Idle Speculations

Username:	Password:
Forgot your password?	Remember Me

ridiculousness
I think hyperbolic discounting is pretty awesome.

Could possible use that as part of a Sims-like AI? If I understand correctly, the Sims AI has a very simple find-best-action core. The core accepts "advertisements" of possible actions from objects, describing why someone might want to, say, sit on the couch (it's comfortable). In the actual sims game, the language of the advertisments come in the form of quantities of various needs (comfort, social, food, bladder).

The difference here would be that the "currencies" would be happiness at various delays. Then the sims could choose between actions like eat ice cream (happiness now, unhappiness later) and throw out all icecream (unhappiness now, happiness later), and evaluate them according to their own discounting schedule.

I doodled a little bit of code related to this idea, but I didn't get very far.

( source code goes behind cuts )

Monday
I take the bus to work, which involves walking to one bus stop, waiting, and riding the (slow) bus, and then walking from the bus stop to work.

On mondays, it is very difficult to convince myself to leave promptly. I left a full 45 minutes after I had planned to leave, and spent the entire hour-plus transportation episode castigating myself for insufficient willpower, and telling myself that my tardiness will get me fired.

However, when I arrive, I learn that in fact it is "only" 8:30, and I am the first to arrive in my (10-programmer) group. Which feels good.

I'm not sure I understand Brad De Long, but what I don't understand appeals to me: http://delong.typepad.com/sdj/2009/04/hoisted-from-the-archives-a-non-socratic-dialogue-on-social-welfare-functions.html

My new favorite noun phrase: priceless antiquarian thinking engine

Challenge rationality
a
I think sometimes (at least on TV) doctors do tests which are "challenges", stressing one part of the patient's system selectively. If you're interested in techniques for enhancing human rationality, you might challenge your individuals with some really blunt manipulations, like getting them very drunk, or requiring them to count backwards by sevens at the same time. Then your experimental group is the one that gets both the challenge and your technique-in-question, and your control group gets only the challenge.

The idea is that differences in degrees of success might be easier to perceive if the "lab rats" are considerably stupider than the experimenters setting up and evaluating the scores, but that the techniques for succeeding while artificially stupid carry over into success without actually requiring the challenge. At the very least, it increases safety - sometimes people are stupid for one reason or another, and having a safety net would improve the process.

I like finding real words that sound fake. For today: psychomotor. Of course, the fake word would be a noun, and the real word is an adjective.

Can we use it in a sentence? Yes, yes we can: "Features of caffeine intoxication include [...] periods of inexhaustibility and psychomotor agitation."

laptop sociability
There is a trend, in my apartment, towards people sitting "together", everyone with laptops on their laps, reading/writing/playing the internet.

Until recently, I did not take part, because I did not have a laptop, merely a palmtop. Due to some annoying misconfiguration between the palmtop and my apartment's router, it was useless for browsing the internet at home. (which is very nearly useless in general).

Now I have a laptop, and can partake in the communal internet-drinking. It is indeed mildly pleasant.

I watched "Mean Girls". Initially miscategorized it as a "Lindsay Lohan movie"; it's not. It's a Tina Fey movie, and it's pretty good.

lesswrong is pretty fun.

StickK
Economists have the notion of a "commitment mechanism". StickK lowers the barriers to individuals utilizing financial commitment mechanisms. In order to explain this, I need to divert a bit. Economists analyze "games", which is only somewhat related to the usual notion of a game. An extensive-form game is something like a tree describing possible futures.

A malicious genius is considering whether to dose the dashing protagonist with a toxin. The toxin is known to be invariably fatal unless counteracted, and the malicious genius has the only antidote. The antagonist knows that the protagonist will face a choice: Either open a specific locked box containing, among other things, the antidote - surviving, but furthering the antagonist's wicked plan, or refuse to open the box, dying, and foiling the plan.

We analyze this as an extensive form game: The antagonist has a choice to dose or not to dose. If dose, then protagonist gets a choice, to die or not to die.

If only the protagonist was not so very very rational! Because the protagonist is known to be very very rational, the antagonist knows that the protagonist will choose to live, and thereby further the antagonist's plan.

A commitment mechanism, then, is the protagonist (rationally) sabotaging their rationality before the antagonist has an opportunity to dose. The "irrational revenge circuit" will revenge harm at any cost, even an irrationally high cost. Even antagonists, step carefully around people with revenge circuits installed. (Note: Yes, evolution has already installed some of these.)

StickK allows you, in one brief burst of action, to do something like sabotaging your rationality. It isn't rational to revenge at the cost of your life, or to send your money to an anti-charity, but it might be rational to risk sending your money to an anti-charity.

Entities that you identify with
For example, your future self. Is this entity easy to pick out of a crowd?

Are you your tomorrow self after going to sleep tonight?

If you say "yes", then suppose that there was some discovery that you are in fact not conscious in a specific part of your sleep cycle. Your consciousness has a gap in it! Is this a problem?

Are you your self after surgery on your heart? The doctors will require stopping your heart to do the surgery (it's that kind of surgery). You might, in fact, be "clinically dead". Your life has a gap in it! Is this a problem?

Are you your self after many of the atoms in your body have been replaced via normal living? What if there was an "atom-flushing" treatment that somehow replaced all of the atoms in your body with new atoms? Your substance is vanishing! Is this a problem?

Are you your self after a stroke that causes you to lose knowledge? What if it caused you to lose emotions? What if you seemed like a completely different person afterward?

