Gyroid Blog

In my post Instatiating a Digital Human I claimed it was necessary for causality to connect the states of a computer simulating a digital human for her to experience consciousness.  I claimed that a computer generating random states that just happened to mimic a computer simulating the digital human, would not generate an experienced consciousness.  Here a present a counter-argument.

Firstly we assume that our algorithm that simulates the digital human is reasonable resilient to errors.  Imagine that after each propagation step we introduce a small amount of noise into the simulation (i.e. we replace a small number of the bits by random bits).  Since our algorithm is resilient our digital human doesn’t die but just becomes very slightly dopier.  As we introduce more and more randomness she will gradually become more senile until eventually she ceases to be conscious.

Now imagine this same situation, where the random bits we introduce, just by chance happen to be the same as the bits they are replacing.  In the limit where we replace all the bits, this is simply the example mentioned earlier of a randomly generated set of states that happen to mimic a digital human, thus we proscribe no consciousness.  However, we can go in a continuous manner from this entirely random digital human, to the entirely deterministic digital human, simply by reducing the number of bits that are randomised.  To the entirely deterministic digital human we proscribe consciousness.  What then do we proscribe to the simulations that lie in between?  Does some lesser form of consciousness exist?  It makes me very suspicious of my earlier argument.

In a previous post “Quantifying Existence” I posed the dilemma of a digital human named Jim, who had to make a decision. To make this decision he needed to know whether he valued each simulation of himself equally or whether he valued each unique simulation of himself equally (i.e. does a second simulation that is identical to one already running add any value). Rephrased the question was whether some conscious experiences could be made to exist more by simulating them on several computers. Possible answers are:
1- Each unique consiousness should be valued equally
2- Each instantiation of a consciousness should be valued equally
3- Something else

A similar question of quantifying our existence comes about if one interprets quantum mechanics in a many worlds manner. For example, if we perform the double slit experiment with a single photon, the result is a superposition of different worlds in which the photon was detected in different positions. Each of these worlds is described by a wavevector and has an amplitude associated with it. The square of this amplitude is the probability that we will find ourselves in that world. That is, all worlds are not equal, some are more probable than others. It seems reasonable to me, to say that some worlds exist more than others. From this it follows that some instantiations of me (those in more strongly existing worlds) exist more than other instantiations of me.

This contradicts answer number 1 above, since we value some consciousnesses (the more probable ones) more than others. Answer 1 is incompatible with a many worlds interpretation of quantum mechanics. Thus although I find answer 1 far more intuitive, I am inclined to go with answer number 2 appropriately adjusted so that we have:
Each instantiation of a consciousness should be valued according to the probability of it occuring.

This does lead us to some funny conclusions though. The most notable is that all other things being equal, a digital human should try to get as many copies of himself running as possible. Imagine a digital human embodied in a robot body. The sensory input could be piped to a billion identical copies of the digital human running in computers around the world, and the output from the digital humans sent back to the robot. The digital human would not notice that he was running on a billions computer but would nevertheless exist much more than he would otherwise.

Currently, I am biting this bullet, but am not happy about it. Some consolation is that such a strategy would probably not survive in a competitive environment since it would use resources without providing any competitive benefit (added existence being of no advantage to winning competition for resources).

Consider the following  dilemma of a digital human named Jim:

Tomorrow 10 copies of him will be made.  8 of these will be run for one week in a virtual Fiji environment, 1 will be run in a virtual Aspen ski resort environment, and 1 will be run in a virtual Swiss Alps ski resort environment.  Since the 8 copies in Fiji are in an identical environment and the simulation is deterministic the states of the computers will be identical throughout the simulations.  The two ski resort simulations will naturally be different.  Jim has the choice, imposed upon him by experimental virtual psychology graduate students of the future, to take either a virtual scuba diving kit, or a virtual skiing kit with him.  He will have this kit with him in all 10 simulations.  If he thought there was a 50% chance of going to Fiji versus going to a ski resort he would value the two kits equally.  This is Jim’s last holiday.  After one week of virtual time in the virtual environments all his copies will be terminated.  Which kit should Jim take with him?

Assume that it is possible to create a digital simulation of a human who experiences “what it’s like to be a human”. The digital human at any point in time can be described by the internal state of the computer along with the rules that generate each progressive state from the previous one. The question I am interested in here is “When does the ‘what it’s like to be a human’ occur?”. Approaches to this question are:

1 - A Platonic position to simulation, asserts that all possible simulations exist independently of physical instantiation. This is argued by the roboticist, Hans Moravec, in his 1998 essay “Simulation, Consciousness, Experience” and is more or less espoused by Greg Egan in his 1995 science fiction novel “Permutation City” with his “Dust” theory of reality.

2 - Pete Mandik, suggests in a blog post I read the other day, that we could store all of these internal states on a large disk, and that that would instantiate the consciousness since the temporal relations have simply been replaced by non-temporal ones.

3 - Peter in his “On Philosophy” blog considers a computer that generates random internal states and by chance produces a series that describe a digital human. He argues that the digital human would experience consciousness.

My objection to 2 and 3 is that the causality linking the different digital human states has been removed. The “what it’s like to be a human” is the generation of each state from the previous one, rather than the collection of states itself. Correlations are typically an indication of causality but do not imply it.

