February 28, 2024 • 32 mins
We dive into one of Jason's all-time favorite games: Robo Rally, where you program little robots to play capture the flag while shooting each other with lasers and avoiding deathtraps. Also, some stuff about remote-controlled bacteria and computers destroying the world with paperclips.
Timestamps0:04 - Remote-controlled bacteria 2:18 - Robo Rally background 7:54 - Game mechanics and updates 12:07 - CPUs, GPUs, and computing 17:32 - Machine learning 22:12 - Factory automation 25:38 - Grades and final thoughts
LinksGaming with Science™ is produced with the help of the University of Georgia and is distributed under a Creative Commons Attribution-Noncommercial (CC BY-NC 4.0) license.
Full TranscriptBrian 0:04 Hello, and welcome to the gaming with science podcast where we talk about the science behind some of your favorite games.
Jason 0:10 Today we'll be talking about Robo rally by renegade game studios. Well, welcome to another episode of Gaming with science. I'm Jason.
Brian 0:18 I'm Brian.
Jason 0:19 And today we'll be talking about Robo Rally. Well, before we get into the main topic, though, fun science fact. So Brian, your turn this time? What fun science thing Have you learned in the past bit?
Brian 0:28 So yes, what did I find for us this week, based on the inspiration of Robo Rally and expressing my very severe biology bias, I found an interesting story about remote control the bacteria, maybe were more remote activated than remote controlled, there's a particular strain of E. coli that's approved for medical use in humans. And it can preferentially be taken up by cancer cells, you inject the bacteria into the bloodstream, and they will colonize cancer cells, because they're pretty good at living with less oxygen and solid tumors will often have a lower oxygen environment inside of them. They carry a type of engineered gene that can be turned on by heat very specifically. And by getting them to turn on this gene, you can have the make anti-cancer drugs, for instance. Now how do you turn this on inside of a human being, you basically use a combination of soundwaves to raise temperature in a very specific location at the site of the tumor, which is now colonized by these bacteria. And you kind of like trigger them to maybe not detonate but just start pumping out things that will kill cancer cells.
Jason 1:34 So you basically turn E. coli into a bunch of little suicide bombers.
Brian 1:37 Well, a bunch of little Yeah, a bunch of little attack robots, but a little attack drones saboteur. Yes, saboteur is for sure. Under normal circumstances, you probably don't want E. Coli in your cells, but the enemy of my enemy, I suppose,
Jason 1:49 as long as they don't cure the disease by killing the host. If they're approved for clinical trials, then I assume that little hurdle has been passed. Yeah,
Brian 1:57 you're you're able to use this inside of people, there is a strain of E coli you can inject into someone's bloodstream, and that is an approved form of therapy.
Jason 2:05 Okay... Well, on to the actual topic for today, which is Robo rally. I wanted to do this as soon as I thought about this podcast. Robo rally has been one of my favorite games, since I first played it way back in college. It actually has an interesting history. So it was first published back in 1994. It was first designed in 1985 by Richard Garfield, whose name you might recognize if you're in the gaming area, because he took it to a little gaming company called Wizards of the Coast, who told him that it looked like a great game, but it'd be too expensive for them
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Brian (00:04):
Hello, and welcome to the gaming with science podcast
where we talk about the sciencebehind some of your favorite
games.
Jason (00:10):
Today we'll be talking about Robo rally by renegade
game studios. Well, welcome toanother episode of Gaming with
science. I'm Jason.
Brian (00:18):
I'm Brian.
Jason (00:19):
And today we'll be talking about Robo Rally. Well,
before we get into the maintopic, though, fun science fact.
So Brian, your turn this time?What fun science thing Have you
learned in the past bit?
Brian (00:28):
So yes, what did I find for us this week, based on the
inspiration of Robo Rally andexpressing my very severe
biology bias, I found aninteresting story about remote
control the bacteria, maybe weremore remote activated than
remote controlled, there's aparticular strain of E. coli
that's approved for medical usein humans. And it can
(00:48):
preferentially be taken up bycancer cells, you inject the
bacteria into the bloodstream,and they will colonize cancer
cells, because they're prettygood at living with less oxygen
and solid tumors will often havea lower oxygen environment
inside of them. They carry atype of engineered gene that can
be turned on by heat veryspecifically. And by getting
(01:11):
them to turn on this gene, youcan have the make anti-cancer
drugs, for instance. Now how doyou turn this on inside of a
human being, you basically use acombination of soundwaves to
raise temperature in a veryspecific location at the site of
the tumor, which is nowcolonized by these bacteria. And
you kind of like trigger them tomaybe not detonate but just
(01:31):
start pumping out things thatwill kill cancer cells.
Jason (01:34):
So you basically turn E. coli into a bunch of little
suicide bombers.
Brian (01:37):
Well, a bunch of little Yeah, a bunch of little attack
robots, but a little attackdrones saboteur. Yes, saboteur
is for sure. Under normalcircumstances, you probably
don't want E. Coli in yourcells, but the enemy of my
enemy, I suppose,
Jason (01:49):
as long as they don't cure the disease by killing the
host. If they're approved forclinical trials, then I assume
that little hurdle has beenpassed. Yeah,
Brian (01:57):
you're you're able to use this inside of people, there is
a strain of E coli you caninject into someone's
bloodstream, and that is anapproved form of therapy.
Jason (02:05):
Okay... Well, on to the actual topic for today, which is
Robo rally. I wanted to do thisas soon as I thought about this
podcast. Robo rally has been oneof my favorite games, since I
first played it way back incollege. It actually has an
interesting history. So it wasfirst published back in 1994. It
was first designed in 1985 byRichard Garfield, whose name you
(02:28):
might recognize if you're in thegaming area, because he took it
to a little gaming companycalled Wizards of the Coast, who
told him that it looked like agreat game, but it'd be too
expensive for them to produce.So they wanted something that
would be cheaper and easier forpeople to carry around. They
could play at a convention. Sohe spent a few years and came up
with this little like unknowncard game called Magic the
Gathering. And after that becamea smash success was there said
(02:50):
okay, maybe we can publish therobot game now. So interesting
sidenote, Richard Garfield, he'snot just some random game
designer, I think, based on thetime and it looks like he
designed Robo rally while he wasgetting his Bachelor's in
Computer mathematics. And he didMagic the Gathering while
getting his PhD in combinatorialmathematics. So he has the
actual like scientificcomputational chops behind this,
(03:12):
and I think it shows in the gamedesign. Anyway, it's gone
through a few iterations.There's the original 1994
release. There's the 2005rerelease under Avalon Hill,
that's the one that I originallyowned. Then it got released
again in 2016, with a majorrules upgrade. And then the one
we're going to be talking aboutis the current edition, the
2023, one by Renegade GamesStudios, which mostly builds off
(03:35):
the 2016 edition with a fewlittle tweaks in terms of like
product quality and tiny littlerules tweaks, as far as I can
tell,
Brian (03:42):
oh, wow. So this is the third edition of this game at
this point, basically.
