with dinner. Our daily choice of dinner affects our finances, our health, and much more by extension. We can end up broke, obese, and diseased if we make poor dinner choices too often for too long. Alternatively, we can fuel our larger ambitions by making economical, efficient, and healthy choices.
This first animation shows an initial decision to eat out followed by a filtering out of unlikely choices. “Frankie’s” and “Shiro” remain. These two choices are predicted based on the pattern of past choices. Thus they are likely to be acceptable, perhaps even ideal, and so they are very easy to say yes to.
The second animation shows an initial decision to eat at home. Presumably the cluster was explored in the morning or the prior evening so that the beef stew ingredients could be prepared and placed in the slow-cooker. Otherwise, on short notice, a sandwich might seem the better choice. Either choice is inherently a reflection of recent behavioral patterns and so is very likely to be accepted. It’s unlikely that any the options presented will ever be a surprise.
To gain this decision-making help, all you need to do is press the right button after dinner, add new buttons when there’s a new dinner, and occasionally reorganize.
What you end up with is a very easy way to first narrow down your high-level preference, like whether to eat at home or eat out, and then perhaps further narrow down the degree of planning, effort, or cost, and then end up with an agreeable suggestion. Or you can filter down directly via the “Dinner” context if it’s approaching dinner time and you have no opinion at all. Always, you may freely explore the patterns of your past choices for inspiration. Whatever your mood or level of energy, you have help at your fingertip.
By pressing these buttons, at minimum you gain a useful tool to help you make important and regular decisions that tend to pop up at the exact time you’d rather not think about anything at all. That’s nice. You also gain the ability to see your patterns laid out so you can consciously accept or reject choices you know are especially good or bad for you. It adds up to an escape valve or a pressure release, one small tiny step at a time. After several of these small steps, you might finally have the time and energy to make headway on the parts of life you care much more about.
And, of course, when you experience how this system of buttons helps with the dinner dilemma, you may be tempted to expand your button cluster to include lunch, breakfast, and snacking. Then perhaps to meal planning, grocery shopping, and exercise. Perhaps also to weekend plans, and activities with the kids, and who to reconnect with, and what to do when you’re bored. All the same class of problem, and only one of many classes of problems this system of interactive contextual prediction helps you solve. More to come.
My project Benome is the implementation of the ideas expressed in this series. Still in its infancy, Benome is free, Open Source, and uses one weird trick to dial privacy up to eleven.