Hip
Connection level weights in a simulation are software data, and the instantiated algorithms used to process them are software. In the brain they are neuronal connections and underlying biochemistry. Neural networks are not the brain. The brain can be described in terms of software, but the description is not reality.
First, much of the physical function in brain architecture is defined by unmodifiable neural structure. Its not random. It has evolved to operate in very specific conditions, and to do very specific things. The brain is not an undifferentiated single hidden layer neural network, its a highly adapted complex architecture that in neural networks terms includes many specialist neural modules, with specialized function, that interact in complex ways that includes feedback loops. Much of that specialized function is unmodifiable.
Specific subsections and regions appear to have a high degree of plasticity, where such connections can be modified. If our entire brain suddenly did that though we would be suddenly dead. Much of the architecture, especially deep body functions (hormones, metabolism, etc) cannot work this way. The visual cortex may have some limited localized plasticity, but it cannot work this way. Higher brain functions are highly plastic though, and it appears that a large part of motor function has high plasticity. I suspect this has a lot to do with memory and consciousness, though that does not explain plasticity in motor function. That is a behaviour and stimuli driven adaptive mechanism that I do not understand, despite that I studied computer models of such things.
Modelling such architecture is what my Ph.D. was aimed at. I designed a neural network simulator that would run complex architectures, including feedback loops, and used biological simulation equations not the standard comp sci ones. However my math brain died at that time, and the huge volumes of numbers coming up on my UNIX terminal became squiggles. A friend of mine who was working on a similar project helped me build that software, and as my brain went from bad to worse I did less and less, but I still understand the principles behind it.
The assertion that the brain runs software is, as I have explained before, a category mistake. Describing the software level as data is a system theoretic abstraction, or a symbolic abstraction. Its actually physical structure with elecrochemical activity.
You can describe a car in terms of the theoretical principles of the internal combustion engine, but asking a computer modeller who has never even held a spanner to fix it is a bad idea.
The notion that the brain is equivalent to a Turing machine is debunked, as I have said before. Its a fundamental error in reasoning. An hypothetical Turing machine could in principle simulate a brain, but it does not work the other way around. That is not equivalence.
When someone talks about separating the hardware and software level, they are making exactly the kind of mistake I am arguing against. There is a sense in which the physical function of the brain can be described in terms of software, or symbolic processing or whatever. That is just a model though, an artificial abstraction. It also hides what is happening according to modern neuroscience.
The brain does not process software. In order to describe it in such terms you have to drastically modify the definitions of what is being discussed. The brain has a deep architecture, and has evolved to process the environment in very specific ways. During development some of the brain architecture is modified by the environment, both internal and external, and this operates in two phases. First it determines some of the connectivity in the brain, and then it determines what connectivity is lost during early childhood. The brain adapts in part by losing connections that are not used. Learning and adapting over a lifetime could be considered a third phase, though in practice this occurs all the time from even before birth.
In modern neuroscience its generally agreed (currently, but things change) that most brain processing is automatic and not conscious. It does not use rules.
I might describe a tree in terms of fractal rules or other rules I can simulate on a computer. That doesn't make my model equivalent to a tree though. Its a model of an abstraction derived from studying a tree. That simulation would have data used by software. That data would be physically or representationally equivalent to aspects of the tree physical structure and architecture. That doesn't make the tree a Turing machine, or mean the tree processes data using software or software equivalents.
As soon as people start thinking of the brain as a computer, running software, they start making mistakes. The model is wrong.