Jason (03:46):
Basiicaly or 2.5, or something, there's only two
really different editions,there's the original one, which
is like 94, and 2005. And thenthere's the 2016 2023, although
there's some some minor tweaks,so it's more like 2.5 edition.
Brian (04:01):
So it's just like Dungeons and Dragons, you skip
over one edition.
Jason (04:04):
Something like that. Yes. And the Board Game Geek ranks on
these are all over the place. Imean, the originals, the highest
rank that around 500, 2016 isabout 1500. The current one is
around 5000. But I think there'sa bias there in terms of just
how many people have reviewedthem, because the current one
actually has the highest averagerating among users. But it's got
(04:25):
the lowest rank. So there'ssomething with the algorithm
putting it there, but the peoplewho have ranked it on average
seem to like the most recent onebest. And I've got to say after
playing it, I kind of like it.There's a lot of quality of life
changes that happened from themy original version to this one
that I like it's a little bitmore streamlined. There's some
of the clunkiness that has goneout. I do enjoy this version
better. As far as what the gameconsists of, for those of you
(04:48):
who've never played it. The ideaof the game is that you're
playing these little robots thatare running around the factory
floor playing basically battlebot Capture the Flag, they're
trying to touch a little flagson the board. and shooting each
other with lasers. And if thatwere all it were, it'd be, it'd
be an OK game. But the thing is,this is a nightmarish factory.
(05:08):
And so there are conveyor beltsand bottomless pits, and pushers
and lasers, some of theexpansions, you can get to have
water or things, the old oneshave like oil slicks, and flame
throwers, crushers, there's allsorts of stuff going on on these
boards. And your goal is to moveyour robot around the
battlefield. Now, the main thingthat makes this challenging is
(05:29):
that you do this by virtue ofhaving a stack of cards that are
your programming cards, you drawup to nine every turn, and then
you put five of them downfacedown in a row. And those are
your next five moves. So youhave to program your robot five
moves at a time to move itaround the board. If it were
just an empty, featureless void,this would be trivial, it would
(05:50):
not be a problem. But the factis that with all the board
elements going on and otherplayers going around, you have
to keep in your mind visualizingwhere will my robot be, which
direction will be facing, whatboard elements will be changing
things, and what my other peopledo to screw me over. So really,
the strategy in the game comesfrom being able to visualize
multiple steps ahead and keepall these different moving parts
(06:11):
in your head and how they affectwhat your robot will be doing.
And a lot of the fun comes fromthat going wrong, either for
yourself or my personal favoritebeing able to screw over other
people by running into them, orpushing them off the plant
track, or anything like that. Soit's a bunch of little computer
controlled chaos, basically. Andso why is it on here, because
it's actually not trying to be ascience game. And most of the
(06:34):
games we're aiming to do in thispodcast are science focused?
Well, I mean, the primaryreasons, because it's one of my
favorite games, and that's oneof the hosts, I can do that. But
the other one is that itactually is a pretty decent
representation of computerprogramming. For my day job,
I've been doing computer codingfor oof, 20 years now?,
(06:54):
something like that, ever sincegraduate school. And playing the
game actually feels a lot likeprogramming a computer, you've
got to think several stepsahead, you have very specific
incremental steps you can do inthe game, it's like you move
forward two spaces you turn,right, you make a U-turn
something like that very smalldefined steps that you have to
piece together into a much morecomplicated whole to in order to
(07:18):
accomplish some objective. Andas happens with real computer
programming, things go wrong andcrap happens. And what I thought
would be great, I make somemistake, or I forgot about
something on the board andeverything goes wrong, because
in the game, if you turn rightinstead of left, or if their
conveyor belt moves you twospaces when you thought it would
move you one, suddenly yourentire program is off. And
(07:40):
instead of touching the flag,this turn, you instead end up
falling off the bottomless pitor ramming into a wall or
getting shot by four differentlaser beams or something crazy
like that,
Brian (07:49):
you're still running that program, the fact that you made
a mistake doesn't matter, youstill have to deal with the
consequences.
Jason (07:54):
Yeah, and so the main parts of the game are the actual
boards that you go on. Thisversion comes with four double
sided boards, there's alreadysome expansions out, you can get
to have additional ones withsome additional board hazards.
You can also just find theseonline, not necessarily the
copyrighted ones that come withthe game, but people have liked
the game for 30 years at thispoint. And so people have just
(08:17):
made custom boards or iconelements, so you can download
custom sheets to print out. Infact, I think the quintessential
one of that is you can look upYouTube videos of people doing a
like life sized one made out ofLego robots at GenCon a few
years ago. So you can watchpeople programming them and
seeing these life side robots,which they made look like some
of the robots in the game. Andsome of them are like R2D2 and
(08:40):
Wally and such moving aroundthis life sized board. So
anyway, you've got your boards,you've got your minis, you've
got your cards, and there's afew other things, some tokens
and like little energy cubes,but the main things are the
board that you move around tothe robots you're moving and the
card to use the program alongwith a shared set of damaged
cards and upgrade cards thatrepresent when you take damage
(09:01):
that kind of fill your deck withuseless stuff or random stuff.
And the upgrade cards which letyou do extra things.
Brian (09:09):
So one thing about this new version is the actual bot
minis got a significant upgrade,right?
Jason (09:15):
Yeah, so previously, they're just unpainted plastic
miniatures. For any of you thathave the old version, good life
hack, you can use those littleplastic things that go around
house keys. You can put thosearound the base to differentiate
them if you're if you like mehave no skill at painting
miniatures, but these ones areactually all pre painted minis
there's only six instead of theoriginal eight, so maybe they're
(09:36):
aiming for a smaller playercount. But yes, they're pre
painted. The original originalgame was actually pewter minis
which are really high qualitybut also kind of expensive and
apparently some peoplecomplained about that at the
time because while nice, it didmake the price significantly
higher. Okay, so Robo rallybuilds itself as being for two
to six players, ages 12 and up.Again, you can play with younger
(09:58):
kids but if you want to play,especially the more advanced
courses, probably on the upperend of that, I know my daughter
used to love not playing thegame, she wanted to set up the
board for us to play and she wasa sadist, she would make the
most difficult hardest board shecould possibly do when she was
like eight, because she didn'twant to play it, she just wanted
(10:19):
to watch us suffer. She hasthankfully gotten beyond that a
little bit. Normal game times itclaims is about 45 to 90
minutes. Obviously, that's veryscalable. You can do this on
just a single board with asingle flag, in which case, it
can be over in 15 minutes if youplay it fast. Or you can have
multiple boards hooked togetherwith multiple flags all over the
place. And you could do a two oreven three hour game. I mean,
(10:42):
theoretically, if you get abunch of the expansions, you
could make an absolutely massiveboard that takes probably
multiple days to run. But whywould someone do that to
themselves? It's a game enjoyfor the time it is, Brian,
you're usually talking about themetaphor of the game. Well, the
metaphor of this game is thatbasically, this is what happens
after the lights go out at thefactory. So the humans go home,
then all the robots power up andthey do this little racing while
(11:04):
the humans aren't there to stopthem. I think in previous
editions, they actually saidthis is a highly advanced
automated factory in the future.And the AI's that run it are
just super, super bored. And sothis is how they're entertaining
themselves.
