Doug Freyburger <dfreybur@yahoo.com> wrote:
>William Conrad Halliburton wrote:
>>What would be the hardest part about creating a bot to ascend?
>The game includes events from a random number generator so it can't be
>scripted in any simple sense.

The same is true of both Rogue and Angband, both of which have bots which
can win the game on a regular basis, so this is not the root of the
problem.

I'm not going to say this is the hardest part, but certainly one real
problem will be that NetHack involves difficult strategic decisions. It
doesn't matter how good the bot's tactics are (ie, how good it is at
exploring levels and dispatching monsters) if it can't make a coherent
decision about where it's going; unlike in Angband, sticking on level 1
for ages is actively counterproductive. Sokoban first, or the Mines, or
maybe just go for the minetown? Where to make caches? Shall I carry these
potentially useful scrolls, risking them being destroyed by fire, or cache
them? When is it right to blitz straight through levels, and when to
explore? Shall I let the pet fluff itself up? Is it worth taking the time
to rob this shop? Scum this altar? Shall I alchemise these potions now, or
wait until I have more?

It's worth taking the time to point out that time is a real issue in
NetHack's early game; a move wasted is, well, a move wasted. You're
constantly draining food, for one thing. So a bot cannot simply bimble
about in the way the Angband Borg does.

The next observation is that this does not apply to Rogue or Angband - if
it did, these problems would obviously be solved, and I'd be talking
rubbish. In Angband, time is not an issue. If you are safe from the
regular monsters at a given depth and can escape from out-of-depth
monsters, you can continue exploring that depth indefinitely, and your
character only becomes stronger by scoring XP, getting money, and perhaps
even finding decent items; the only question is when to descend in order
to accelerate the rate at which you gain goodies.

This is also Angband's only strategic decision; the depth at which to
operate. However, this is an easy decision for the Borg, because it has a
very high boredom threshold and simply always operates at a depth at which
it feels completely safe; it has full information on the damage that can
be done by potential attackers.

Angband also has an equipment optimisation game, but that is actually
easier for a computer that can quickly consider all possible combinations
of the existing armour, weapons and support items. Last of all, Angband's
tactical game is easier for a computer to approach, because essentially
all resources - ammo, potions, scrolls - can be readily replaced, so it
need not make decisions about whether the current situation merits using a
finite resource up.

In Rogue, there is only one strategic decision; whether to
explore each level fully or dive as soon as you have found the food and
the stairs, and I suspect the answer is always "dive". Time is an issue in
Rogue, but it's so overwhelming an issue that the opposite situation
applies as in Angband; Rog-o-matic need not make decisions about whether
to spend moves on anything other than finding the food and stairs,
dispatching monsters and analysing loot, because the answer is always
"no". There's nothing else to be done; so essentially all Rog-o-matic
need do is have tactical skill at dispatching monsters and exploring the
level.

I shouldn't take this as dimishing the achievements of the authors of
Rog-o-matic or the Angband Borg; both were clearly extremely difficult to
write, and the latter in particular is an amazing achievement. However, I
think one can see that a NH bot would be much harder still, and hence I
believe it is effectively impossible.