Embedded Prophesy Devices

When you walk on a sidewalk you are pretty sure it won't kill you as you step on it. Somehow, you know that moving your foot forward and shifting your full weight on it will lead to you being one step closer to where you want to be, with no averse effects. It won't, for instance, bite your foot; it won't move suddenly out of your way; it won't propel you backwards; and it won't turn into a 10-thousand-strong procession of Brazilian carnival dancers that trample you to death at the beat of a Samba. Not in the next step, nor in any of the steps that you intend to take until you reach your destination.

But you (as in "the biological system that is now parsing and comprehending these words") are a bunch of neurons and other connected cells nested deep inside the reddish darkness of flesh and bones. All you know about the sidewalk is that it reflected photons in a certain pattern—shades of gray, rectangles, perspective effects...—in the past. Even assuming that those photons you scooped up with the two pinholes on the surface of your skull is perfectly accurate, all you have is outdated information. You have zero information about what the sidewalk will be like when you set your next foot down.

Of course, you somehow know it won't have changed much. Not in most cases. How do you know that? The answer is that you have a model of the sidewalk inside your head.

Somewhere, spread across a great number of interlinked neurons, you have a replica of the sidewalk, and of every other solid walking surface, that you use to predict the future. As soon as you get enough photons in patterns that tickle your "sidewalk" neural network strongly enough, a prophecy is made: you can move your feet in a walking motion and all that will happen is that you move forward, no explosions, no harm, no one dressed in glittering costumes shaking their behinds at you.

The "sidewalk model" is there in your head first and foremost so that you can predict the future and get to where you're trying to get. And, similarly, prophesy is the function of the "door handle model" that you fire up when you feel like being on the other side of the door; and of the "Jim model" when you're considering inviting your friend Jim to a jog in the park; the "H key model" when you want the letter H to appear on your monitor; the "my skin" model when you decide how hard to scratch your itchy arm without shredding it; and all the other things you do, second by second, as you live your life.

Without a vast, detailed replica of the world encoded as models inside your body, you wouldn't survive one day.

Those models are not the same as the physical things they are modeling. In fact, they are very different from most points of view—a network of a few thousand neurons looks and feels nothing like a door handle. That's alright. They don't need to be the same, as long as they behave similarly enough to produce reliable prophecies when and where you need them.

How do these models work? We don't know the details yet but, from first principles, we can make a couple of assumptions.

First, the predictions made by these models are not, in general, on/off statements but probability distributions. I find that a "tree of possibilities" is a great way to think about it.

Given what we know about the system under consideration and the specific circumstances, we have a range of possible outcomes, some more probable than others. This is the part that really needs to match the modeled system well for the model to be of any value.

The second thing we can say about how mental models work is that they must be made up of heuristics. Since the models in our heads have physical configurations very different from anything existing outside mammal skulls, they can't possibly be subjected to the same mechanisms of energy transfer and physical forces as the things they are meant to prophesy about. In other words, you don't need to actually rotate your synapses 30 degrees in order to imagine turning a door handle. What's probably be happening is a collection of if-this-then-that rules of thumb.

• "If the apple has nothing to support it, then it falls down"
• "If I put my hand on that flame, then I will feel great pain"
• "If I yell at Lucy, then she will be upset and this and that other bad thing will happen"

When we say that we "learn" something, what we really mean is that we're creating or upgrading our mental models. You can do that by being taught by others or simply by observing what happens around you. The vast majority of what you learn happens through the latter, direct experience. You don't need to be taught that the apple will fall, you observe all sorts of things fall from a young age, and form your mental model of that phenomenon on your own. Reading about Newton's theory of gravity might give you more nuance and expand the applicability of that model, but the core of your predictive power takes root well before you even learn to read.

Some cognitive psychologists speak of the huge role that metaphors play in the way we talk and reason about the world. Perhaps metaphors are how the brain optimizes its memory storage and access: if I already have a "model for X" with a tree of possibilities and heuristics that are similar to those of system Y, why forge a new one?

Notice that I made a silent conceptual leap a few paragraph ago. I went from talking about models in general to mental models, happening inside brains. That's the kind of models we are most familiar with, and those we can do something about. But mental models are only a subset of all possible future-predicting models. What does a "non-mental" model look like?

A first category of non-mental models is the collection of things we know instinctively, without the need to learn them through experience. These are the models hard-wired into our nervous systems by evolution, like the innate fear of wild predators, and the sense of disgust at dirty or decaying things. While they are still centered on neural processes, these models don't require consciousness. They make predictions—the tiger will eat me, the rotten fruit will make me sick—and those predictions automatically trigger an appropriate physical reaction. This toolkit of prophesy devices we share with all other animals.

But we can expand the category further. Forget about a brain, even a nervous system is optional for the formation of internal models of the world. For example, most bacteria are capable of chemotaxis, the directed movement as a response to a chemical signal. Take E. coli, the most widely studied single-cell organism. E. coli has sensors on its surface to detect favorable chemicals, and can swim to where the concentration of those chemicals is higher. That means that, somewhere inside it, the bacterium is making predictions about the future: "moving now might get me to a place with more food." No thinking is involved, of course. The prediction directly triggers, by chemical means, the activation of the flagella on the cell's membrane, causing it to move (it needs some trial and error to find the right direction, though).

This is quite amazing when you think about it. Not only is a rule-of-thumb-driven simulation of the external world constantly running inside every skull in existence, but that same kind of simulation, to various degrees of sophistication, is running somewhere in the body of every living being. In this sense, prophesy is one of the core functions of what we call Life. ●