Brian (11:16):
But, that's not the metaphor anymore? It's not the
super advanced AI?
Jason (11:19):
No, no, this is just what the robots are. autonomous
robots are just battling witheach other for
entertainment.Yep. Becausethat's what you do when the
humans go home.
Brian (11:28):
Yeah, I guess if you could just be reassembled, and
it doesn't really matter if youfall into a bottomless pit, then
why not?
Unknown (11:33):
Yes. And that's definitely one of the quality of
life upgrades is that previouseditions, you had limited number
of lives, which if you lostthem, then you're out of the
game. And that's just not fun tojust sit on the sidelines
watching everyone else. So nowyou have infinite respawns,
although you do take a littlebit of a hit every time, just so
it's not free. Especiallybecause there is a valid
strategy of touching the flag ofone point, killing yourself so
(11:53):
that you respond closer to yournext flag. And you can basically
get a jump on that.
Brian (11:57):
Yeah, we actually did that in one of our family play
sessions, I think. So like,well, if you just dive yourself
down to this pit, you'll be in amore tactical position for the
next flag.
Jason (12:06):
Yep, All right. Now as for the actual science here, so
I admit, when I first put thisup, I knew I wanted to do Robo
Rally, I didn't really knowwhere the science would be. So I
started looking at it andlooking at the pieces. And the
part that really stuck is theprogramming phase where you put
down the five cards, and theycall that the register. So there
are five registers each turn,and you have to do those five in
(12:29):
order as you lay them down. NowI knew that registers were
something in computerprogramming, but I didn't really
know what so I started lookingup and then I went down a rabbit
hole. Because it turns out thishas to do with the way CPU
architecture is built thedifference between CPUs and
GPUs, which we'll get tocryptocurrency machine learning,
like this is like literally thecore of all computation here, in
(12:52):
this little board game, the fivecard register, roughly speaking,
well, similar to that, thatcomputers can do more than five
things. But yeah, because theregister turns out the register
is part of the CPU, the centralprocessing unit, that is what
makes a computer run, it's whathandles all the computation and
data and stuff. And the registeris what actually does those
computations. And it can onlyhold a small number of things at
(13:13):
a time. And kind of the size ofthat register determines the
quality of your CPU. A lot ofyou have probably heard about
like 32 bit architecture versusthe 64 bit architecture. And the
64 bit architecture is the newerthat's determines how much stuff
can actually be held in the CPU64 bits. And it just lets it do
(13:34):
more things at once and handlelarger numbers. Now, the
interesting thing here is when Istarted looking into it, I've
heard about CPUs and GPUsgraphical processing units,
because they turned out they'revery useful for certain types of
computation. They were actuallyoriginally designed for what the
name says graphical processing.So these are the things running
in your game consoles,PlayStations, Xbox, etc, to do
(13:55):
these high end 3D graphics, butthen people found out they were
really useful for all sorts ofother things, the biggest ones
probably being machine learning.So programming, these AI
algorithms, including thingslike chat, GTP, and Dali, and
these other big AI programs, andthen cryptocurrency mining,
specifically Bitcoin, butpresumably also the others. And
the reason has to do with theway they're built. So a central
(14:18):
processing unit, the one that'sin most people's computers, its
goal is to be able to doeverything. So it can be highly
flexible. It can take all sortsof different things in it can
take different processingfunctions and different needs,
and it can move them around andallocate resources and be very,
very flexible. But because ofthat it's not super fast,
(14:39):
relatively speaking. I mean,obviously, nowadays, chips are
actually quite fast relative toprevious ones. But relative to
the other person in town, theGPU, CPUs are actually kind of
slow, because they have to havethat flexibility. A GPU is not
flexible. It has much, much lessability to do other types of
programming or do with differenttypes of programming, but what
(15:02):
it does is it does a certaintype of calculations over and
over and over again very, verywell. It's basically set up to
do many, many more times thiscalculation in parallel, thus
making that particularcalculation faster. Now, this is
really useful for applicationswhere you essentially have to do
the same thing. a bajilliontimes, like with graphics
(15:22):
processing, you just have torender the screen. That's all
you're doing. It's always theexact same thing. Just render
what the screen looks like, withcrypto mining, you have to do
the I actually don't know howBitcoin crypto mining was it
something about hash codes,curious
Brian (15:38):
primes or something, I don't know.
Jason (15:40):
Something like that. I don't do crypto mining, I don't
understand it. But lots ofpeople are trying to make lots
of money by using GPUs to dothat. And then machine learning
it's training. It's crunchingall the data and running all
these different algorithms onit, actually not running that
many different now the samealgorithm just many, many times.
And so that's why GPUs are sofavorable for some things. And
(16:01):
that's why there's actually ashortage of them right now. I
was talking to someone the otherday, they said that someone I
think they were saying the UKhas basically bought all GPU
units that are going to beproduced in the next six months
already, like they're backloggedat this point. Now, I suspect
that's a little bit of anexaggeration, but it gives you
an idea, these things are inreally hot demand precisely
because of their ability to runthese sorts of computation. I
(16:24):
suspect the UK wants them notfor crypto mining, but probably
for machine learningapplications.
Brian (16:29):
Interesting, so a GPU is good at doing one thing, it's
it's a brute force solution toone type of calculation.
Jason (16:37):
Yeah, basically, someone made the comparison that a CPU
is like a fighter jet. It'sfast, it's maneuverable, it can
do all sorts of things. But youcan't actually carry that much
stuff in it. So if your goal isto move something from point A
to point B, you don't want touse a fighter jet. Whereas if
you have like a shipping barge,like it's not fast, it's not
maneuverable, but it can move aton of stuff. And so by virtue
(17:01):
of having the GPU being able tomove a ton of calculations, the
net effect is that you're ableto do those calculations much,
much faster. A differentcomparison someone made is that
a CPU is like a small team ofgeniuses who can do anything you
want them to do, but they take alittle while to learn the new
system and get it set up andgoing. Whereas the GPU is like a
(17:24):
an army of people who may not begeniuses, they're just Okay
people, but you have them doingthe same thing over and over and
over again. And so they just dueto the scale of how many you
have working, they're able toget it done quickly.
Brian (17:37):
So in the metaphor of Robo rally, we're dealing with a
CPU a flexible programmableregister,
Jason (17:43):
pretty much yeah, this it's too simple to be a GPU.
