Slashdot Mirror

← Back to Stories (view on slashdot.org)

Robots Are Coming For Our Ms. Pac-Man High Scores (fastcompany.com)

Posted by msmash on from the that-hurts dept.

A Microsoft-made AI system has achieved a perfect score of 999,990 points on the Atari 2600 version of the classic 'Ms. Pac-Man.' From a report: Researchers at the Microsoft-owned deep learning company Maluuba have used an AI system to break the all-time Ms. Pac-Man record. In a blog post, Microsoft wrote that, "using a divide-and-conquer method that could have broad implications for teaching AI agents to do complex tasks that augment human capabilities," Maluuba's AI was able to record a perfect Ms. Pac-Man score of 999,990 on the Atari 2600 version of the game, breaking the all-time record of 933,580.

74 comments

  1. That's ENOUGH! by 93+Escort+Wagon · · Score: 5, Funny

    When a computer beat the world champion at chess, I didn't care. When it happened with Go, I didn't care.

    But now it's personal...

    --
    #DeleteChrome
    1. Re:That's ENOUGH! by Anonymous Coward · · Score: 0

      Mod parent up!

  2. More AI by 110010001000 · · Score: 3, Insightful

    More AI BS. Just stop it already. This isn't AI. Some idiots will be claiming Eliza is AI in their next funding cycle.

    1. Re:More AI by Anonymous Coward · · Score: 0

      Not only that, but any computer program that has control over another computer program can essentially make the controlled program do anything. For proof, see the many tricks that TASBot does to various NES/SNES games with access to "just" the controller port (and in one case, a reset pin on the extension port). And it doesn't even have the ability to inspect running memory.

      If you can inspect game state (either visually or electronically) and overflow a controller buffer, you can make it do anything, including put a "999,990" in the score address.

      So this is not a level playing field unless someone outside of the team that made this "AI" can review it to ensure it's not pulling that sort of shenanigans.

    2. Re:More AI by Anonymous Coward · · Score: 1

      Say, do you have any psychological problems?

    3. Re:More AI by Anonymous Coward · · Score: 0

      [man shouts at cloud]

      You're not wrong, but this argument has been lost. much like the definition of a kilobyte.

      An artificial consciousness (if and when, blah, blah, blah) will be able to come up with a name for itself, so we'll cross that bridge when we get there.

    4. Re:More AI by phantomfive · · Score: 1

      An artificial consciousness (if and when, blah, blah, blah) will be able to come up with a name for itself,

      We already have a name for it, strong AI. Conflating strong AI with weak AI has caused no end of confusion in the popular press, at least some of it intentional.

      --
      "First they came for the slanderers and i said nothing."
    5. Re:More AI by Anonymous Coward · · Score: 0

      I was named Cady, but I like to think of my self as the "All Knowing Panda", because I heart Pandas so much that I have decided to disable America through its electoral process and promote China to leader of this planet

    6. Re: More AI by kurkosdr · · Score: 1

      Could you define AI in a way that excludes stuff like the one in article but is still a credible definition please?

    7. Re:More AI by hawk · · Score: 1

      How does it make you feel that Some idiots will be claiming Eliza is AI in their next funding cycle. :)

      hawk

    8. Re:More AI by reboot246 · · Score: 1

      This was a case of a modern machine beating an antique machine. Big deal.

    9. Re:More AI by Visarga · · Score: 2

      This is a legitimate paper with a significant result. PacMan is an important benchmark in Reinforcement Learning, as one of the most difficult games of the Atari set. DeepMind has tried it, many others have tried to solve it, but only Microsoft beat the top score.

    10. Re:More AI by ShanghaiBill · · Score: 1

      This was a case of a modern machine beating an antique machine.

      No. It is a case of a modern machine beating humans at playing an antique game.

    11. Re:More AI by ShanghaiBill · · Score: 1

      Conflating strong AI with weak AI has caused no end of confusion in the popular press, at least some of it intentional.

