Part 0: Biological Imperative
I wrote a proposition about intelligence that I think is cool. Colliding Minds Part 1: Mental Modeling the Future. This post provides some of the backstory behind the proposition, some of the biological evidence to support the proposal. It’s not extensive nor is it meant to be authoritative. It’s more of a written sketch that illustrates some of the inspirations for the proposal.
I’m going to ask you to take a trip with me in time, from the primordial beginning of life…
Imagine if you can, being a single celled life form. It may have some cilia (hair like projections) or a flagella (tail like projection) or maybe pseudopods (body morphing like an amoeba). It probably has some form of sensory capability, an energy attraction of some kind (thermal, photonic, chemical) because it needs that energy to live and it has evolved that attraction over millions of generations.
It has no nerves and no brain of course, it’s just one cell. But it happily moves along, adjusting course toward it’s goal, growing, feeding, excreting and procreating. It might have an anti-attraction to something, cold or toxins or predators but no real response other than to move away. It’s more of a desire to move towards a more favorable environment that a desire to ‘run away’ from danger. So, sometimes things eat it, sometimes it gets stuck, sometimes it seeks out freedom. It’s a pretty haphazard existence really. Not much of a base for intelligence. But it is turning into an information processing machine, of that there is no doubt.
Now imagine being a primitive multi-cellular animal, like a hydra or a water bear. It now has new opportunities as each of it’s cells can now collect information independently and share that information with it’s neighbors. A new capability arises wherein cells on one part of it’s being can have a different set of sensory data from the other parts of it’s being. Over millions, billions — probably trillions, of generations the ability to communicate this sensory data is selected for and primitive neuron cells are evolved, specializing in transmitting an electrical or chemical signal from one part of it’s being to the other parts. It now has a very simple neural network, one that can learn how to coordinate it’s actions through what amounts to a sensory feedback loop.
About this time it also discovers that it has some new sensory organs, sensory mechanoreceptors like proprioreceptors and nociceptors. Actually it has a lot of them and they are sending feedback from all around your being — internal and external, which is now somewhat of a body — a body of cells that is. It probably even have some ganglion, or clusters of nerve cell bodies that are helping to respond to incoming signals (afferent nerves) by monitoring the frequency and volume of signals coming in from areas of it’s body, sharing this information with motor nerves (efferent nerves) which then send new signals to muscles or other responsive tissue. Coordinated movement based on a concentration of sensory data coming in… awesome. No brain yet though, just some ganglion to aggregate sensory data, measure it and then respond with instruction signals after some useful criteria is met.
It doesn’t really start to get interesting until these ganglion nerve cells get together and start comparing notes though, forming a plexus of ganglion. Think about it. Each ganglion is a cluster of nerve tissue that is measuring the frequency of incoming signals, which themselves are being tightly regulated based on the intensity of the sensation being transmitted. You’ve suddenly got a lot of data flying around electrically and chemically and the beginnings of complex structures that can start to coordinate their actions in surprising new ways.
Okay, this all makes sense but why is it significant in regards to intelligence? Well besides having a history of sensation (which we’ll talk about soon enough) it also gets you space and dimensions. By noting the relative position of the body over time and comparing with external stimuli, you can get a measurement of an external object in relation to the self. It’s like feeling in the dark for us humans. We can develop an impression of an object by reaching out and touching it in several places. Using this information you can establish the point in space relative to yourself where you touched or collided with the object. This of course works even better with something you can ‘multi-touch’, such as when holding it in your hand, as you get even more sensor data coming in parallel bursts and fewer variables to account for.
In short, our multi-cellular, ganglionated being is getting it’s first model of self and of the external world. It might even be able to model it’s interactions with the outside world. A bunch of nerve tissue clumped up in close proximity to each other may have generated the world’s first “mental” model, millions of years ago. Animals, even the most primitive (and some say even plants) may be vastly more intelligent than we give them credit for. On the other hand it also becomes quite clear that certain types of intelligence are limited by the sheer quantity and quality of sensory data a being has access to.
Towards general intelligence in higher order animal species.
So far we’ve looked at a general hypothesis of collision as a mechanism for intelligence and covered supporting background evidence from the natural world. We’ve also briefly mentioned how it can be applied to higher order abstract thinking. Let’s dig in a little more deeply and approach some fundamental knowledge we humans have, other animals appear to have and that should be derivable from any theory of general intelligence such as the proposed hypothesis.
So much of our understanding of animal intelligence is based on behavioral observations, so little is based on the biology that underpins the behavior. When a rat navigates a maze to find a treat, we know that it learns the maze, we know that there’s activity in the hippocampus and of course in the basal ganglia which is where all higher order animals control movement. What is so clear to me is that these areas of the brain are not working in isolation, physically or temporally. Recent studies have shown that the animals not only record their experiences in memory, they even may replay them to help in determining future decision making choices. A simulation is the best explanation for how this would work. Extending the collision hypothesis may help to illuminate the mechanics for how this would work while further refining the hypothesis itself. In this example we can think of the rat as overlaying prior experiences with current incoming sensory information, assessing how closely the two models align, how close they come to a collision. The more data available in both the prior experience and the current experience, the more detailed the model will be and the higher the fidelity of the simulation.
Consider that the rat however is far more intelligent and capable of much more sophisticated decision making, both strategic and tactical. Of course it has a fully functioning mammalian brain, not just a nerve net or simple ganglia of nerve cell-bodies — a distinct advantage. However, it is still dependent on the same exact information as what is available to the most primitive animals. What makes the difference is the amount of information being collected and how the data is being used. The first of these is easily accounted for; more densely packed sensor cells, more variety of input. The second is a more nuanced topic and one we’ll touch on in more detail in the course of examining a model for artificial intelligence in Colliding Minds Part 2: Intelligent Machines (not yet published).
Wherein I propose a theory of mind based on comparing mental models as simulations used for making predictions and grounded in biochemical processes.