An AI created by a Microsoft-owned machine/deep learning team has completely conquered Ms. Pac-Man, achieving a perfect score:
At long last, the perfect score for arcade classic Ms. Pac-Man has been achieved, though not by a human. Maluuba — a deep learning team acquired by Microsoft in January — has created an AI system that's learned how to reach the game's maximum point value of 999,900 on Atari 2600, using a unique combination of reinforcement learning with a divide-and-conquer method.
AI researchers have a documented penchant for using video games to test machine learning; they better mimic real-world chaos in a controlled environment versus more static games like chess. In 2015, Google's DeepMind AI was able to learn how to master 49 Atari games using reinforcement learning, which provides positive or negative feedback each time the AI attempts to solve a problem.
Though AI has conquered a wealth of retro games, Ms. Pac-Man has remained elusive for years, due to the game's intentional lack of predictability. Turns out it's a toughie for humans as well. Many have tried to reach Ms. Pac-Man's top score, only coming as close as 266,330 on the Atari 2600 version. The game's elusive 999,900 number though, has so far only been achieved by mortals via cheats.
(Score: 3, Insightful) by AthanasiusKircher on Thursday June 15 2017, @02:58PM
Hmm... well, a quick search tells me that the world record for arcade Ms. Pac-Man is over 900,000. I don't know how competitive this stuff is for people who specifically want to beat the Atari version. And I don't know how the implementation may differ.
Also in this case, is it really an "AI" achievement compared to humans, or an agility achievement? I'm assuming the AI didn't have to compete by physically manipulating a joystick or controller? If humans could control the Atari interface with their minds, would they be able to get a perfect score too?
(Score: 0) by Anonymous Coward on Thursday June 15 2017, @03:00PM (8 children)
Surely all of the game's "randomness" comes from nothing much more sophisticated than an LCG. It should therefore be possible to determine the internal state of the generator by observing a relatively small number of monster movements and/or fruit spawns, thus enabling an AI to accurately predict all future movements.
(Score: 4, Insightful) by KGIII on Thursday June 15 2017, @03:19PM (7 children)
Yup.
Random is one of my least favorite and most favorite subjects. At best, as I'm unfamiliar with the specific architecture, this game has a fairly trivial PRNG.
Why do I like random, as a topic? It boils down to a philosophical argument. Throughout history, we've thought of many things as being random. We would attribute those things to "gods" or "fate." Yet, as we learned more, we discovered they're not really random - but actually able to be understood and predicted.
We currently believe that the decay of an radioactive atom is random, yet that even falls within certain boundaries. Is it truly random, or is it just still poorly understood?
That said, given the level of tech that was available for this game, and in that era, I'm pretty sure it's absolutely not true random.
"So long and thanks for all the fish."
(Score: 2) by mhajicek on Thursday June 15 2017, @06:36PM (4 children)
That's how I feel about quantum physics.
The spacelike surfaces of time foliations can have a cusp at the surface of discontinuity. - P. Hajicek
(Score: 1, Redundant) by KGIII on Thursday June 15 2017, @07:46PM (3 children)
Quantum Physics has a whole lot of repeatability - but is not yet fully understood.
I guess, how to say this? Hmm... I'm not yet ready to pop the cork on a bottle of bubbly until we've a unified theory that has been demonstrated to be repeatably sound. Yes, yes I do realize that's a pretty high expectation, and may never be possible - and unlikely in my lifetime.
It's interesting that many consider Einstein's biggest mistake to be, "God does not play dice." Yet, here I am, largely saying the same thing - albeit with a bit of a modification. Meh... If I'm wrong, I am in good company. And, no... I'm not saying that random doesn't exist. I'm saying that I remain a bit skeptical. I have a pretty decent grasp on the science involved, but my opinion is certainly not an authority. By the way, his use of God was metaphorical.
"So long and thanks for all the fish."
(Score: 1, Offtopic) by FatPhil on Thursday June 15 2017, @08:36PM (1 child)
"I believe in Spinoza’s God, who reveals himself in the harmony of all that exists, not in a God who concerns himself with the fate and the doings of mankind." -- Einstein
Some would place that as a type of pantheism, though Einstein himself shied away from that term.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 1, Offtopic) by KGIII on Thursday June 15 2017, @09:30PM
Yeah, I couldn't really think of a good way to describe it. If I had to speculate, I'd probably say he was probably agnostic, at least in determination of a traditional deity.
I'm not sure if the reference really speaks to a higher power, or to a higher order? The latter not needing a deity, nor implying a deity or purpose.
