Friday, May 15, 2009

Of modelling Human beings and pseudo random number generators

While have a chat with my friend Abhishek over my previous post on irrational behavior of humans, I started thinking on how and if humans can be represented by random numbers (Afterall each one us is unique and essentially random).

I believe (as do a lot of evolutionary scientists) that our reaction to any stimulus depends on a lot of factors, like the genes we carry, the natural environment we have lived in, the kind of thoughts we have, our experiences, stimuli from the recent past etc. Evidently, there are far too many variables that can affect our decisions.

So is there a way to this apparent madness ? Can we call these effects totally random? Abhishek sure seems to think so. But I doubt it. If everyone had totally random characteristics, then we would have had a zero (neutral) average reaction to any stimulus. But, on the contrary, masses behave more in unison. So the seemingly random "personality trait numbers" have a non-zero mean. But this mean can only be seen in a "large enough" group.

So can traits of human beings be thought of as being generated by a pseudo-random number generator? After all, these can have a mean, are random for all practical purposes but everyone knows that these have been derived from some determinstic method. I really dont know. But the idea itself seemed too good not to be posted

Stumble Upon Toolbar

4 comments:

Abhishek Kannur said...

I did not think of the averaging-out-to-zero point. This way, the problem becomes even more complex considering no one really knows the portion creating the majority (as you mentioned).

Man, this is so arbit!

Ankit Ashok said...

ya, it is very difficult to determine what is a large enough sample to average out the idiosyncrasies.

It indeed is complex !

RSP said...

i think, humans should be modeled as vectors in n dimension with each coefficient being a random number.

also i dont think averaging out goes to 0, it goes to a constant (prolly 1/2,if random nos are between 0 and 1)

any 2 junta have some common n diff properties so they cant be called random/independent too..

Ankit Ashok said...

Dear RSP

Please try to write in a more formal way, so that it remains intelligible.

My response is based on my guess as to what you intended to say.

Choice of the range of the random numbers used to model human behavior can be arbitrary - But it makes sense to choose values that lie on either ends of spectrum. (We often talk of human behavior as contrasting between complete opposites - kind vs brutal, docile vs rebel etc.)

So in the ideal case, we should look at a zero mean.

The problem arises when measures on these scales tilt towards either side - and we end up with a non-zero mean, or if you stick to your 0 to 1 convention, a mean not equal to half.

The independence of the values of a particular individual vis-a-vis another one comes from the assumption that the second person has no bearing on the first person's personality. Eg - I may have a risk taking value of lets say .9 (on a scale of -1 to 1) another person called Dan has the same value. But we haven't met/interacted ever before - in fact we don't even know of the other person's existence. Do you think that our values are dependent?