Brian (17:45):
So what would a GPU be in Robo rally?
Jason (17:49):
Ooh, that I don't know if it could be represented in the
game. Because unless you weredoing something where you were
actually trying to learn thegame, by playing it against
itself, it's almost like you'dhave to be a bit of a metagame
where you use it to play thegame a whole bunch of times to
learn the strategies and thenimplement them on the individual
(18:09):
CPU calculations. Because Ican't think of any way where you
want to have like 12 differentregisters going at once and all
your different robots going indifferent directions to figure
out which one actually works.
Brian (18:19):
You're, you're running an army of bots instead of one bot.
Jason (18:23):
Yeah, although people have done that, not to play the
game, but as a teaching tool. Sowhen I was looking into this, I
found that Robo rally has beenused for a long time to teach
computation to people, to highschool students and
undergraduates and such,sometimes it's really simple.
It's just a basic board. Andthey'll just have the robot that
they write the programming codeto help it navigate around
(18:44):
obstacles and end up getting tothe flag. That's pretty simple.
But I saw one person who hadenough that they were actually
doing machine learning on it. Soit was the students tasks to
train a machine learningalgorithm to play Robo rally by
itself. It's not explicitlyprogrammed that here's the flag,
make sure you go forward towardsthe flag, turn to avoid
obstacles, but rather just playthe game a bajillion times, and
(19:07):
learn the rules so that you canplay it on your own. This is
things like Deep Mind and stuffdid with AlphaGo. The original
chess program was deep blue, Ithink. And it was more of a
brute force programming. Butmodern ones I suspect use
machine learning like this.They're using it for go for
poker for pretty much all thethings you're doing now with
(19:28):
games, they're not trying toexplicitly program in the rules
of the game. They're just tryingto have the computer essentially
play against itself a wholebunch of times and learn the
rules.
Brian (19:36):
That's interesting, because in those systems, I've
seen people do things like thisto try to teach an AI how to
play Pokemon and you need to setthe rules up very carefully to
reward and punish appropriately.And I know the flags are the
objective, but how's it going toaccidentally find the flags?
train it to Oh, don't go off theboard or you can't stand still
(19:56):
or stuff like that.
Jason (19:58):
Yeah, basically, and again, And I don't know the
algorithmic details of this, Iknow some of the terminology.
But basically, when it doessomething that you want, which
the first many hundreds or 1000sof times will basically be by
random chance, it gets rewarded.So something about the way code
is executed, that time getsstrengthened, so it's more
likely to happen again. Whereasif something bad happens, you go
(20:19):
off the board, you fall down apit, whatever, then you get
punished. And we're literallytalking 1000s upon 1000s of
plays, just to get the firststep, and then you iteratively
go there. So these complexmachine learning algorithms that
can play Go and chess andpokimane, I've seen Minecraft
and StarCraft being worked on,they take probably millions to
(20:42):
billions of plays to learn therules, basically. But by the end
of it, they're actually reallygood. In fact, I remember when
AlphaGo beat the world Gochampion. The thing with it was
that because it had playedagainst itself, instead of
learning from past human ones,it came up with strategies that
humans don't do. Because Go istaught from essentially master
(21:02):
to student you learn from otherhumans. And so there's a bit of
culture in terms of like, Oh,these are the kinds of moves you
make, like chess has certainopening moves and such. The
computer didn't care. It justdid whatever it happened to
find. And so it found somesolutions that were way outside
the box, as far as human Goplaying was, someone described
it as Go from Mars, in someways, that was probably give it
(21:24):
an edge to beat the humans justbecause it did things that they
weren't expecting,
Brian (21:27):
Sort of developed its own culture. Here's my biology bias.
Again, this sounds an awful lotlike natural selection.
Jason (21:33):
best. And then you'd mutate them again, and so on. If
you look under the hood, whatthey're doing, they're actually
making many, many, manydifferent versions of these AIs,
randomly mutating them, keepingthe ones that do best mutating
those again, over and over andover again. So they're
(21:55):
iteratively, improving it. Andthey are essentially evolving
computers that can do thesetasks.
Brian (22:00):
I guess it's almost an extreme example of artificial
selection, because you've setthe task in front of it that you
wanted to do, but you could doit millions, billions of times.
Jason (22:10):
Yeah. And there's some really good YouTube videos on
this. So it's CPG. Gray has agood one on just general
artificial intelligencetraining. And then there's a
bunch of people that actuallyshow you what it looks like to
train an AI to do something liketo have a make a little AR
avatar walk, just giving itbasic instructions or play
various games are such they'reall over YouTube. So it's
(22:31):
interesting. I mean, it's veryfascinating. Watching the
computer learn to do things alsomay be a little bit scary as
people are realizing what's chatGTP as we're getting ones that
are good enough to mimic a humanand do things. I'm not worried
about computers taking over theworld yet, although that
actually kind of leads into thethird thing I wanted to talk
about. Because looking intothis, like I found stuff about
(22:53):
basic computer programming, Ifound stuff about CPUs and GPUs,
the last one I looked into wasautomated manufacturing. This is
sort of like the quintessentialend goal of replacing people
with robots in factories, whichis where you have a factory that
is essentially completelyrobotic, there are no humans
there. Or maybe there's like oneto make sure things don't break,
(23:14):
or maybe a few people doingquality control. But otherwise,
the factory runs itself. So thecompany Phillips, that makes
razors, they have a factory likethis in the Netherlands,
apparently there's got they'vegot some humans there that are
only for quality control. Andthen this next one actually made
me laugh. So there's a companycalled FANUC, F A N U C don't
know how to pronounce that. InJapan, they have a an automated
(23:37):
factory, where the robots aremaking more robots. And they can
do about 50 robots per day, theywork 24/7, they can go a full
month without any humanschecking in. And it has the
advantages that they don't haveto have lighting or heat or air
conditioning or anything likethat, that the humans need. But
as I read that, I had to thinkHave you not seen any science
fiction films about how robotsactually do take over the world?
(24:00):
The point at which you haverobots making more robots is the
point at which they start takingover the world.
Brian (24:06):
Oh, they have. That's why they did it. What sorts of
robots are they making?
Jason (24:10):
I don't know. I mean, they could just be other
manufacturing robots and such.The thing is like, I'm actually
not concerned about robotstaking over the world in terms
like, oh, they suddenly developsentience and want to command
themselves and be autonomous andget rid of their human over
masters. I don't think we canmake AI that good yet. I'm more
(24:30):
worried about what someonecalled, I think it was termed
the paperclip problem. All youneed is for a sufficiently
powerful AI whose job it is tomake paperclips. decide the best
way to do that is to convert allother mass on the planet into
paperclips. And that's not beingable to stop it. It has no
intelligence as far as we wouldunderstand it. It has no
morality. It's not evil. It'sjust doing its job in a very
(24:54):
efficient and kind ofunfortunate way. That's the kind
of AI I'm worried about is whereit will do what we have
programmed it to do so well thatwe suffer unintended
consequences from it. Probablynot from paperclips. But well,
this is not the time to get intoa spiral off tangent in terms of
what social media and all thatsort of stuff is doing with AI.
