After a brief discussion on another blog pertaining to whether or not humans possess some kind of an innate ontology or other forms of what I would call innate knowledge, I decided to expand on my reply to that blog post.
While I agree that at least most of our knowledge is acquired through learning, specifically through the acquisition and use of memorized patterns of perception (as this is generally how I would define knowledge), I also believe that there are at least some innate forms of knowledge, including some that would likely result from certain aspects of our brain’s innate neurological configuration and implementation strategy. This proposed form of innate knowledge would seem to bestow a foundation for later acquiring the bulk of our knowledge that is accomplished through learning. This foundation would perhaps be best described as a fundamental scaffold of our ontology and thus an innate aspect that our continually developing ontology is based on.
My basic contention is that the hierarchical configuration of neuronal connections in our brains is highly analogous to the hierarchical relationships utilized to produce our conceptualization of reality. In order for us to make sense of the world, our brains seem to fracture reality into many discrete elements, properties, concepts, propositions, etc., which are all connected to each other through various causal relationships or what some might call semantic hierarchies. So it seems plausible if not likely that the brain is accomplishing a fundamental aspect of our ontology by our utilizing an innate hardware schema that involves neurological branching.
As the evidence in the neurosciences suggests, it certainly appears that our acquisition of knowledge through learning what those discrete elements, properties, concepts, propositions, etc., are, involves synaptogenesis followed by pruning, modifying, and reshaping a hierarchical neurological configuration, in order to end up with a more specific hierarchical neurological arrangement, and one that more accurately correlates with the reality we are interacting with and learning about through our sensory organs. Since the specific arrangement that eventually forms couldn’t have been entirely coded for in our DNA (due to it’s extremely high level of complexity and information density), it ultimately had to be fine-tuned to this level of complexity after it’s initial pre-sensory configuration developed. Nevertheless, the DNA sequences that were naturally selected for to produce the highly capable brains of human beings (as opposed to the DNA that guides the formation of the brain of a much less intelligent animal), clearly have encoded increasingly more effective hardware implementation strategies than our evolutionary ancestors. These naturally selected neurological strategies seem to control what particular types of causal patterns the brain is theoretically capable of recognizing (including some upper limit of complexity), and they also seem to control how the brain stores and organizes these patterns for later use. So overall, my contention is that these naturally selected strategies in themselves are a type of knowledge, because they seem to provide the very foundation for our initial ontology.
Based on my understanding, after many of the initial activity-independent mechanisms for neural development have occurred in some region of the developing brain such as cellular differentiation, cellular migration, axon guidance, and some amount of synapse formation, then the activity-dependent mechanisms for neuronal development (such as neural activity caused by the sensory organs in the process of learning), finally begin to modify those synapses and axons into a new hierarchical arrangement. It is especially worth noting that even though much of the synapse formation during neural development is mediated by activity-dependent mechanisms, such as the aforementioned neural activity produced by the sensory organs during perceptual development and learning, there is also spontaneous neural activity forming many of these synapses even before any sensory input is present, thus contributing to the innate neurological configuration (i.e. that which is formed before any sensation or learning has occurred).
Thus, the subsequent hierarchy formed through neural/sensory stimulation via learning appears to begin from a parent hierarchical starting point based on neural developmental processes that are coded for in our DNA as well as synaptogenic mechanisms involving spontaneous pre-sensory neural activity. So our brain’s innate (i.e. pre-sensory) configuration likely contributes to our making sense of the world by providing a starting point that reflects the fundamental hierarchical nature of reality that all subsequent knowledge is built off of. In other words, it seems that if our mature conceptualization of reality involves a very specific type of hierarchy, then an innate/pre-sensory hierarchical schema of neurons would be a plausible if not expected physical foundation for it (see Edelman’s Theory of Neuronal Group Selection within this link for more empirical support of these points).
Additionally, if the brain’s wiring has evolved in order to see dimensions of difference in the world (unique sensory/perceptual patterns that is, such as quantity, colors, sounds, tastes, smells, etc.), then it would make sense that the brain can give any particular pattern an identity by having a unique schema of hardware or unique use of said hardware to perceive such a pattern and distinguish it from other patterns. After the brain does this, the patterns are then arguably organized by the logical absolutes. For example, if the hardware scheme or process used to detect a particular pattern “A” exists and all other patterns we perceive have or are given their own unique hardware-based identity (i.e. “not-A” a.k.a. B, C, D, etc.), then the brain would effectively be wired such that pattern “A” = pattern “A” (law of identity), any other pattern which we can call “not-A” does not equal pattern “A” (law of non-contradiction), and any pattern must either be “A” or some other pattern even if brand new, which we can also call “not-A” (law of the excluded middle). So by the brain giving a pattern a physical identity (i.e. a specific type of hardware configuration in our brain that when activated, represents a detection of one specific pattern), our brains effectively produce the logical absolutes by nature of the brain’s innate wiring strategy which it uses to distinguish one pattern from another. So although it may be true that there can’t be any patterns stored in the brain until after learning begins (through sensory experience), the fact that the DNA-mediated brain wiring strategy inherently involves eventually giving a particular learned pattern a unique neurological hardware identity to distinguish it from other stored patterns, suggests that the logical absolutes themselves are an innate and implicit property of how the brain stores recognized patterns.
In short, if it is true that any and all forms of reasoning as well as the ability to accumulate knowledge simply requires logic and the recognition of causal patterns, and if the brain’s innate neurological configuration schema provides the starting foundation for both, then it would seem reasonable to conclude that the brain has at least some types of innate knowledge.
The trouble with this, is that it is hopelessly vague.
I suggest you sit down and try to design an AI system based on these ideas. That will force you to try to flesh out the details. And then you will find that some of them cannot be fleshed out.
You may say that this is hopelessly vague, but I’d argue that it nevertheless shows an innate property of the brain that our developing ontology is based on. Furthermore, I’d argue that the concept of hierarchical connections and relationships isn’t really a vague concept at all. It is fairly well defined as the concept of connections having different hierarchies and thus different levels of connections and interconnections that involve branching, as if from a tree, where those branches have their own sub-branches, and where some of those sub-branches may have direct connections with branches at various higher levels up the tree.
That one would or wouldn’t make an AI system based on these apparent inherent principles in the brain doesn’t show that those apparent principles aren’t the case for our brain. The design for an AI system has pragmatic limitations and thus likely wouldn’t exactly mimic the way our brain develops over time with physical rewiring mediated by various biological processes involving self-assembly and chemical changes over an extremely large number of molecular components. Rather, it would likely be accomplished by software that serves to produce and utilize similar hierarchical relationships between data that’s processed.
Having said that, I think that there will likely be various parallels to our own brains’ configuration in efficacious AI design that involves hierarchical pattern recognition, even if it is accomplished through other means or platforms than a biological brain utilizes.
I suggest you try and design an AI system that ISN’T based on hierarchical pattern recognition and hierarchical relationships, and thus those that parallel the brain configuration I’ve discussed here. I’m guessing you won’t get very far.