I find 1 harder to argue against, since I’m inclined to assume that all possible worlds exist, which would naturally include all simulations. If all possible “what it’s like to be a human” experiences exist independently then it is difficult to talk about when a physical system instantiates them. We can get around this by claiming that instantiating the digital human in our physical world makes her exist more. This added existence occurs when we generate each state from the previous one.

good list of commands here
http://www.tcl.tk/man/tcl8.3/TclCmd/contents.htm

Problem: a CNAME redirection was not working

Solution: Changing DNS zone file from

blog 10800 IN CNAME gyroid.wordpress.com

to

blog 10800 IN CNAME gyroid.wordpress.com.

i.e. I was missing a full stop/period

Notes on BASH shell scripting

cat - read content of file
/dev/null - file that eats any stream sent to it - produces an empty file if used as input i.e.

good introduction at
http://www.linuxien.com/docs/bash/shellscript.html

This is a list of shell commands that are useful in Linux but that I use infrequently enough to forget. So I’m writing them down here so I can find them easily. I´ll add to the list as I have to look things up.

whereis
If you’re trying to find out where a tool that you use is, i.e. for some reason you want to know where gnuplot is, then you type

 whereis gnuplot

which will get a response like

 gnuplot: /usr/bin/gnuplot /usr/share/gnuplot /usr/share/man/man1/gnuplot.1.gz

find
Find a file with a given name e.g.

 find Espresso*/samples/ -name "*diatoms*"

gives

Espresso_New2/samples/lj_diatoms.tcl

Espresso_New2/samples/lj_diatoms.tcl~

This post is me organizing some of my thoughts on my metaphysical beliefs.  Upon reading Wikipedia I’ve decided I’m a Metaphysical Naturalist.  This seems to be defined mostly by the statement that “mental properties … are causally derived from, and ontologically dependent on, systems of nonmental properties, powers, or things”.  I guess same would apply to anything else such as gods, spirits, etc.  They’re allowed to exist but they can be explained physically and so are deprived of their supernaturalness.

But this definition is more in telling us what naturalism is not than what it is.  We cannot simply equate naturalism with materialism, since physics is far more than just matter. Modern reductionist physics is more a collection of mathematical equations, and it seems strange to assign any of the concepts that emerge from these equations such as space, energy, matter, time or causality to a higher existential status than the others.

From this standpoint the Universe is the “solving” of a certain set of equations with constraints, such as an initial condition (I’ll ignore the randomness introduced by quantum mechanics here to keep things simpler). An important point is that the Universe is not one and the same as the equations and their constraints, but rather one and the same as the “solving” of them. Computer simulation provides a nice analogy to make this clearer. A description of all the equations of the some universe, along with the initial conditions would give us everything we needed to simulate the universe (aside from the really big computer), but the universe wouldn’t be conjured into existence until the simulation was started.

We can be a bit more general and talk about the minimum information necessary to describe the universe (to start a simulation of it), rather than a system of equations and constraints. What empirical evidence tells us is that their is a whole lot of pattern in the Universe (causality, matter, space…) and it seems that the minimum information necessary to describe it is quite small (a few equations and and initial conditions), however the solution of these equations brings a Universe of incredible complexity into being. Consciousness is then emergent from the process of “solving” the system.

That’s more or less my take on naturalism. The difficult questions are then how to define information, since it’s hard to think what this means without an interpreter. Could there even be a well defined answer to what the minimum information necessary to describe the Universe is? And then why our particular set of information has been “solved”?

This is a little demonstration that one can observe continuous economic growth but end up at the exact same place you started. It may, however, be entirely wrong! I’m not very sure about my definitions for economic growth and inflation. The main thing is I get to play with typing in LaTeX equations (nice list of symbols here).

Start off assuming we have two commodities in our economy A and B and that A serves as the currency. The state of our economy is described by the prices of the two commodities, and . Production/Consumption is and .

The size of our economy, is thus
1)

The growth rate of our economy, is
2)

The value of a basket of goods that describe typical consumption, is
3) (same as size of economy in this simple example)

Then inflation, , is defined (I think) as the effect of price changes on the value of the basket
4)

Real economic growth, , is then the measured growth minus inflation
5)

Imagine that A and B have similar production costs so that at economic equilibrium . If consumers decide they like B a bit more than they previously did then the price of B will slightly increase, the production of B will increase and production of A will decrease and then the price of B will settle back down to 1. If we assume total consumption is constant, , then real economic growth is
6)
which will be positive so we experience positive economic growth.

If then consumers decide they don’t like B so much after all, then the price of B will drop below 1, the production of B will decrease, production of A will increase, and then the price of B will rise back to 1. Again we experience a period of positive economic growth although we end up with the economy in exactly the same situation as when we started.

The real growth rate then is a measure of how quickly the economy is approaching what we ideally would like. However what we would like is not at all constant. For an absolute measure of how the economy is growing we would need to measure it by some fixed critieria that was independent of consumer desires, but then it’s not clear what good criteria would be. It would be interesting to know how much of typical economic growth comes from fluctuations in consumer preferences relative to more absolute measures of the economy.

I googled a bit and couldn’t find any one talking about this but I don’t know this field well enough even to get the search terms right. But it’s a nice equation practice post!