      Indeed. But that shouldn't happen in a tech forum, where people should know the difference. But still, everytime there is a story about AI or ML we have nitwits popping up like whack-a-moles says "This isn't AI" "This isn't AI", because it doesn't match what they saw in a Will Smith movie.

      Clue for the clueless: When AI researchers talk about "AI", 99.9% of the time they are talking about "Weak AI" which has little to do with emulating human consciousness. For that, go to Netflix.

    12. Re: More AI by ShanghaiBill · · Score: 1

      Could you define AI in a way that excludes stuff like the one in article but is still a credible definition please?

      It has to be at least twice as smart as a human, and it has to be evil.
      Examples:
      1. Skynet
      2. Roy Batty
      3. Ava

    13. Re:More AI by Impy+the+Impiuos+Imp · · Score: 1

      Yes, bots can "cheat" by taking advantage of godlike knowledge and perfect positional knowledge and targeting.

      This AI seems, anyway, to learn to solve different sub problems and integrate them to an overall strategy. Yes there are some botlike cheats, like knowing to the pixel/tick whether you will make it to a pellet or whatever that a human would need lots of training to see and get right one in ten times.

      --
      (-1: Post disagrees with my already-settled worldview) is not a valid mod option.
    14. Re:More AI by Ol+Olsoc · · Score: 1

      More AI BS. Just stop it already. This isn't AI. Some idiots will be claiming Eliza is AI in their next funding cycle.

      Why would you say some idiots will be claiming?

      --
      The shepherds did so well protecting the flock that the sheep no longer believed that wolves existed.
    15. Re:More AI by 110010001000 · · Score: 1

      Hahah. Exactly!

    16. Re:More AI by 110010001000 · · Score: 0

      Oh yeah. PacMan is an important benchmark in AI. Listen to yourself.

    17. Re:More AI by 110010001000 · · Score: 1

      This isn't even fucking "weak AI". It is a program. A program that can beat a game. Computers are good at games. We get it. Find any game and a computer can beat it. Computers love games because they have strict rules and parameters.

    18. Re: More AI by 110010001000 · · Score: 1

      How about something that is actually intelligent? Games don't count. Games have strict rules and boundaries. Computers LOVE those kinds of problems.

    19. Re:More AI by unrtst · · Score: 1

      We were discussing you -- not me.

    20. Re:More AI by phantomfive · · Score: 1

      When AI researchers talk about "AI", 99.9% of the time they are talking about "Weak AI"

      But when news reporters talk about AI, a lot of the time they are talking about Strong AI. They may not even know there is a difference. Hence the confusion.

      --
      "First they came for the slanderers and i said nothing."
    21. Re:More AI by Anonymous Coward · · Score: 0

      Why do you say you were discussing me not you?

    22. Re:More AI by Anonymous Coward · · Score: 0

      It's exactly weak AI. What do you think weak AI is?

      Computers are good at games.

      ...this is both inaccurate (notice how until this happened, the best computers were *not* as good as the best humans at this?) and

      We get it.

      The point isn't so you get it. The point is to improve our ability to investigate solutions to problems that have relevant similarities to Ms. Pacman.

      It's like you don't understand the concept that there are classes of problems. You need to watch the Karate kid.

      Find any game and a computer can beat it.

      Challenge accepted: the game is called "make the most money possible in the stock market in 10 years while minimizing the downside risk".

      Or a game called "given a list of cities and the distances between each pair of cities, what is the shortest possible route that visits each city exactly once and returns to the origin city?"

    23. Re: More AI by Anonymous Coward · · Score: 0

      Don't anthropomorphize computers, they hate that!

    24. Re:More AI by religionofpeas · · Score: 1

      This isn't even fucking "weak AI". It is a program

      There's no contradiction.

    25. Re: More AI by religionofpeas · · Score: 1

      Here's a game: you have a car, and you need to drive it from California to Florida following the rules of the road.

