Author Topic: When do you think A.I. will be able to break the Donkey Kong world record?  (Read 6428 times)

0 Members and 1 Guest are viewing this topic.

Offline Zmick_tricK

  • Full Member
  • *
  • Posts: 54
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #15 on: February 10, 2018, 11:50:50 AM »
After rethinking the question, I came back here to clarify and correct some of what I said- only to find that KongTower already beat me to the punch! You are, of course, correct that I was missing the key component to what AI actually is, and that would be the I, or, intelligence. I was conceptualizing the technology in the same manner as a brute force method of solving a board or logic game- using software to run simulations of every possibility, discarding failed results and keeping sucessful results, to arrive at a clean decision tree that leads to perfect play. I was imagining, basically, a program that writes its own TAS for the game,  which is not at all what we're talking about here, even if that would be interesting in itself.

Maybe a better analogy for AI gaming, and maybe KongTower can help verify or correct me here, would be if Johnny 5 took to playing Donkey Kong and kept playing and playing until he became better than all humans. The difference being that instead of a physical robot, Johny 5's play is generated by code that directly inputs controls into the game instead of using a joystick and buttons.
DKF Blogger

Offline KongTower

  • Jr. Member
  • *
  • Posts: 35
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #16 on: February 10, 2018, 03:15:08 PM »
After rethinking the question, I came back here to clarify and correct some of what I said- only to find that KongTower already beat me to the punch! You are, of course, correct that I was missing the key component to what AI actually is, and that would be the I, or, intelligence. I was conceptualizing the technology in the same manner as a brute force method of solving a board or logic game- using software to run simulations of every possibility, discarding failed results and keeping sucessful results, to arrive at a clean decision tree that leads to perfect play. I was imagining, basically, a program that writes its own TAS for the game,  which is not at all what we're talking about here, even if that would be interesting in itself.

Maybe a better analogy for AI gaming, and maybe KongTower can help verify or correct me here, would be if Johnny 5 took to playing Donkey Kong and kept playing and playing until he became better than all humans. The difference being that instead of a physical robot, Johny 5's play is generated by code that directly inputs controls into the game instead of using a joystick and buttons.

Close.  The ML approach would absolutely positively want all the .inp files from MARP and a way to generate video from them as "labelled input" for the "training phase" of the process.

Playing is the "inference stage."  MINIMAX and trees of all possible things that could happen is the old way.  The new way is training with massive amounts of image bandwidth fed to a computer brain that is trained with these images.

It's actually more like how they created a girl in Weird Science than the traditional concept of what most people trained in AI from 1980's through mid-2000's had learned.  It's only a couple years that this has been known as the breakthrough.

We do a lot of research into this at my work.  Another buddy has used Machine Learning to earn the #1 and #2 spot in the top-grossing iPhone apps.  He was my game tester at Atari that I hired three times, twice at Atari, and later at a motion-based coin-op game company.

His machine learning maximizes the ROI for advertising for his two iPhone apps.  I think they were making 80M per month each a year ago, and I'd expect more by now.

ML has taken the world by storm already.  It was very unanticipated four years ago that'd we be here this fast, even, much less the crazy visions of what's coming in just a few more years.

Offline serphintizer

  • Senior Member
  • *
  • Posts: 132
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #17 on: February 11, 2018, 12:33:24 PM »
 <confused>
"Kill screen? Ha! It's for tourists."  - Mr. X
DK 1 Million Point Scorer Twitch TV  Streamer DK 1.1 Million Point Scorer DK Killscreener

Offline TheSunshineFund

  • Spring Jumper
  • *
  • Posts: 1109
  • Why not play some Kangaroo today?
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #18 on: February 11, 2018, 01:43:36 PM »


It's actually more like how they created a girl in Weird Science than the traditional concept of what most people trained in AI from 1980's through mid-2000's had learned.  It's only a couple years that this has been known as the breakthrough.


Wait....we can do this?
And as the cattlerack was cleared
The floor was brightly stained
And larger though we grew in size
Not a thing was gained
The fate as so ordained
Twitch TV  Streamer DK Killscreener DK Jr. Killscreener DKF Blogger Crazy Kong Killscreener

Offline QAOP Spaceman

  • Elite Member
  • *
  • Posts: 295
jammvyy: sory today i am rambunkshess

jammvyy: shis

jammvyy: rambakshis
Twitch TV  Streamer Crazy Kong Killscreener

Offline KongTower

  • Jr. Member
  • *
  • Posts: 35
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #20 on: February 12, 2018, 11:54:15 AM »
I honestly think that rubber hitting the road helps people the most.  From what I have understood about the game of Donkey Kong is that under emulation, the random number generation process is compromised.  On the real hardware, it has been reported to me that randomness is based on electromagnetic noise.  An intelligence that has perfect timing but cannot predict the EMF in the environment cannot make incremental improvements on a process with an anticipated result.

In any case, MAME emulation of Donkey Kong is a trivial matter for a traditional AI to explore even without machine learning.

When I first learned about MAME and MARP, it was pretty obvious that cheating was simple to do.  If you playback one input file while recording another, the new input file you recorded would take all of the samples from the input file that was being played back and you can easily append one file to another.  By using the F10 key, you can turn throttling off and advance to where the game is going to be looking at your inputs more quickly to continue creating the input file you want to represent a high score.

