halyavin wrote:
Some top demos are incredibly good. Don't you think that this demos (at least partly) obtained with using "drod solver" - a computer program that can bruteforce move sequences and suggest optimal moves?

halyavin wrote:
Some top demos are incredibly good. Don't you think that this demos (at least partly) obtained with using "drod solver" - a computer program that can bruteforce move sequences and suggest optimal moves?

Umm.. that's a pretty serious accusation to make so lightly. Obviously that would be cheating. Could you give a few examples of demos you feel were obtained by a computer?

Anyway, a computer could only optimise short move sequences. Something like 15, maybe 20 moves. Humans can do that too, it only takes a little longer, and we often make mistakes because we don't really do an exhaustive search - our minds tend to eliminate "obvious" bad moves. Solving DROD rooms by computer is waayyyy without reach of current technology. I doubt there will ever be computers fast enough. It will certainly take much faster computers, plus some major breakthrough in AI theory.

Dali wrote:
I admit that some demos are nearly perfect and very fast. I remember Erik's demos in KDD in DROD: Architects' Edition and it was really fast.

halyavin wrote
I think it is impossible to search all variants after 10 moves in roach-fighting, but computer can search only good variants as chess programs do. You don't even need AI - the goodness of variant can be programmed specially for each room based on it idea (which you have to think out yourself). And core dual processors are not exotic already (I hope that I will have one eventually).

mrimer wrote:
Iirc, in the ending credits to either 1.5 or 1.6, Erik mentions that Matthew Daly made these demo recordings and that he sped them up a bit.

Pinnacle wrote:
If a room requires 100 moves (a really short one) the possibilities are 1.378 * 10^104.
If the solver can attempt 2.2 billion per second, it would take 1.986 * 10^87 years to solve it. (5.993 * 10^77 times the age of the universe)
So... no. Not at all.

Fafnir wrote:
One last point - if I wanted to cheat, which would be a rather pointless endeavour since most of my recent demos don't seem to want to upload, I would probably use some sort of modified keylogger as a "portable checkpoint". The idea would be to record all keystrokes, then play them back upon death - effectively reloading. Are there safeguards against this in the code? I think it should be possible to trap keypress messages without returning them to the OS, and thereby any keyloggers. [If there are no safeguards, mods feel free to delete this paragraph.]

Schik wrote:
If the top players have computers from far in the future, then maybe they have brute force solvers. If not, then I'd say they're just far sadder than I could ever hope to be.

Niccus wrote:
On the solver:

How will it tell if the room/level is impossible?
(this means: impossible to clear, leave, etc)

(this is a comment on feasibility, mind you)

EDIT: Oof, wrong question, sorry.

state_queue = get_all_starting_states()

while(state_queue isn't empty) {
  current_state = pop state_queue;
  foreach new_state (get_next_states(current_state)) {
     next if we've already found next_state;
     next if new_state is a death state;
     next if new_state is leaving an unconquered room
     if(new_state is leaving a conquered room) {
       print "The room is not impossible"
       end;
     }
     push new_state into state_queue;
  }
}

print "The room is impossible"

;)

TripleM wrote:
Could possibly let the same AI play the same room multiple times so it can 'learn'.

Obviously it would be only open to people who knew how to program in some language.

NiroZ wrote:

halyavin wrote
As for chess, it has taken people hundreds of years to figure out good strategies, which have then been programmed directly into a computer.

You are wrong, computer algorithms to play chess are not based on good "human" strategies, they are just brute force with some heuristic optimization.

The problem of solving a drod room is NP-complete, as already someone has stated, so i'm not sure if there are possible implementations to solve drod puzzles. It might be possible, I'm not interested in investigating that though
Even if it is possible to solve drod rooms it would most certainly be much slower than a human player to do it.

I'm thinking about 1 thing that no one has mentioned here though, what if you use the computer to play for you when you "tell" the computer the exact perfect moves to solve a specific puzzle you already know the answer.
You can take all time you want to figure out the "perfect" solution and then just make a script to do it, wouldn't that be easy, and much faster than a human player?

schep wrote:
I've thought of this sort of DROD AI a few times before, never seriously. But usually as something interactive: >SOLVE ROOM. Don't know how. Type STRAT to list goal tree. >DROP RED DOOR. Don't know how to reach (5,35) past goblins...

Kemosabe-TBC wrote (with bad quote-tags):
computer algorithms to play chess are not based on good "human" strategies, they are just brute force with some heuristic optimization.

Kemosabe-TBC wrote:

I'm thinking about 1 thing that no one has mentioned here though, what if you use the computer to play for you when you "tell" the computer the exact perfect moves to solve a specific puzzle you already know the answer.
You can take all time you want to figure out the "perfect" solution and then just make a script to do it, wouldn't that be easy, and much faster than a human player?

×
First Delver! (I was the first non-tester/dev
to conquer TCB.)
d/dy

zex20913 wrote:
Somebody once did something like this (he skyrocketed from ~20 to ~7 in two weeks). We determined that he was feeding the movesets from #1 demos into his computer to obtain many many first place ties in very very little time. (Compare 1:00 to :07 for completion times.)

schep wrote:

zex20913 wrote:
Somebody once did something like this (he skyrocketed from ~20 to ~7 in two weeks). We determined that he was feeding the movesets from #1 demos into his computer to obtain many many first place ties in very very little time. (Compare 1:00 to :07 for completion times.)

Of course, demo duration has absolutely nothing to do with demo ranks and scoring, right? Copying other people's demos in order to tie them was the immoral cheating part. The impossibly short times on the demos was just another clue that something was foul.

Not that I'm describing how to get away with this sort of thing to anyone. If somebody copied a bunch of #1 demos move-for-move with random reasonable delays inserted, they'd still eventually be noticed and caught, just for having so many #1s, all identical to other people's.

×
First Delver! (I was the first non-tester/dev
to conquer TCB.)
d/dy

schep wrote:

TripleM wrote:
Could possibly let the same AI play the same room multiple times so it can 'learn'.

But what AI isn't going to try out possibilities in its own memory before 'submitting' them, whatever that means? Doesn't seem there's much point to multiple runs.

Getting such a thing capable of solving even a few simple rooms would be a tall order. The thought of preparing a computer to battle three roaches puts me off from planning any further.

Kemosabe-TBC wrote:

NiroZ wrote:

halyavin wrote I wrote
As for chess, it has taken people hundreds of years to figure out good strategies, which have then been programmed directly into a computer.

You are wrong, computer algorithms to play chess are not based on good "human" strategies, they are just brute force with some heuristic optimization.

Kemosabe-TBC wrote:
I'm thinking about 1 thing that no one has mentioned here though, what if you use the computer to play for you when you "tell" the computer the exact perfect moves to solve a specific puzzle you already know the answer.
You can take all time you want to figure out the "perfect" solution and then just make a script to do it, wouldn't that be easy, and much faster than a human player?

×Well if I'm going to get modded down for this, I might as well go

Tahnan wrote:
But I (usually) have enough fun with trapdoors--as opposed to orb puzzles--that I'm not sure I'd want to.