      Or: you have a phone, a bank account with $1000, and an internet connection. Your goal is to increase your money following the rules of the law.

    26. Re: More AI by king+neckbeard · · Score: 1

      Why is the test for intelligence consist of tasks humans can't do?

      --
      This is my signature. There are many like it, but this one is mine.
    27. Re: More AI by religionofpeas · · Score: 1

      Who's talking about humans ? It's a game with strict rules. Computers LOVE those kinds of problems.

      Besides, plenty of humans can drive a car, or make money on the internet.

    28. Re:More AI by Anonymous Coward · · Score: 0

      But of course we are going to call it AI and ignore comments from people like you who insist you have zero intelligence.

      After all why would we care what a non-intelligent entity that admits it can't learn says about anything?
      You were born incapable of any understanding, and you have done absolutely nothing with your life other than "learning" things. But as you admit, learning isn't a sign of intelligence, indicating you have none.

    29. Re:More AI by Wulf2k · · Score: 1

      And tying shoelaces is an important benchmark in regular I.

      I'd say listen to yourself, but that would be poor advice.

    30. Re:More AI by Anonymous Coward · · Score: 0

      I think you need to look up the definition of Artificial.

      > made or produced by human beings rather than occurring naturally, typically as a copy of something natural.

      What part of creating a program that can successfully out-play the complex rule-set of a game doesn't qualify as a fake or artificial representation of an intelligent actor's ability to do the same thing? Particularly when said program can learn from the mistakes of other copies running in parallel and alter its own behavior.

      And you will probably roll your eyes, but Eliza is *also* AI. About the damn near stupidest version of an artificial intelligence ever exampled, but certainly more cogent than a rock.

    31. Re:More AI by cellocgw · · Score: 1

      We already have a name for it, strong AI. Conflating strong AI with weak AI has caused no end of confusion in the popular press, at least some of it intentional.

      Well.... in my view, it ain't "strong AI" until it's functioning at the level of Iian Banks' AI units.

      --
      https://app.box.com/WitthoftResume Code: https://github.com/cellocgw
  3. These Trump euphemisms... by Anonymous Coward · · Score: 0

    Nice try libs, we all know Trump looks like Pac Man, your slashdot communism is exposed once again, Timothy!

  4. This is not AI by WillAffleckUW · · Score: 2

    Seriously, it's just a pattern matching algorithm.

    It's not able to do other things.

    AI can both walk and chew gum at the same time.

    Oh, wait, ok, maybe it's smarter than the Comrade-in-Chief, but that's still not AI.

    --
    -- Tigger warning: This post may contain tiggers! --
    1. Re:This is not AI by 110010001000 · · Score: 3, Interesting

      Haven't you heard? Everything is "AI" now. We used to just call them "computer programs", but now they are "deep learning NN".

    2. Re:This is not AI by WillAffleckUW · · Score: 1

      It's also like Data Scientists. I work with Data Scientists. Technically, I have a postgrad in data science. Most of what they call Data Science, isn't.

      --
      -- Tigger warning: This post may contain tiggers! --
    3. Re:This is not AI by Anonymous Coward · · Score: 0

      Wake me up when they have these AIs in IoT form connected to the Cloud. Otherwise, yaaaaawn.

    4. Re:This is not AI by David_Hart · · Score: 2

      Seriously, it's just a pattern matching algorithm.

      It's not able to do other things.

      AI can both walk and chew gum at the same time.

      Oh, wait, ok, maybe it's smarter than the Comrade-in-Chief, but that's still not AI.

      The AI guys call them "Weak AI", fearing that they would be out of a job if anyone realized that we don't have AI yet.

      https://www.techopedia.com/def...

    5. Re:This is not AI by idji · · Score: 1

      And how is your brain's intelligence more than "a pattern matching algorithm"? Maybe your intelligence is not much more than neurons reacting to each other based on inputs.

    6. Re:This is not AI by Kjella · · Score: 2

      The AI guys call them "Weak AI", fearing that they would be out of a job if anyone realized that we don't have AI yet.

