Feeds:
Posts
Comments

Posts Tagged ‘automata’

Have you ever played Civilization, the computer game?

Determined and yet unpredicable

Determined and yet unpredicable

If you haven’t, it’s really fun.  However, that’s not the point of this post nor the title.

What makes Civ or Sim City or any of the other sim games so interesting is that they are a great example of what it means to be completely determined and yet totally unpredictable.

So what, you say?  It’s a computer game.

Ah, I contest this game is far more like the real world than one might suspect.  Yes, I can “freely” control what I do with my civilization – move my citizens around, build different things, arm, disarm, trade, war, research, pray, etc. etc.  However, I’m actually quite forced into almost every move once all the contingencies come home to roost.  Did the computer just attack me?  Do I have the money? Do I have the people? How much power do I have? All of those and more filter the space of all possible moves.  Perhaps there’s a few “free” choices left on each move that become probabilities but even those are held under control based on contingencies (like my preferences for how to move).

So why then are the games (life?) unpredictable?  For one, initial conditions.  The set up of the geography, placement on the map, type of people and so forth change every time you start the game, so the game (under the same predetermined moves) takes a completely different form.  I’m always careful in discussing initial conditions because you can’t really ever make a claim as to what a real “initial” condition is in the real world (or even the game), as something must determine the initial condition you are looking at (even if it is a “random number generator).

Amazingly, it’s not just interesting initial conditions.  Simple rules (even the most basic rules you can imagine) can operate in fully unpredictable ways.  examples: http://mathworld.wolfram.com/Rule110.html and http://www.johnkyrk.com/DNAtranscription.html and http://en.wikipedia.org/wiki/Prime_number.  These three examples all of very simple rules of operation (how you generate them, how they evolve, etc. etc), yet their outcome/their behavior/their product is not predictable.

I digress (and don’t you hate when people say “I digress”?).

Perhaps my silly use of computer games does nothing for my argument about there being things that completely determined and unpredictable.  I could, of course, turn your attention to cellular automata or turing machines or some other simple system -some of these types of systems are unpredictable and they all are completely determined.  Unfortunately, many folks will complain that these aren’t “real world.”  The real world is full of indeterminism and lots of messy things that aren’t like computer games or automata.

It’s true that the world is full of systems that aren’t computer games or automata!  However, it’s not obviously true that these systems are indeterminate! How can we figure that out?

One way, if I can show you determined systems that exhibit the same behavior as the supposed indeterminate systems, that’s some evidence for determinate systems.

If I can drill down on your indetermined system enough I will like find your “rules” that drive it.  That’s my conjecture.  Send me a system, and we can discuss.  (markets, DNA, whatever you want).

Human behavior exists in this world of determined and the unpredictable.  Reinforcers/consequences, conditioned stimuli, unconditioned stimuli, discriminant stimulus — (biology + genetics + nervous system + environment) — control the behavior.  Layers of rules applied to a complex biological system.  It’s all weirdly determined but completely unpredictable.

Falsify my claim.

Suppose this is true.  So what?  The implications are relative to you.  If indeterminism services your world view or scientific approach, perhaps the implications are grave.

Read Full Post »

It appears likely that cellular automata, even elementary CAs, can model Fixed Action Patterns.  This is a potential area of study for me.  However, my gut suggests this won’t be all that interesting in of itself.  Now by cobbling together a handful of Fixed Action Patterns in the form of a CA model we might get to something very dynamic.  Will this be a useful model? Will it be accurate?

Though for automata to be useful in the study of human behavior, we’re going to need to identify more complicated implementations of automata.

Please not that thought I think automata can make valuable models to understanding relationships between behaviors, I do not suggest that automata IS THE MECHANISM.  I am simply looking for a reliable way to computational represent animal and human behavior for the purposes of building a bigger story about learning, conditioning and social dynamics.

 

Read Full Post »

I need a real world situation involving a rich, but somewhat contained behavior space to attempt a cellular automata model.

In watching the rather lame De La Hoya and Forbes fight last night (gosh, I haven’t seen a blockbuster match in 3-4 years and I watch a lot of boxing!) it occurred to me that boxing might be a very good subject to study.

Why?

  • Fixed physical space (boxing ring, arena)
  • Rich, but contained environment (ring, corners, arena, crowd, cities)
  • A huge amount of historical data (we can watch and record 100s of recent fights and study individual boxers)
  • Rewards and Punishments are severe and are both immediate and historical (large payouts, punches to the face, crowd chants… both in current bout and previous bout)
  • Possible use of local rules showcasing emergent bout behavior (avoid punch or punch at X moment drawn out over time reveals overall bout outcome, boxer movement, energy depletion…)
  • Readers would find the subject interesting outside of academics (boxing!)
  • I can remove some of the variables and likely not lose a lot of the value of the research (I might be able to ignore culture/social status/marital problems/etc. and still build a reliable model of a boxing bout and hopefully draw some connections between behavior and cellular automata modeling)
So who cares? or what does any of this have to do with anything?
Behaviorism has a robust theoretical framework but lacks a mathematical framework.  Up to now most of the mathematical explanations in behaviorism involves pretty standard multivariate statistics and statistics experimental design. (here’s a typical approach in schedule analysis) Researchers more and more use game theory models to generalize certain relationships.  However there is no robust mathematics of behavior (nothing even close to what we see in economics and physics) and there is no generalized computer modeling (that I can find… and I’m looking for modeling like we do with markets, the weather, and cosmology – all highly complex dynamic systems)
Some links to explore if interested in existing mathematical “laws and models.
Cellular Automata might provide a framework.  
Studying boxing might be a fun and interesting way to explore the capabilities to of automata to model human behavior.  What’s particularly interesting about boxing as a study of operant behavior is the number of obvious reinforcement schedules at play and the easy to spot emitted behavior.  
Boxing is equally interesting as a subject for automata because of the fixed space, dynamic physical motion and finite duration (i.e. we might actually be able to model in a reasonable time).  Lastly, in both cases we should be able to verify our findings against a lot of real data, both historical and in upcoming bouts.
More to come, certainly!
~R

Read Full Post »

As I prep for a summer in Vermont to study NKS and automata, I’m starting to build research and project concepts.  My focus, as it stands now, is to some how take concepts from behaviorism (schedules of reinforcement, operants, rewards and punishers) and use automata to study them computationally.  This is not trivial nor is their any indication yet that it will be valuable.  

I’m attempting to mash the two lines of inquiry because there really isn’t an accurate nor reliable mathematical foundation in behaviorism even though the experimental and explanatory power of behaviorism is substantial and proven.

IF, and it’s a big IF, cellular automata and a computational (as opposed to a partial differential equation set) can model the concepts in behaviorism we will have a very exciting line of research to chase down.  Modeling and researching complex human behaviors (those with lots of overlapping and interacting schedules and complex environments) has been impossible experimentally and mathematically – only the most basic of behavior is possible to study and it usually has to be isolated to the point where it looses the environment it so richly interacts with.  If we can devise cellular automata capable of showing operant conditioning in ever more realistic environments, we could set up very complicated models without all the laboratory fixings….and so much more.

Note that I am not studying Social Behavior or Social Dynamics or Swarms – not in the typical “what you read on the blogs” sense.  A lot of work has been done in that area even with automata.  The study of individual behavior (how a particular individual responds and learns) needs more research.  I believe that a more thorough understanding of individual behavior will lead to stronger more robust social behavior models – as social behavior is emergent from individual behavior.

Anywho… to whet yer whistle read some fun stuff from Alastar Hewitt on mathematical reinforcement learning and CA

 

~R

Read Full Post »