Are you your children? Are you your clone, grown from your DNA?

Suppose a genome-editing virus transforms you into (technically) a different species: mutually infertile with unedited humans but otherwise identical. Are you the same person?

Are you your life's work, a body of text that might be reread by students for centuries? What if we videotaped every moment of your life? From every angle?

Are you your entire future light cone?

( I think the moral of these little questions is clear. )

I'm excited about possibly using max-sat algorithms to attack subgraph isomorphism and compare the results to the (ancient, discredited) structure mapping analogy-finding algorithms.

Commentary on sf/f books, photoshopped onto the cover of the books in question, replacing the title.

Some of the ingredients of humor are truth, insult and exclusivity. The entire endeavor is exclusive in that you need to have read the books in order to understand the comments about the books. Further, you need to be familiar enough with the books to recognize them based only on the cover image, not the title. The insult is clear enough. The truth in these examples varies, but the best ones are the most truthful.

I don't know where I got this idea (Steven Pinker? Marvin Minsky?) but someone proposed that laughter might have evolved as a technique for several lower-status individuals to get the timing correct in a coordinated attack on a higher-status individual. The three qualities above certainly fit the theory. Laughter is contagious, which fits. If a high-status individual is laughed at, they are lowered in status, which fits. Laughing together at the same things is a bonding experience, which fits.

Laughing also lowers the threshold for people to laugh again. I don't know whether that fits. It's possible that there's a generic priming mechanism going on. Maybe if you didn't get the timing right the first time, lowering the threshold will cause the contagion to spread faster, and improve the timing.

Laughing is also somewhat involuntary. I'm not sure how this fits. Is it that the timing has to be so tight that you don't have time to involve your deliberative brain? Or is game theoretic signalling involved here?

I had this idea a while ago: brute force search guided by simple type theory, filtered by tests.

The internet makes me cry. But in a good way.

To investigate later: Obtiva and 8th Light

There was a contest, a while ago, for "General Game Playing". The programs compete to read a game specification and then play that game.

I wonder whether this would be an effective programming technique. Write the task that you want the computer to accomplish as a game against Nature, and then hand the specification to a general game player.

I made a little system dynamics model of a "shishi-odoshi". I think this is a good model for the relaxation oscillators used in BEAM robotics.

It's not a good model by any measure, but it was fun to make.

There are three stocks:

The amount of water in the cup
The position of the cup
The velocity of the cup

The amount of water in the cup is increased by a constant slow inflow. If position is below a threshold, it is decreased by a fast outflow. The position is increased by the velocity. The velocity is affected by five forces: Gravity (a constant), Friction (correlated with velocity and in the opposite direction), Water weight (proportional to the amount of water in the cup), and two "saturation" forces, preventing the cup from going beyond the upper or lower limits.

I'm not sure how to model the "saturation" forces well. My current version uses a double exponential once the position is beyond the limit, but it's still a bit bouncy.

Post Modern Programming is funny, in an interesting, oblique truth kind of way. I certainly don't subscribe to postmodernism, and I'm not sure the authors do either. This example is pretty awesome:

"The task is to instruct a computer to print a table of the first thousand prime
numbers, 2 being considered the first prime number.

To write this program, we first connected our computer to the Internet,
downloaded some music from Napster, and then read our email. (You have
to receive email to perform a workday [11]). We received 25 pieces of email of
which 16 were advertisements for Internet pornography, administrivia, or invi-
tations to invest in Nigerian currency trades. After dealing with this email, we
typed “calculate prime numbers” into Google. This found several web sites
regarding prime numbers, and some more pornography. After a while, we
were interrupted, and so moved on to the prime number web sites. In particu-
lar, http://www.2357.a-tu.net includes the “ALGOMATH” C library for
calculating prime numbers; another site included an EXCEL macro which was
to complex to understand. Although we had not programmed in C for years,
after downloading and compiling the library (by typing make), we noticed the
documentation included the following program:


        int *pointer , c=0;
        if((pointer = am_primes_array(4, 3)) == NULL)
              printf("not enough memory\n");
        while( *(pointer+c)){
                                   printf("%d\n",*(pointer+c));
                   c++;
            }
        return;

We cut and pasted this program into a file and compiled it several times,
having to add a few extra lines (e.g. main () {). Eventually we ran it, and in-
deed it appeared to generate three prime numbers larger than four. We edited
the parameters to am_primes_array to (2,1000), and then ran the output
through wc -l to check that it had printed 1000 numbers."

Hum. So, I studied a little python (it's very nice), and a little javascript (I don't really understand how to do objects yet).

Exciting things: Google has a C++ styleguide, and a lintish style checker. The v8 javascript engine is fun to install and play with.

Seriously, to install and play with it, I did these commands:

svn checkout http://v8.googlecode.com/svn/trunk/ v8
cd v8
ls
sudo apt-get install scons
scons sample=shell
sudo install shell /usr/local/bin/v8

Looking at the C++ code for the v8 shell (v8/samples/shell.cc) was fun. It's pretty easy to extend it with a new primitive, not knowing anything, just working by analogy.

I think it might be fun to have a game where the standard console is a scripting console, like python, lua, or the javascript console in your browser. The actions that your character takes would be primitives, like "go_north(); look()". The game would incorporate the possibility that people write scripts while playing the game, like: "i = 0; while( i < 3 ) { go_north(); ++i; }". Maybe SVG together with javascript?