"So long and thanks for all the fish."
(Score: 0) by Anonymous Coward on Friday June 16 2017, @01:31PM
Well, you should have popped that bubbly a very long time ago. Our theories are enormously repeatable and sound, and there are no unification issues between QCD and gravity until you get to the Planck scale. With QCD, the Standard Model and all those particles fall out by setting, what, one or two free parameters? Your "pretty high expectation" is really "nothing short of perfection". Not even mathematics can have that level of self-consistency.
(Score: 0) by Anonymous Coward on Thursday June 15 2017, @06:59PM (1 child)
side comment: quantum mechanics gives truly random numbers.
Bell's inequalities being broken is experimental evidence of that.
(Score: 2) by KGIII on Thursday June 15 2017, @07:39PM
We think... Again, I'd suggest that it becomes a philosophical question, at some point. I'm familiar with Bell's theorem. I believe the best way to sum that up is that local deterministic and local random variables can not reproduce what is predicted by QM.
It's important that I point out that I'm a mathematician and not a physicist.
What I'd suggest is that, and again this is pure philosophy at this point, QM is not totally understood - can we agree on that? If we can agree on that, I'd speculate that it remains possible that we simply don't have enough data to make predictions accurately. Again, I point to the vast history we have of belief in random events when they were truly just not understood.
To be clear, I'm not saying that I disbelieve that it is random. I am saying that I have a niggling doubt, even though QM appears to be well verified. I'd further suspect that that's partially due to my own biases, as my experience in academia largely centered around mathematics. I will even go so far as to say that I'm absolutely unqualified to say they're wrong.
There's a lot that remains to be understood and quantified. That's really the only defense I have for my lingering doubts.
"So long and thanks for all the fish."
(Score: -1, Offtopic) by Anonymous Coward on Thursday June 15 2017, @03:05PM
Dog Shit Eater
(Score: 3, Interesting) by looorg on Thursday June 15 2017, @03:17PM (15 children)
Have it repeated it over and over again or was this some kind of one-time fluke? One would think the score could be different depending on which score-fruit drops and such. Did it die anytime during the run? Also how come it always seems to be the Atari 2600 versions of games? Are they some how easier or?
(Score: 3, Interesting) by AthanasiusKircher on Thursday June 15 2017, @04:30PM (14 children)
I just assumed it was the availability of convenient emulators, which means you can basically run the games on any modern OS. The "gold standard" for gaming records for old arcade games seems to be... well, actual arcade versions, no? (I frankly no very little about this world, but that's the sense I have.) But then you're dealing with custom hardware issues potentially, etc. Much easier to get your AI to just interact with one standard emulator.
Which actually brings up a legitimate issue (which I already hinted above) -- most old arcade games depended on agility with controls as necessary for success. How adept you were at moving joysticks around and hitting buttons really was probably more important than your "intelligence" or related strategic abilities. I want to see an "AI" that not only learns the sequence of events necessary to beat a game -- but can actually physically figure it out how to manipulate the controls to do so too. I know that seems like a bit of a different kind of task, but in this case it's a relevant issue concerning how the game is "beaten." (And there are plenty of tasks in the world humans might like AI to help with that will require physical manipulation of things in the real world -- and learning how to handle that in a real-world situation -- even though most AI applications so far seem centered on internal computation, etc.)
(Score: 4, Interesting) by KGIII on Thursday June 15 2017, @04:46PM (10 children)
To add to this, Mrs. Pac-Man has been used for AI testing in the past. Specifically, the 2600 version. (I think any carts for newer systems were just simple ports. I am not an authoritative source, however.)
This topic popped up years ago and I'm unable to find the source. If my memory is correct, there's a difference in the behavior of the ghosts and more random behavior. In the arcade version, the ghosts would head to their corners and would leave the box one by one. In the 2600 version, they'd all leave the box at the same time and not head immediately to their corners. The ghosts also all have different behaviors, meaning Blinky behaves differently than Pinky.
Which kinda touches back into my post further up the thread. There are different values for random. Some things are less random than others, but still considered random. (Not to be confused with true random.) There is more randomness in the 2600 version than there is in the arcade version - meaning that the ghosts are less predictable. Instead of going to their respective corners, they'll behave differently and this is inconsistent between game plays - though still not true random. (Not even close to true random, I *do* have a whole rant about the subject of randomness but you don't want to hear it.) IIRC, in the arcade version, they'll head to their respective corners and then converge on the player - whilst the 2600 variant would have them behave much more aggressively.
