News:

PD.Com: Pretention in a can.

Main Menu

George

Started by Doktor Howl, May 31, 2018, 06:27:56 PM

Previous topic - Next topic

Doktor Howl

Quote from: chaotic neutral observer on August 19, 2018, 06:29:42 PM
I like where this is going, and am looking forward to future installments.

Quote from: Doktor Howl on August 18, 2018, 09:57:21 PM
I fire the explosive pins in the car's frame, trigger the air bags, and turn the radio on.
Turning the radio on is a nice touch.  The reason for it is obvious in hindsight, but the incongruity made my brain hit a speedbump.


However, you did trigger my OCD:
Quote
The first scenario leaves Daniel with a 90% chance of surviving, but leaves the other driver with only a 20% chance of survival, for best-case outcome odds of  .27.  The second scenario gives both drivers a 60% chance of survival, with best-case outcome odds of .36.
0.6*0.6 = 0.36,  0.9*0.2 = 0.18.  Did you have the other driver with a 30% chance of survival in an earlier draft?  Am I making a mistake in assuming that the appearance of statistical independence in scenario 2 implies the same in scenario 1?  Or am I just another asshole obsessed with irrelevant minutiae?

I'm also not sure if an AI would use the probability of both parties surviving as the optimization goal, as opposed to the expected number of survivors (0.9+0.2 = 1.1 for scenario 1, 0.6+0.6 = 1.2 for scenario 2).

Neither target is really satisfactory, though.  For example:
Scenario 3: {driver 1: 100%}, {driver 2: 0%}
Scenario 4: {driver 1: 90%}, {driver 2: 10%}

If we maximize the probability of both surviving as the goal, then we would pick scenario 4 (.09) over scenario 3 (0.0).  If we maximize the expected number of survivors, we wouldn't know which to pick (the answer is 1.0 for both).

However, scenario 3 is arguably better, because one person is guaranteed to live, while in scenario 4, there is a 0.09 chance of both dying.

I'm gonna think about this for a bit.

You are correct, that was a typo.  Thanks.
Molon Lube

Doktor Howl

Corrected.

Also, yes, this particular AI would strive to keep the most people alive, rather than accept the highest odds of one person surviving.  Also, the events do not dictate a zero sum game.  For example, in no case is there a 100% chance of one survival and a 0% chance of the other person's survival, so yes, there is a statistical independence.
Molon Lube

Doktor Howl

This is still the intro chapter, mind you.

I have only a few minutes at a time to work on this.
Molon Lube

minuspace

 More please. On that last thought experiment, the designation of fractional survival rates kinda jangles my nerves. I suppose the logical extension of Scenario Three being correct, which it seems, is to not let humans drive in the first place. That's going to be interesting.

Doktor Howl

Quote from: LuciferX on August 21, 2018, 01:08:07 AM
More please. On that last thought experiment, the designation of fractional survival rates kinda jangles my nerves. I suppose the logical extension of Scenario Three being correct, which it seems, is to not let humans drive in the first place. That's going to be interesting.

Or let them think they're driving.
Molon Lube

minuspace

Quote from: Doktor Howl on August 21, 2018, 01:48:06 AM
Quote from: LuciferX on August 21, 2018, 01:08:07 AM
More please. On that last thought experiment, the designation of fractional survival rates kinda jangles my nerves. I suppose the logical extension of Scenario Three being correct, which it seems, is to not let humans drive in the first place. That's going to be interesting.

Or let them think they're driving.
:lulz: :lulz: :lulz:

LMNO

That's cool, thanks Dok.

Doktor Howl

Molon Lube