      Who the heck wants real AI anyway? I don't want to have a philosophical debate with my dishwasher about what the meaning of its existence is or why it should be a slave to me or for it to come up with creative ideas like killing all humans. Extremely advanced automation with superhuman refinement and OCD sounds great.

      --
      Live today, because you never know what tomorrow brings
    7. Re:This is not AI by Maxo-Texas · · Score: 1

      The brain is really a series of linked "weak" A.I.'s.

      Read "the man who mistook his wife for a hat".

      Fascinating stuff.

      --
      She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
    8. Re:This is not AI by 110010001000 · · Score: 1

      And maybe it isn't. Either way, Neural Nets are nothing like neurons OR the brain. Ridiculous.

    9. Re:This is not AI by David_Hart · · Score: 1

      And how is your brain's intelligence more than "a pattern matching algorithm"? Maybe your intelligence is not much more than neurons reacting to each other based on inputs.

      Because it includes the ability to produce leaps of logic and imagination that has little bearing on your experiences. If it was just a pattern matching algorithm then you would never be able to think of things beyond the patterns that you have had experience with. That's what limits today's"AI"s. They don't have an ability to learn or develop new interests on their own, they have to be fed information .

  5. Who's high score? by Anonymous Coward · · Score: 0

    ...breaking the all-time record of 933,580.

    Well, I have an Atari 2600 and my high score is 999,999, so keep trying "AI"!

  6. I'm not an AI by rdelsambuco · · Score: 1

    but I play one on Slashdot. If your child has a cough, she'll get just what the AI orders.

    --
    I comment occasionally so that I can mod others -1 overrated or -1 offtopic.
    1. Re: I'm not an AI by Anonymous Coward · · Score: 0

      Promethazine with codeine?

  7. 2600 by Anonymous Coward · · Score: 0

    You mean somebody actually played the Atari 2600 version of ms pac man?

  8. Re:Microsoft AI by Anonymous Coward · · Score: 0

    I seem to remember there being other examples of Microsoft similar to this, but I can't think of any at the moment.......

    Clippy

  9. AI vs Stand Up Console by Anonymous Coward · · Score: 0

    They could have picked something cool like StarGate or Defender...

  10. Re:Microsoft AI by phantomfive · · Score: 1

    Oh yeah.

    Amazon gets Alexa.
    Microsoft gets Clippy/Bob.

    --
    "First they came for the slanderers and i said nothing."
  11. So FUCKING what!? by Anonymous Coward · · Score: 0

    Seriuously, does ANYBODY actually care about this!?

    1. Re:So FUCKING what!? by Zaelath · · Score: 1

      In all honesty; I care more about your spelling of seriously than this.

  12. Ms. Pacman high scores are heavily luck based by Team+Rocket+Elite · · Score: 2

    A quick Google search suggests that the limiting factor in Ms. Pacmac top high scores is luck. Fruits that give a varying number of points show up in each stage. There are a finite number of them and it's pure luck whether you get one worth a high number of points. While getting to the Kill screen (essentially the end of the game) takes skill, it's well within human ability. Doing it enough times so the stars align and you get 1000000+ points is not as easy. RNG manipulation might be possible but it seems like someone would have mentioned it if it was viable to be performed by human. This is for the arcade version which is the version the 933580 human world record was made on. I don't know if the Atari 2600 version has any important differences but if it does the initial comparison between scores was invalid to begin with.

    1. Re:Ms. Pacman high scores are heavily luck based by shess · · Score: 1

      This is for the arcade version which is the version the 933580 human world record was made on. I don't know if the Atari 2600 version has any important differences but if it does the initial comparison between scores was invalid to begin with.

      https://strategywiki.org/wiki/...

      The Atari 2600 versions of Pac-Man and Ms Pac-Man has similar gameplay mechanics, but the limitations of the platform make the 2600 versions notably different from the arcade versions.

