Frankenstein

For all those I would ever care to show this one, already know what triggered it.

I was thoroughly bored on a Sunday afternoon, so I was going through some of the older and the most interesting chats I have ever had with people, I came across this one with a friend of mine, who I consider has quite an intelligent imagination and an appetite for abstractness.

So he got metaphysical and asked me what did I think was “thought”? And I answered – A thought can be described as a function in the neural/neuronal space. ( I love being a theoretician :D).

The thought stuck in my mind the whole day that resulted in the following hypothesis.

Firstly why a thought should be a function in the neural/neuronal space. Since when you “think” about something, its not merely a “thought”, it is a combination of certain visual, audio (namely all sensory) memories of a certain event and your reactions to them, which can be quantified as certain sets of neurons firing at the same time or in succession to each other over some short period of time. So imagine if every neuron represent one dimension, for ‘n’ neurons involved for a certain kind of thought, a thought can be described as a trajectory in n-dimensional neuronal space.

There can always be a function approximation for any given trajectory, using machine learning methods like neural networks. (Are we approaching an infinite loop here ?)

Anyways, since the dimensions we would be dealing here would be huge, we can first use dimension reduction techniques like PCA (Principal Component Analysis ), and extract the most relevant features ( in simple words, the neurons that really matter and contribute to the “thought”). The trajectory can be approximated by two methods.

One, you can use a neural network, where you don’t actually get the function describing the trajectory, but get a model, which would predict the output, given an input i.e you can find out what kind of “thought” would one have if an another set of neurons are triggered. But for some reason, the idea doesn’t seem very alluring to me, one reason might be that sometime in the past, attempts were made to model the brain using neural network, hence it would be like using method ‘A’ to model ‘A’. ( That was what I meant when I said infinite loop ), the rhetorical skepticism that we would as human beings would never able to understand a human brain, because it would be limited by the fact that a brain is trying to understand itself and would result in never ending loop of reasoning and how something greater than human brain is required to do the job.

Second, one can use something called Kernel Methods widely used in Machine Learning these days. What you try and do is find a feature space where the trajectory is a hyperplane ( like a straight line in 2-dimension, plane in 3 dimension etc ). The mapping is done by something called a kernel function. One can have polynomial, guassian, hyperbolic tan as kernel functions. Hence the mapping can also be done to the feature space with infinite dimensions ( for example in guassian kernel , u use the exponent ‘e’ function which is an infinite expansion of polynomials ).

Again it is easier said than done. The main problem is how do you choose the kernel function ? How do you know in what feature space would the trajectory be a hyperplane? Unfortunately there is no correct answer for it. But there are techniques like kernel boosting, where one starts with a set of what are called ‘weak kernels’, and try to find the accurate kernel as a linear combination of these kernels.

On an optimistic note, you would have an approximate function describing a particular kind of thought ( isn’t that a treat for mathematicians and cognitive scientists ?? )

This might be just a beginning. Once the functions for all “thoughts” are known, they can be combined in “some” way to define a personality or the person himself.

Imagine an equation describing yourself, whatever you u could ever think, talk or do.

Did we just create a Frankenstein ?