(25:14):
That's where it concerns me. Butthankfully, Robo rally is just
cute little robots playing. Whenthat laser tag gets actually
they're shooting each othertrying to blow each other up. So
cute little robots playingbattle bot, capture the flag in
a factory at night when thehumans have gone home.
Brian (25:28):
It's full contact laser tag.
Jason (25:30):
Yes. Oh, definitely pushing is a big part of this.
There's nothing better thanbeing able to push someone's
robot one space to the side andthrow off their entire plans.
Brian (25:39):
Yeah, we went pretty pretty far away from I can't
remember which direction aconveyor belt goes to AI is
making paper clips that convertthe entire planet into paper
clips.
Jason (25:48):
Yes, well, I mean, maybe we'll be better off and we'll
just have the AI is will convertthe entire planet into
computational infrastructure forthem to play Go against each
other. That may be more likewhere we're heading now. But
yes, we did have that issuewhere you cannot remember which
way conveyor belts go. So I knowany game in the future, I just
need to introduce conveyorbelts, so I can win.
Brian (26:08):
But how well do you think the aim of Robo rally sort of
represents the science of themetaphor? Is it doing a good
job?
Jason (26:16):
So this was tricky for me. And I was thinking about
this because we wanted to giveletter grades like how well does
this actually represent thescience of running a robot. And
on the one hand, there's notthat much science here, I mean,
I did have to go looking alittle bit to try to find
something because it really isjust Battle Bot Capture the
Flag. That's what the game istrying to be. It's not trying to
encapsulate a scientificproject. But on the other hand,
(26:37):
playing the game feels likewriting computer code, it
actually feels very similar tome. And I can see it being a
good introductory thing forlike, middle schoolers or such
to teach them the very basicsof, hey, this is how programming
goes. And such. And so I think,for that point, in terms of
capturing the the feel, and theessence of writing code of
programming a computer, I thinkit does pretty well. I mean, if
(26:59):
I were to give it a grade, I'dprobably give it. Well, here's
the thing as just pure scienceportrayal, probably like a B,
B+. But if you take it like howmuch science is actually trying
to convey, I'd bump it up to anA or an A-, because it's not
trying to convey a lot ofscience. It's just trying to be
fun. And using a little bit ofcomputer science to do that. And
(27:19):
it does that little bit quitewell.
Brian (27:21):
Okay, well if we're going to look at it just purely from
the science perspective, youthink maybe a B+ then?
Jason (27:26):
something like that. And that's mostly just because it
doesn't have that much in it.
Brian (27:29):
Yeah, this is not an inherent objective of the game.
It's there, but you kind ofgotta go looking for it.
Jason (27:35):
Yeah, which is not a problem. Like not all games need
to have something in thescience. So
Brian (27:40):
Well, that's true. But our games do you have to have at
least a little bit. So what doesthis game feel like to play? So
let's see. Not facts, butfeelings on this. For me, it
makes me feel like I'm crazy.
Jason (27:54):
How so? like, like, I can see frustration. But what do you
mean crazy?
Brian (27:58):
It makes me feel like I am five years old and can't
remember left from right.
Jason (28:02):
Okay yes, that happens. There have definitely been times
I turned left when I meant toturn right. Yeah, I think one of
our games that happened at leastonce, possibly twice.
Brian (28:09):
It's interesting to me that the metaphor of the game is
no longer I am an advanced AIbecause if I am an advanced AI,
I evidently am one that cannotsolve basic CAPTCHAs of what is
a left and what is it a right,so maybe in that way, sure. I
don't mind playing Robo rally,it's fine. I'm not good at the
game. So it's really aboutfeeling that I am offering very
little competition for someoneI'm playing with. But as long as
(28:31):
they don't mind, I don't mindbeing a bad player at the game.
It's enjoyable to watch yourrobot get pushed in unexpected
ways.
Jason (28:38):
I totally agree. In fact, it was infamous in my family
that we owned this game. And itwas my favorite game for like
five or six years before Iactually won a game. But I still
loved it. It's one of thosegames where I don't care if I
win. It's just fun to play. Andsometimes it's even more fun to
lose spectacularly.
Brian (28:56):
So for those of us might be more videogame inclined for
anybody who played Portal 2 theend of the game involves sort of
a collaborative work of tworobots trying to solve a puzzle
and get through a complexfactory. That's a collaborative
game. In a way Robo rally feelsa little bit like that. But you
are not working together. Youare explicitly working against
each other. But it would beinteresting to see what a
(29:18):
collaborative form of Robo rallywould look like.
Jason (29:21):
I bet people could hack that and now you have me wanting
to make the portal gun upgradefor you just be insane. Although
there are teleporters and one ofthe expansions so actually not
that crazy.
Brian (29:31):
That can be one of the upgrade cards. Yeah, your laser
creates a portal on a flatsurface.
Jason (29:36):
Yeah, Okay, so how about you? If you had to grade the
gameplay? How would this go?
Brian (29:41):
Oh, that's difficult for me. Because again, it's like, I
know this is one of yourfavorite games. It's one that
I'm happy to play, but it's notone that I'm super enthusiac.
Yeah, it's not what I'm gonnaget off the shelf. So if it just
my own pure grade, I'm gonnahave to give it a B, B- because
it's not going to be one that'sgoing to be a go to.
Jason (29:57):
Okay, and obviously, you can probably guess I'm gonna
give it an A or an A+, justbecause I think it is a blast to
play, especially if you can getfour or five people so that the
robots are all running into eachother a lot. We played it first
with just two people. And it's,it's okay with two people. But
you don't get that muchinteraction, when you have four
or five, and you're all runninginto each other and shooting
each other, it becomes a lotmore fun, at least from my
(30:18):
definition of fun.
Brian (30:19):
And we've done some of those games with more people.
Luckily, it's not just the twoof us, we do get to test these
games out with a larger playercount. And so we do kind of know
what that's like as well. So youwould recommend it clearly?
Jason (30:31):
I would clearly recommend this. I love this game. And I
actually really liked the rulesupgrade. So I think they did a
lot of good improvements for it.And I think I now prefer the
newest version over the one Ioriginally bought just because
it's a little bit slicker andsmoother. And the good news is
that most of the pieces,especially the boards are
actually still compatible, youjust slap the board down, maybe
figure out how to put a few ofthe new, the new elements on
(30:54):
what stickers are just print offlittle things you can just place
on as temporary tokens orsomething. But otherwise, it's
still completely compatible.
Brian (31:02):
I don't think we talked about this last time, what's the
price point on this.