This makes it more difficult for an "AI" to beat. I'm thinking that I should be using the word "learn," as in, "machine learning."
Anyhow, MS isn't the first to tackle this with AI. Curiously, the fruit is also displayed "randomly." Each fruit has a different point value. I've not read/seen any published papers on this specific game.
Again, I'd like to make it clear that I am not even remotely close to an authoritative source on this - and I smoke a lot of pot. My memory may be fuzzier than that the yogurt in your fridge.
"So long and thanks for all the fish."
(Score: 2) by looorg on Thursday June 15 2017, @05:13PM (3 children)
Have you considered a career in the Secret Service?
(Score: 2, Interesting) by KGIII on Thursday June 15 2017, @05:34PM (2 children)
Yeah, but only in passing. I'm too old, now. I served as a Marine as a way to pay for my education, then I went back in and saved money up so that I could get my Ph.D. 'cause the GI Bill didn't (doesn't still?) even give you a stipend beyond the 4 year mark.
"So long and thanks for all the fish."
(Score: 2) by looorg on Thursday June 15 2017, @11:39PM (1 child)
Damn. I wasn't expecting a serious answer. I was just trying to make a funny in combo with the other news, a few stories down from this one, about how the Secret Service is going to relax about regarding stoners applying for a job.
(Score: 3, Touché) by KGIII on Friday June 16 2017, @12:14AM
I am always serious! (This is not true.)
Also, it's damned funny - I responded to your question and am off topic. Ah well, I blame you. ;-) They can mod this one off-topic, as well. I ain't scared.
"So long and thanks for all the fish."
(Score: 1, Touché) by Anonymous Coward on Thursday June 15 2017, @06:40PM (3 children)
Break out the torches and pitchforks, troops!! Ms. Pac-Man is a strong, independent female who don't need no man!
(Score: 2) by KGIII on Thursday June 15 2017, @07:57PM
LOL Good catch. I'll expect angry people on my lawn.
She was probably fat anyhow. She ran around eating everything. ;-)
"So long and thanks for all the fish."
(Score: 2) by Azuma Hazuki on Thursday June 15 2017, @08:55PM (1 child)
Where exactly did Pac-Baby come from then?
I am "that girl" your mother warned you about...
(Score: 0) by Anonymous Coward on Friday June 16 2017, @03:13PM
that thing where they figured out how to create sperm from bone marrow, duh.
(Score: 0) by Anonymous Coward on Thursday June 15 2017, @10:40PM (1 child)
Actually, I would like to hear this. I know a bit about randomness from my introductory studies in AI, and some amateur interest. What is this rant abuot?
(Score: 2) by KGIII on Friday June 16 2017, @12:31AM
I posted a good portion of it, in a few different replies.
For starters, we call as sorts of things random that really aren't necessarily random. And, worse, we don't actually have the computational ability to determine if they're random. What they are, however, is "random enough." You can find some pretty good sources of random based on things like cosmic rays. There are some who will say it's true random, there are others who'd disagree.
(Hecht, Jeff, and Torrey, Lee; "Scientists Find Sources of Cosmic Rays," New Scientist, 99:764, 1983.)
I guess, if I had to put myself into a category, I'd say I'm an undecided determinist.
If you can get your hands on a copy of What is Random by Ed Beltrami, that is a good start. A decent uni library should have a copy. If you absolutely can't find a copy, let me know. I think I have an extra. I think...
You might be able to find some works from one Avi Wigderson, as well. They tend to work more along the lines of pseudorandomness and computation. There's still a bunch of good information and I believe they're still in academia so their work should be easier to find. As for a specific work, I'm just gonna suggest diving in at whatever seems interesting.
"So long and thanks for all the fish."
(Score: 5, Funny) by bob_super on Thursday June 15 2017, @05:33PM (2 children)
For the sake of realism, does the emulation come with a little brother who Wants!His!Turn!, a friend telling the algorithm to take the opposite direction, that amazing girl who is definitely not interested, and a creepy guy who's eyeing your underage sister while you're playing?
(Score: 1) by DECbot on Thursday June 15 2017, @07:53PM (1 child)
No, but do you have a bunch of nerds with Cheeto fingers and mountain dew breath on your keyboard fiddling with your life purpose and understanding of the world?
cats~$ sudo chown -R us /home/base
(Score: 2) by bob_super on Thursday June 15 2017, @07:56PM
Nyet, Tovarich. No Cheetos since sanctions. Hard times.