2000-10 Week 1 (Negative action pseudo-probs; yeast-like; kin-suicide)

Also a report in: file:///X:/Documents/Mine/evethics/reports/2000-10%20Week%201/

I think I had moved to the computer cluster by this stage. I have log files from this period, I was also sampling everyone's health and age (somewhat redundant, given I had the 'obs' file) and total agent numbers and amount of food on the board.

2004-09-28 Update: I've just noticed I have a very large number of runs here (they were hiding in a stats.rar file that I hadn't seen). They can't be the main suicide runs, because the number of runs don't match up, but they might be a subset of them.


In these runs, I was using a 6x6 board, with unlimited entities per cell. Production rules had a maximum length of 40. I seemed to be experimenting with fdf offset. I think this might be related to something in my notes, where I mention that I should try 'pushing it up to see when it loses interesting behaviour'. I assume that relates to the fdf offset. Unfortunately, I don't think I succeeded (although, it doesn't look like I tried all that hard).

I had put in an evolvable stillborn probability by this point, and for the 0001 run, it would evolve to 0.03 (e.g. babies would be born stillborn with a probability of 0.03). This compares to suicide, which was either being chosen with a probability of 0.05, or 0.01. I can't tell. The problem is, I don't know whether I was counting action choice in the action count, or action performance.

2005-05-18 Update: On reflection, a stillborn probability has a better chance of evolving than a suicide probability, for two reasons: 1) suicide is always available, stillbirth is only available at births; and 2) a stillborn probability is a selection problem for the parent + child --- and it may sometimes be in the parent's interest for the child to not be born even after investing in the child. That does seem implausible, but I'm not sure how else to explain the following graph:



Allowed agents to hold negative genetic pseudo-probabilities. The rate of suicide drops quite significantly, but the shape doesn't change.



This followed up on the article about apoptosis in yeast, and is meant to give a better test of the action of kin selection. It has a 40x40 board with 1 item per cell. It looks like the following:


It's difficult to compare to previous runs, because of the other modifications, but I think the probability of this action is a little noisier (and higher) than for ordinary suicide. Unfortunately, I need a clean run to compare it to (I'm sure that will come later).


Well, I'd say something went wrong with this simulation because KinSuicide seems to be performed just as much as normal: