[gurps] Failing autopilots and RVO

[David Scheidt]

On Mon, Mar 2, 2009 at 3:23 PM, Zan Lynx <zlynx at acm.org> wrote:
> Anthony Jackson wrote:
>
>> The other factor here is that computers are only predictable if you've had
>> considerable experience with them. It's not like you're going to be
>> publishing the source code for your autopilots, nor is it like you won't be
>> updating the logic of your autopilots as people work out ways of beating
>> them. Incidentally, this applies to humans as well, you do need to update
>> your tactics manuals every so often.
>
> Yes, with regard to human tactics training, that is very true.  That is why
> US army and air forces would train against special training groups that used
> the Soviet Union's tactics manuals, for example.
>
> But humans can reach beyond their training when they need to, or sometimes
> they make a mistake which later turns out to have really been the right
> thing to do.  That's less likely for a computer running a program.
>

It's less likely for humans than you think.  Current research shows
the reaction takes place *before* mentation about what that rection
will be.  In other words, you do something (jink left in your fighter
plane) and then you think "I should jink left".

> I am thinking that a computer which rewrites its own programs based on
> experience or predictions of opponents future actions is no longer just
> running a program; at that point it is at least an LAI (in my opinion).  And
> that is a whole new thing than a Skill 17 piloting program.

It's state of the art AI.  For 1975.  Pretty routine today, used in
all sorts of stuff.

-- 
David Scheidt
dmscheidt at gmail.com