How Can We Discover The Salient Attributes of Sentient Brains?
As described in an earlier article, we use the term sentience to describe how a brain, natural or synthetic, may sense the environment and produce meaningful behavior in that environment. We seek to understand the minimal brain architecture, along with its necessary and sufficient salient attributes, that gives rise to sentience.
In this article, we will fly-over approaches to the problem of finding the salient attributes of sentient brains, and consider two alternatives: (1) reverse-engineering natural brains, and (2) observing the behavior of synthetic brains.
Reverse Engineering Natural Brains
When we look at gross anatomy or use electron microscopy of any small area of any brain, we are often left with more questions-- mostly, questions such as "what is this?" and "what is it for?" This is because natural brains are highly-optimized systems which have endured billions of iterations of natural selection-- nature's product release cycle. The optimizations allow the organism to perform better in its environment, but they may obscure the actual salient features themselves. For example, the Cerebellum is a significant feature discovered to be like a numeric coprocessor which provides fine-tuned motor control and is essential for competitive survival in land most animals, yet is not actually necessary for sentience to arise. This gross anatomy lesson tells us that we needn't look to the organization of the cerebellum to understand sentience, for example.
Recent investigations by researchers have been able to produce whole- or partial-brain connectome atlases of organisms such as mice and humans. These results reveal both large-scale and small-scale complexities. At the higher levels, trunks between regions of the cortex can be seen, so that major interaction pathways are suggested. At the lower levels, cannonical microcircuits forming minicolumns and macrocells in cortex are suggested. These findings steer hypotheses in the direction of using these general paradigms in synthetic models.
Scientific research with medical patients having brain lesions or disease has enabled identification of components of the brain which are involved in functional areas. These discoveries have guided brain research for decades. These studies can be very helpful to narrow our search for salient attributes.
Functional MRI similarly provides a moving tool that can be used to identify what areas of the brain are active when performing a task. These studies can also narrow our search for salient attributes.
The necessary and sufficient salient attributes will likely not be the ones we might today. The sentience effect may arise through a complex resonation between large brain areas, or perhaps on a small scale, a simple predictable pattern of connectivity that is repeated across minicolumns that, as a whole, gives rise to the effect. We may discover that sentience is an observable effect, yet not actually a physical thing, in the same way that electrons are not actually things, but rather a collapse of the wave function of the quantum electric field.
The Huge Range of Scale of Natural Brain Plans That Work
At the smallest scale, we start with C. Elegans, a worm about 1mm in length with 302 neurons in its nervous system. This organism has been the subject of exhaustive study in several ways. By 2002, its genome of 100,000,000 base pairs had been completely sequenced. And in 2019, the entire organization of the neurons (including types of neurons and their connections), referred to as a connectome, was recorded. This worm senses the environment and exhibits feeding, avoidance, and mating behaviors, and is successful in the wild on its own. It is interesting to view the C Elegans connectome with this online interactive connectome viewer.
At the higher (but not highest) end of the scale, we find Homo Sapiens with approximately 8.7x1010 (87bn) neurons in its brain. Our scale of complexity is large, from 302 to 8.7x1010 neurons, or eight orders of magnitude.
The number of neurons is not the only measure of complexity-- the connectivity between neurons allows them to interact and operate as a single system. This has been widely-studied and reported for many species; however different brain regions have different organizations and connectivity varies (for example, there simply aren't very many neurons in C. Elegans), but in general, for larger organisms, nature tends to keep afferent synapses on a neuron to under 10,000. Quick math tells us that humans have 8.7x1010 times 104, or 8.7x1014 connections as an upper bound, which is large indeed.
In fact, a wide range of brain sizes have been proven by nature to be successful:
- Pond Snail - 11,000 neurons
- Lobster - 100,000 neurons
- Ant - 250,000 neurons
- Honey Bee - 960,000 neurons
- Cockroach - 1,000,000 neurons
- Frog - 16,000,000 neurons
- Rat - 27,000,000 neurons
- Mouse - 71,000,000 neurons
- Guinea Pig - 240,000,000 neurons
- Octopus - 500,000,000 neurons
- Cat - 760,000,000 neurons
- Dog - ~2,500,000,000 neurons
- Macaque - 3,440,000,000 neurons
- Baboon - 10,950,000,000 neurons
- Chimpanzee - 28,000,000,000 neurons
- Orangutan - 32,000,000,000 neurons
- Gorilla - 34,400,000,000 neurons
- Human - 86,000,000,000 neurons
- African Elephant - 257,000,000,000 neurons
Assuming that there is only one fundamental architecture that causes sentience to arise, nature uses it to deliver sentience on a scale spanning eight orders of complexity. From this observation, we can conclude that it is not necessary to model large organisms; small ones exhibit the same fundamental effect sought.
Nature's Evolution of Brain Architecture
The earliest forms of life, simple microbes, made their appearance on Earth some 3.7bn years ago. After some 1.3bn years, cyanobacteria evolved, making food using water and the Sun's energy, and releasing oxygen. Over the next 500 million years, nature evolved eukarotic cells with their more structured organization. These cells were able to group together, forming the first multicellular sponges, the earliest animals, some 800 million years ago. By the Cambrian Period (some 541-485 million years ago), a multitude of life forms appeared (i.e., trilobytes), with varied body plans, heads and tails. At this point, nervous systems must have started forming to make animal behavior possible. Brains then, however primitive, have been around for a half billion years, although it wasn't until about 360 million years ago that the first mammals appeared, with a small neocortex.
For the next 360 million years, millions of species evolved, each with a different brain plan, and each brain plan was tested with billions if not trillions of animals. Although earlier, simpler brains had a fixed connectome, or wiring, now even the most modest brain plan had a different wire-up for each individual. This means that there is a huge set of possible functional connectomes, so that specific wiring is not a salient attribute of sentient brains, but an implementation detail.
Interestingly, it's not necessary to have a mammilian brain for behavior to emerge; even Drosophila Melanogaster, the common fruit fly with a brain employing fewer than 100,000 neurons, is able to exhibit behavior that enables it to be successful in the world.
Observing Behavior of Synthetic Brains
Another approach to understanding how brains work is to create models from our hypotheses about how brains work, and run those models on real-time simulators to observe their behavior. This approach has the advantage that it is easily supported by the scientific method of hypothesize and test. With the right model building tools and a powerful enough simulation engine, a researcher can construct a large number of models and run them on the simulator in an iterative fashion, revising the models as their behavior is observed, so that it can be established which design paradigms work, and which ones don't.
Real-time simulation is essential to the observation of behavior. A simulator must operate on the same simulation timescale that biology does (1ms), because that is the low-end of the timescale that biological organisms use to interact with their environment.
New modeling tools are necessary to build models that can run on the simulator. NeuroSynthetica's Workbench enables researchers to design models with SOMA™, NeuroSynthetica's modeling language that can be compiled and generated on the simulation server. Changes can be made to the model's source code, and the model regenerated, as necessary during the model's development lifecycle. Visualization tools in the Workbench enable the researcher to monitor the overall simulation as well as individual circuits and neurons, as the real-time simulation runs.
To continue, read the next article about real-time Simulation.
Notes & Sources:
-  Wikipedia, List of Animals by Number of Neurons.
-  Note: Some fixed-wired brains, such as C Elegans' 302-neuron brain, may not exhibit the level of sentience of interest.
-  National Museum of Natural History, Early Life on Earth--Animal Origins.
-  New Scientist, A Brief History of the Brain.