Jason (31:05):
So when I got this, the MSRP was $50. Obviously, you can
get it for less at Big box stufffor Amazon, we always encourage
people to support your localgame stores, which are probably
selling it at full price. So Ijust consider that to be the tax
for keeping my friendly localgame store in business. But I
would rather pay a little bitextra and make sure it's going
in the pocket of someone who ishere and local and who loves
(31:26):
board games then to, Well, let'sbe blunt, Amazon technically has
humans running it. But mostlyit's run by an AI.
Brian (31:33):
So we don't want to support robots?
Jason (31:36):
They're doing just fine on their own. I can go to my
local game store and Amazon willnot care.
Brian (31:42):
$50 actually doesn't seem that bad for a game that you're
gonna get this much replay outof. And with that this was sort
of intrinsic resources availableso many ways to support it. So
many different ways to play itif you want to hack it if you
like it $50 seems like a goodvalue.
Jason (31:56):
Yeah, you get a few replays out of it. It's
definitely worth it. And there'sdefinitely a very devoted fan
base that you can find on theinternet with all sorts of
stuff. All right. Well, I thinkthat's where we're going to wrap
it up. Thank you very mucheveryone for listening. Until
next time, have fun, have goodgames, and we will see you next
time. See ya. This has been thegaming with Science Podcast
copyright 2024. listeners arefree to reuse this recording for
(32:17):
any non commercial purpose aslong as credit is given to
Gaming with Science. Thispodcast is produced with the
support from the University ofGeorgia. All opinions are those
of the hosts and do not implyendorsement by the sponsors. If
you wish to purchase any of thegames we talked about, we
encourage you to do so throughyour friendly local game store.
Thank you and have fun playingdice with the universe
Hello, and welcome to the gaming with science podcast
where we talk about the sciencebehind some of your favorite
games.
Jason (00:10):
Today we'll be talking about Robo rally by renegade
game studios. Well, welcome toanother episode of Gaming with
science. I'm Jason.
Brian (00:18):
I'm Brian.
Jason (00:19):
And today we'll be talking about Robo Rally. Well,
before we get into the maintopic, though, fun science fact.
So Brian, your turn this time?What fun science thing Have you
learned in the past bit?
Brian (00:28):
So yes, what did I find for us this week, based on the
inspiration of Robo Rally andexpressing my very severe
biology bias, I found aninteresting story about remote
control the bacteria, maybe weremore remote activated than
remote controlled, there's aparticular strain of E. coli
that's approved for medical usein humans. And it can
(00:48):
preferentially be taken up bycancer cells, you inject the
bacteria into the bloodstream,and they will colonize cancer
cells, because they're prettygood at living with less oxygen
and solid tumors will often havea lower oxygen environment
inside of them. They carry atype of engineered gene that can
be turned on by heat veryspecifically. And by getting
(01:11):
them to turn on this gene, youcan have the make anti-cancer
drugs, for instance. Now how doyou turn this on inside of a
human being, you basically use acombination of soundwaves to
raise temperature in a veryspecific location at the site of
the tumor, which is nowcolonized by these bacteria. And
you kind of like trigger them tomaybe not detonate but just
(01:31):
start pumping out things thatwill kill cancer cells.
Jason (01:34):
So you basically turn E. coli into a bunch of little
suicide bombers.
Brian (01:37):
Well, a bunch of little Yeah, a bunch of little attack
robots, but a little attackdrones saboteur. Yes, saboteur
is for sure. Under normalcircumstances, you probably
don't want E. Coli in yourcells, but the enemy of my
enemy, I suppose,
Jason (01:49):
as long as they don't cure the disease by killing the
host. If they're approved forclinical trials, then I assume
that little hurdle has beenpassed. Yeah,
Brian (01:57):
you're you're able to use this inside of people, there is
a strain of E coli you caninject into someone's
bloodstream, and that is anapproved form of therapy.
Jason (02:05):
Okay... Well, on to the actual topic for today, which is
Robo rally. I wanted to do thisas soon as I thought about this
podcast. Robo rally has been oneof my favorite games, since I
first played it way back incollege. It actually has an
interesting history. So it wasfirst published back in 1994. It
was first designed in 1985 byRichard Garfield, whose name you
(02:28):
might recognize if you're in thegaming area, because he took it
to a little gaming companycalled Wizards of the Coast, who
told him that it looked like agreat game, but it'd be too
expensive for them to produce.So they wanted something that
would be cheaper and easier forpeople to carry around. They
could play at a convention. Sohe spent a few years and came up
with this little like unknowncard game called Magic the
Gathering. And after that becamea smash success was there said
(02:50):
okay, maybe we can publish therobot game now. So interesting
sidenote, Richard Garfield, he'snot just some random game
designer, I think, based on thetime and it looks like he
designed Robo rally while he wasgetting his Bachelor's in
Computer mathematics. And he didMagic the Gathering while
getting his PhD in combinatorialmathematics. So he has the
actual like scientificcomputational chops behind this,
(03:12):
and I think it shows in the gamedesign. Anyway, it's gone
through a few iterations.There's the original 1994
release. There's the 2005rerelease under Avalon Hill,
that's the one that I originallyowned. Then it got released
again in 2016, with a majorrules upgrade. And then the one
we're going to be talking aboutis the current edition, the
2023, one by Renegade GamesStudios, which mostly builds off
(03:35):
the 2016 edition with a fewlittle tweaks in terms of like
product quality and tiny littlerules tweaks, as far as I can
tell,
Brian (03:42):
oh, wow. So this is the third edition of this game at
this point, basically.
Jason (03:46):
Basiicaly or 2.5, or something, there's only two
really different editions,there's the original one, which
is like 94, and 2005. And thenthere's the 2016 2023, although
there's some some minor tweaks,so it's more like 2.5 edition.
Brian (04:01):
So it's just like Dungeons and Dragons, you skip
over one edition.
Jason (04:04):
Something like that. Yes. And the Board Game Geek ranks on
these are all over the place. Imean, the originals, the highest
rank that around 500, 2016 isabout 1500. The current one is
around 5000. But I think there'sa bias there in terms of just
how many people have reviewedthem, because the current one
actually has the highest averagerating among users. But it's got
(04:25):
the lowest rank. So there'ssomething with the algorithm
putting it there, but the peoplewho have ranked it on average
seem to like the most recent onebest. And I've got to say after
playing it, I kind of like it.There's a lot of quality of life
changes that happened from themy original version to this one
that I like it's a little bitmore streamlined. There's some
of the clunkiness that has goneout. I do enjoy this version
better. As far as what the gameconsists of, for those of you
(04:48):
who've never played it. The ideaof the game is that you're
playing these little robots thatare running around the factory
floor playing basically battlebot Capture the Flag, they're
trying to touch a little flagson the board. and shooting each
other with lasers. And if thatwere all it were, it'd be, it'd
be an OK game. But the thing is,this is a nightmarish factory.