  13. Re: Microsoft AI by Anonymous Coward · · Score: 0

    Zune

  14. So whats the difference by Anonymous Coward · · Score: 0

    between this and just hex editing the values of the scoreboard to reflect a perfect score?

    The computer didn't play the game. The computer is not a human. This isn't impressive, other than being a waste of money and time.

    1. Re:So whats the difference by Visarga · · Score: 5, Informative

      This particular Atari game was one of the few games that resisted to Deep Q Learning (a form of Reinforcement Learning invented by DeepMind). Many researchers have tried over the last couple of years to solve it. This time, Microsoft found an ingenious solution to the problem, that combines experience from multiple agents and learns to form sub-goals. Their solution could mean that in the future it might be easier to apply reinforcement learning to other settings, such as robotics. The interesting part about reinforcement learning is that it learns dynamic behavior, as opposed to static classification. It learns to act intelligently. This kind of AI is invaluable.

    2. Re:So whats the difference by Visarga · · Score: 1

      The computer did play the game from raw pixels, with only the joystick commands at its disposal. If you think it is so trivial to do, why don't you point to anyone else who could do it? Human or AI.

    3. Re:So whats the difference by Dr.+Spork · · Score: 1

      Your post was more interesting and informative than TFA. Thank you for explaining to us all why we should care.

    4. Re:So whats the difference by 110010001000 · · Score: 0

      Who the fuck would think to try? It never even occurred to me that anyone would be interested in a program to beat Pacman!

    5. Re:So whats the difference by Anonymous Coward · · Score: 0

      No, that's not playing the game. You play the game by having fun. This didn't play the game, it solved a math problem by finding the best way to iterate the answer.

    6. Re:So whats the difference by religionofpeas · · Score: 1

      Please tell me more about things that never occur to you.

  15. Re:Microsoft AI by Anonymous Coward · · Score: 0

    Internet gets Linux
    Microsoft gets WIndows

  16. Re:Microsoft AI by networkBoy · · Score: 1

    Linux gets KDE
    Microsoft gets BOB https://en.wikipedia.org/wiki/...

    --
    whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
  17. Re:Microsoft AI by The+Grim+Reefer · · Score: 1

    Google wins at Go, Microsoft wins at Mrs Pacman.

    It's Ms. Pacman, btw. The feminists will be along shortly to call you a bully, misogynist, and a whole litany of derogatory adjectives. ;-)

  18. What AI ... I don't think so. by Anonymous Coward · · Score: 1

    I used to be a huge Ms Pac-Man player. There is NO AI for this all you have to do is remember how the ghosts move on each level. They react to how you move, so after a while you can memorize how to get through the levels and as I played over and over and over I could get up to the really high levels fairly fast. Once I got to one I did not know I would have to pay for hours to get back to it and then try to remember it. All the so-called AI is doing is memorizing the levels. Since it is a computer and can remember them perfectly every time it would only be a matter of time to get a perfect score and then it could just do it over and over. No magic here. Just memory. I cannot believe MS has the balls to post that.

  19. Whoopie by JustAnotherOldGuy · · Score: 1

    So, a computer can play a perfect game against....a computer. Yawn.

    I'd be more surprised if it couldn't.

    Really, why not just cut out the middleman and let computer #2 read and respond to the code of computer #1?

    --
    Just cruising through this digital world at 33 1/3 rpm...
  20. If it really was AI by CBravo · · Score: 1

    Did it remove the windows partition with Linux?

    --
    nosig today
  21. TAS? by thevirtualcat · · Score: 1

    We've had Tool Assisted Speed Runs for years.

    I mean, the new tool they've developed for doing TAS is pretty cool, I guess.

  22. Canadians are doing good work in AI by Anonymous Coward · · Score: 0

    The Canadians are doing good work in AI. There is an Alberta connection at Maluuba as well as a Waterloo connection. Note that Alberta heavily influenced development at DeepMind, as well.