General > Biology

The Geek LoopHole

<< < (4/6) > >>

Numsgil:

--- Quote ---But ok I am in creative mood:

B is a nasty evil robot and starts to attack C
This takes several rounds (lets say 10) during which C (who is really quite nice) is most likely running away at a speed of about 50 and is not attacking B

Meanwhile A is off in search of food which it finds and begins feeding happily away on. Now B and C attack A as well, so A has a tough time.

B is now dead and C's population nearly doubled form killing all the B. (With the help and trouble of A at the same time.)
A is sorrunded by all the C right now. If this keeps up A has a chance because A is a little better then C. But the veg is reaching its reproduction point and its at its lowest. A is out numbered 1 to 5 now. Ah, finaly vegs reproduce. A is now 3 to 15. O no A and C did not get lucky. A = 0 and C = 12.

If and when B catches up with A to attack it, he will be pretty knackered and A will be nice and strong -In PY's dream land maybe so. The only way B will win is if A doesn't fight back. If this is the case then efficiency is irrelevent since B is a Canibot while A is not.
--- End quote ---
Maybe Bots is a computer AI trying to pass the Turing test.

If so, I'd say he's failing.

Botsareus:
Why you say that Num? I did not make it up...


Were is Endy and Welwordion, when you need them. The "proffesionals" solutions , or disbelive rather, sux.

Numsgil:
I just mean that every time you speak, it's like you're on the edge of making sense, but never quite make it.

Sort of like talking with ALICE.

So maybe you're a chat bot trying to pass the Turing test.

Botsareus:
Oh, well anyway , I beleave mutation somtimes produces situations when a robot won because it actualy was better not because it used the loophole to win. The real qustion is: How do you select this spesific robot.

In the current system:

Every second Find Best is called.  To the Best Robot its good factor is increased.

good = good + total population + total children,  every time find best is called.

The robot who has the most good ever during the whole 3 lest simulations is selected as the next step in evolution.

The problem is if this robot kills geeks, then he will cheat to be good.

We need a better way to mesure true skill.

Numsgil:

--- Quote ---We need a better way to mesure true skill.
--- End quote ---
Agree full heartedly.  I don't think number of offspring is a good indication of fitness.

Here's what I would do:

1.  Have the program periodically drop in One, or some other top bot, for like 200 cycles.  Then pull them all out and let the population recover.  There isn't a way to do this yet, but let's ignore that for the moment.

2.  If a bot ever manages to kill a One, give it a reward.  Maybe like 3000 body, or something like that.

3.  Slowly (like over millions of cycles) decrease the time between One's "attacks".

In fact, you could have the program select a random F1 bot to insert for that 200 cycles or whatever.  Reward for kills based on the position of the bot in F1.

Reward bots based on the percentage of the enemy killed, not the number, since Ymir is usually slow to reproduce but impossible to kill compared to Dominator Invincibalis.

Thus you'd be evolving bots to both survive and defeat other bots in F1.

Just be sure you aren't killing off too many bots between "attacks".  And for heaven sakes don't try this to evolve against a viral bot, because the viruses won't ever go away.

Navigation

[0] Message Index

[#] Next page

[*] Previous page