(05:08):
And so there are conveyor beltsand bottomless pits, and pushers
and lasers, some of theexpansions, you can get to have
water or things, the old oneshave like oil slicks, and flame
throwers, crushers, there's allsorts of stuff going on on these
boards. And your goal is to moveyour robot around the
battlefield. Now, the main thingthat makes this challenging is
(05:29):
that you do this by virtue ofhaving a stack of cards that are
your programming cards, you drawup to nine every turn, and then
you put five of them downfacedown in a row. And those are
your next five moves. So youhave to program your robot five
moves at a time to move itaround the board. If it were
just an empty, featureless void,this would be trivial, it would
(05:50):
not be a problem. But the factis that with all the board
elements going on and otherplayers going around, you have
to keep in your mind visualizingwhere will my robot be, which
direction will be facing, whatboard elements will be changing
things, and what my other peopledo to screw me over. So really,
the strategy in the game comesfrom being able to visualize
multiple steps ahead and keepall these different moving parts
(06:11):
in your head and how they affectwhat your robot will be doing.
And a lot of the fun comes fromthat going wrong, either for
yourself or my personal favoritebeing able to screw over other
people by running into them, orpushing them off the plant
track, or anything like that. Soit's a bunch of little computer
controlled chaos, basically. Andso why is it on here, because
it's actually not trying to be ascience game. And most of the
(06:34):
games we're aiming to do in thispodcast are science focused?
Well, I mean, the primaryreasons, because it's one of my
favorite games, and that's oneof the hosts, I can do that. But
the other one is that itactually is a pretty decent
representation of computerprogramming. For my day job,
I've been doing computer codingfor oof, 20 years now?,
(06:54):
something like that, ever sincegraduate school. And playing the
game actually feels a lot likeprogramming a computer, you've
got to think several stepsahead, you have very specific
incremental steps you can do inthe game, it's like you move
forward two spaces you turn,right, you make a U-turn
something like that very smalldefined steps that you have to
piece together into a much morecomplicated whole to in order to
(07:18):
accomplish some objective. Andas happens with real computer
programming, things go wrong andcrap happens. And what I thought
would be great, I make somemistake, or I forgot about
something on the board andeverything goes wrong, because
in the game, if you turn rightinstead of left, or if their
conveyor belt moves you twospaces when you thought it would
move you one, suddenly yourentire program is off. And
(07:40):
instead of touching the flag,this turn, you instead end up
falling off the bottomless pitor ramming into a wall or
getting shot by four differentlaser beams or something crazy
like that,
Brian (07:49):
you're still running that program, the fact that you made
a mistake doesn't matter, youstill have to deal with the
consequences.
Jason (07:54):
Yeah, and so the main parts of the game are the actual
boards that you go on. Thisversion comes with four double
sided boards, there's alreadysome expansions out, you can get
to have additional ones withsome additional board hazards.
You can also just find theseonline, not necessarily the
copyrighted ones that come withthe game, but people have liked
the game for 30 years at thispoint. And so people have just
(08:17):
made custom boards or iconelements, so you can download
custom sheets to print out. Infact, I think the quintessential
one of that is you can look upYouTube videos of people doing a
like life sized one made out ofLego robots at GenCon a few
years ago. So you can watchpeople programming them and
seeing these life side robots,which they made look like some
of the robots in the game. Andsome of them are like R2D2 and
(08:40):
Wally and such moving aroundthis life sized board. So
anyway, you've got your boards,you've got your minis, you've
got your cards, and there's afew other things, some tokens
and like little energy cubes,but the main things are the
board that you move around tothe robots you're moving and the
card to use the program alongwith a shared set of damaged
cards and upgrade cards thatrepresent when you take damage
(09:01):
that kind of fill your deck withuseless stuff or random stuff.
And the upgrade cards which letyou do extra things.
Brian (09:09):
So one thing about this new version is the actual bot
minis got a significant upgrade,right?
Jason (09:15):
Yeah, so previously, they're just unpainted plastic
miniatures. For any of you thathave the old version, good life
hack, you can use those littleplastic things that go around
house keys. You can put thosearound the base to differentiate
them if you're if you like mehave no skill at painting
miniatures, but these ones areactually all pre painted minis
there's only six instead of theoriginal eight, so maybe they're
(09:36):
aiming for a smaller playercount. But yes, they're pre
painted. The original originalgame was actually pewter minis
which are really high qualitybut also kind of expensive and
apparently some peoplecomplained about that at the
time because while nice, it didmake the price significantly
higher. Okay, so Robo rallybuilds itself as being for two
to six players, ages 12 and up.Again, you can play with younger
(09:58):
kids but if you want to play,especially the more advanced
courses, probably on the upperend of that, I know my daughter
used to love not playing thegame, she wanted to set up the
board for us to play and she wasa sadist, she would make the
most difficult hardest board shecould possibly do when she was
like eight, because she didn'twant to play it, she just wanted
(10:19):
to watch us suffer. She hasthankfully gotten beyond that a
little bit. Normal game times itclaims is about 45 to 90
minutes. Obviously, that's veryscalable. You can do this on
just a single board with asingle flag, in which case, it
can be over in 15 minutes if youplay it fast. Or you can have
multiple boards hooked togetherwith multiple flags all over the
place. And you could do a two oreven three hour game. I mean,
(10:42):
theoretically, if you get abunch of the expansions, you
could make an absolutely massiveboard that takes probably
multiple days to run. But whywould someone do that to
themselves? It's a game enjoyfor the time it is, Brian,
you're usually talking about themetaphor of the game. Well, the
metaphor of this game is thatbasically, this is what happens
after the lights go out at thefactory. So the humans go home,
then all the robots power up andthey do this little racing while
(11:04):
the humans aren't there to stopthem. I think in previous
editions, they actually saidthis is a highly advanced
automated factory in the future.And the AI's that run it are
just super, super bored. And sothis is how they're entertaining
themselves.
Brian (11:16):
But, that's not the metaphor anymore? It's not the
super advanced AI?
Jason (11:19):
No, no, this is just what the robots are. autonomous
robots are just battling witheach other for
entertainment.Yep. Becausethat's what you do when the
humans go home.
Brian (11:28):
Yeah, I guess if you could just be reassembled, and
it doesn't really matter if youfall into a bottomless pit, then
why not?
Unknown (11:33):
Yes. And that's definitely one of the quality of
life upgrades is that previouseditions, you had limited number
of lives, which if you lostthem, then you're out of the
game. And that's just not fun tojust sit on the sidelines
watching everyone else. So nowyou have infinite respawns,
although you do take a littlebit of a hit every time, just so
it's not free. Especiallybecause there is a valid
strategy of touching the flag ofone point, killing yourself so
(11:53):
that you respond closer to yournext flag. And you can basically
get a jump on that.
Brian (11:57):
Yeah, we actually did that in one of our family play
sessions, I think. So like,well, if you just dive yourself
down to this pit, you'll be in amore tactical position for the
next flag.