I found this out on my own, as I'm sure many others did as well, while recording MARP scores with various incantations of MAME needed to do the best recording for each game.

The cheaters guide to improving skill also includes save-and-restore state.  You can easily get to a stage that you are having troubles with and save your state and be able to play that game in emulation and restore immediately to that state from the command line to practice your troubling area.  I had brought this idea up to Steve Wiebe as a means to practicing the harder stages.  At the time, most of the point pressing going on was on the barrel level as I expect that it is today.

But, personally, I don't think that MAME being used to build up motor memory on more challenging stages is anything to frown on.  And the reason why is that nobody can ever prove where you obtained your skill.  If you restore-state to the same point in the game that normally you would have to play for 2 or more hours to get to would take, you can build up skill in that area so many times faster by restoring state.  If you don't do it your enemy will, and as long as there is no way to prove that you have done this in order to train, people are going to do it, so you might as well go ahead and do it yourself.

Steve never showed any interest at all in MAME, and I don't blame him.  So many reasons to just focus on the rules being put forth to him by Twin Galaxies and not to worry about anything that might be interpreted as cheating given the history of cheating in competitive arcade game playing.

But today, I'd love to hear a roll call from anyone willing to having had practiced harder stages in Donkey Kong by using MAME as a tool to build up motor-memory for the game.

Offline WCopeland

  • Spring Jumper
  • *
  • Posts: 587
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #21 on: February 12, 2018, 12:16:05 PM »
But today, I'd love to hear a roll call from anyone willing to having had practiced harder stages in Donkey Kong by using MAME as a tool to build up motor-memory for the game.

Literally 99% of active members of this forum.
Twitch TV  Streamer DK Killscreener DK 1 Million Point Scorer DK Online Open #4 Winner DK 1.1 Million Point Scorer Crazy Kong Killscreener DKF Blogger Pauline WR Holder DK 1st Man Million DK 1st man 1.1 Million DK Jr. Killscreener 1 Million Point Trifecta DK 1.2 Million Point Scorer

Offline KongTower

  • Jr. Member
  • *
  • Posts: 35
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #22 on: February 12, 2018, 01:58:08 PM »
But today, I'd love to hear a roll call from anyone willing to having had practiced harder stages in Donkey Kong by using MAME as a tool to build up motor-memory for the game.

Literally 99% of active members of this forum.

What percentage would it be if Billy Mitchell and Steve Wiebe were members of this forum?

Offline f_symbols

  • Spring Jumper
  • *
  • Posts: 802
  • wht u mean
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #23 on: February 12, 2018, 07:25:55 PM »
I'm not saying this to sound high and mighty, but to the best of my knowledge I am the only active player who has never played DK in mame.  I always assumed that I was in a boat with only  <Wiebe> and  <Billy> in this regard (and maybe  <Kuh> ), and that is the reason I chose the "long road" so to speak. It wouldn't surprise me to learn that all 3 of them have played on mame (specifically to hone their skills on certain screens), I'm sure there were plenty of 3-500K players back in the 80's and early 90's that took this route, if only by necessity. 

To answer your question, If you counted  <Wiebe> and  <Billy> in the percentages, it would still be over 99% as Wes stated.  Learning without mame is dumb and takes infinitely longer, from my experience.  <YSG>
DK Remix Killscreener Twitch TV  Streamer DK Killscreener DKF Blogger DK 1 Million Point Scorer

Offline Zmick_tricK

  • Full Member
  • *
  • Posts: 54
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #24 on: February 12, 2018, 07:53:19 PM »
I've heard the argument that one shouldn't use emulator functions to practice and, similarly, I've heard that practicing Connect Four in perfect-play software is "cheating". (if the latter sounds out of place, allow me to add that C4 has been a fixture in my life and, when I'm de-rusted, am an expert player in the game). If it's simply a preference, like with f_symbols, that's one's prerogative.

To actually insist or demand that others do the same is just ROFL To me, it's like saying you can only practice kicking field goals during a real game or that you can only practice running during a race.

Thankfully, at least 99% of people here would agree with me and the whole argument is moot  <Allen><Sanders>
DKF Blogger

Offline KongTower

  • Jr. Member
  • *
  • Posts: 35
Re: When do you think A.I. will be able to break the Donkey Kong world record?
« Reply #25 on: February 12, 2018, 08:52:30 PM »
Thanks for the responses, guys.  I've never used any methods to cheat or to practice higher levels myself and 100% of my skills were accumulated by age 14 at the latest.

The main reason I wanted to record my 300,000 score was to challenge myself.  It was a surprise to me that I set a 300,000 score while playing with Wiebe, and I was wanting to see if I could do it again.

I've almost broke 200,000 in the last couple of days, but I have done nothing to cheat or practice at all.  But I did break patterns to attempt pushing points and failed immediately each time.

I made about four noteworthy attempts at scores, and I've gotten 186k first try, 123k second try, 88k third try, 122k third try.  Those are scores typical of my childhood play.  The 200+ scores are relatively uncommon for me, but I do think I could swing a wolfmame 300+ and I'd like to update my score before anyone questions if I have cheated.  Steve Wiebe is my only witness that I've ever even scored over 300,000 in real life besides an arcade owner when I was a teenager who is probably not even alive any longer.

I've only played a few hours (less than 100 anyway) of Donkey Kong since the 80's, including the few games with Steve and on MAME subsequent to that.