Jason (12:06):
Yep, All right. Now as for the actual science here, so
I admit, when I first put thisup, I knew I wanted to do Robo
Rally, I didn't really knowwhere the science would be. So I
started looking at it andlooking at the pieces. And the
part that really stuck is theprogramming phase where you put
down the five cards, and theycall that the register. So there
are five registers each turn,and you have to do those five in
(12:29):
order as you lay them down. NowI knew that registers were
something in computerprogramming, but I didn't really
know what so I started lookingup and then I went down a rabbit
hole. Because it turns out thishas to do with the way CPU
architecture is built thedifference between CPUs and
GPUs, which we'll get tocryptocurrency machine learning,
like this is like literally thecore of all computation here, in
(12:52):
this little board game, the fivecard register, roughly speaking,
well, similar to that, thatcomputers can do more than five
things. But yeah, because theregister turns out the register
is part of the CPU, the centralprocessing unit, that is what
makes a computer run, it's whathandles all the computation and
data and stuff. And the registeris what actually does those
computations. And it can onlyhold a small number of things at
(13:13):
a time. And kind of the size ofthat register determines the
quality of your CPU. A lot ofyou have probably heard about
like 32 bit architecture versusthe 64 bit architecture. And the
64 bit architecture is the newerthat's determines how much stuff
can actually be held in the CPU64 bits. And it just lets it do
(13:34):
more things at once and handlelarger numbers. Now, the
interesting thing here is when Istarted looking into it, I've
heard about CPUs and GPUsgraphical processing units,
because they turned out they'revery useful for certain types of
computation. They were actuallyoriginally designed for what the
name says graphical processing.So these are the things running
in your game consoles,PlayStations, Xbox, etc, to do
(13:55):
these high end 3D graphics, butthen people found out they were
really useful for all sorts ofother things, the biggest ones
probably being machine learning.So programming, these AI
algorithms, including thingslike chat, GTP, and Dali, and
these other big AI programs, andthen cryptocurrency mining,
specifically Bitcoin, butpresumably also the others. And
the reason has to do with theway they're built. So a central
(14:18):
processing unit, the one that'sin most people's computers, its
goal is to be able to doeverything. So it can be highly
flexible. It can take all sortsof different things in it can
take different processingfunctions and different needs,
and it can move them around andallocate resources and be very,
very flexible. But because ofthat it's not super fast,
(14:39):
relatively speaking. I mean,obviously, nowadays, chips are
actually quite fast relative toprevious ones. But relative to
the other person in town, theGPU, CPUs are actually kind of
slow, because they have to havethat flexibility. A GPU is not
flexible. It has much, much lessability to do other types of
programming or do with differenttypes of programming, but what
(15:02):
it does is it does a certaintype of calculations over and
over and over again very, verywell. It's basically set up to
do many, many more times thiscalculation in parallel, thus
making that particularcalculation faster. Now, this is
really useful for applicationswhere you essentially have to do
the same thing. a bajilliontimes, like with graphics
(15:22):
processing, you just have torender the screen. That's all
you're doing. It's always theexact same thing. Just render
what the screen looks like, withcrypto mining, you have to do
the I actually don't know howBitcoin crypto mining was it
something about hash codes,curious
Brian (15:38):
primes or something, I don't know.
Jason (15:40):
Something like that. I don't do crypto mining, I don't
understand it. But lots ofpeople are trying to make lots
of money by using GPUs to dothat. And then machine learning
it's training. It's crunchingall the data and running all
these different algorithms onit, actually not running that
many different now the samealgorithm just many, many times.
And so that's why GPUs are sofavorable for some things. And
(16:01):
that's why there's actually ashortage of them right now. I
was talking to someone the otherday, they said that someone I
think they were saying the UKhas basically bought all GPU
units that are going to beproduced in the next six months
already, like they're backloggedat this point. Now, I suspect
that's a little bit of anexaggeration, but it gives you
an idea, these things are inreally hot demand precisely
because of their ability to runthese sorts of computation. I
(16:24):
suspect the UK wants them notfor crypto mining, but probably
for machine learningapplications.
Brian (16:29):
Interesting, so a GPU is good at doing one thing, it's
it's a brute force solution toone type of calculation.
Jason (16:37):
Yeah, basically, someone made the comparison that a CPU
is like a fighter jet. It'sfast, it's maneuverable, it can
do all sorts of things. But youcan't actually carry that much
stuff in it. So if your goal isto move something from point A
to point B, you don't want touse a fighter jet. Whereas if
you have like a shipping barge,like it's not fast, it's not
maneuverable, but it can move aton of stuff. And so by virtue
(17:01):
of having the GPU being able tomove a ton of calculations, the
net effect is that you're ableto do those calculations much,
much faster. A differentcomparison someone made is that
a CPU is like a small team ofgeniuses who can do anything you
want them to do, but they take alittle while to learn the new
system and get it set up andgoing. Whereas the GPU is like a
(17:24):
an army of people who may not begeniuses, they're just Okay
people, but you have them doingthe same thing over and over and
over again. And so they just dueto the scale of how many you
have working, they're able toget it done quickly.
Brian (17:37):
So in the metaphor of Robo rally, we're dealing with a
CPU a flexible programmableregister,
Jason (17:43):
pretty much yeah, this it's too simple to be a GPU.
Brian (17:45):
So what would a GPU be in Robo rally?
Jason (17:49):
Ooh, that I don't know if it could be represented in the
game. Because unless you weredoing something where you were
actually trying to learn thegame, by playing it against
itself, it's almost like you'dhave to be a bit of a metagame
where you use it to play thegame a whole bunch of times to
learn the strategies and thenimplement them on the individual
(18:09):
CPU calculations. Because Ican't think of any way where you
want to have like 12 differentregisters going at once and all
your different robots going indifferent directions to figure
out which one actually works.
Brian (18:19):
You're, you're running an army of bots instead of one bot.
Jason (18:23):
Yeah, although people have done that, not to play the
game, but as a teaching tool. Sowhen I was looking into this, I
found that Robo rally has beenused for a long time to teach
computation to people, to highschool students and
undergraduates and such,sometimes it's really simple.
It's just a basic board. Andthey'll just have the robot that
they write the programming codeto help it navigate around
(18:44):
obstacles and end up getting tothe flag. That's pretty simple.
But I saw one person who hadenough that they were actually
doing machine learning on it. Soit was the students tasks to
train a machine learningalgorithm to play Robo rally by
itself. It's not explicitlyprogrammed that here's the flag,
make sure you go forward towardsthe flag, turn to avoid
obstacles, but rather just playthe game a bajillion times, and
(19:07):
learn the rules so that you canplay it on your own. This is
things like Deep Mind and stuffdid with AlphaGo. The original
chess program was deep blue, Ithink. And it was more of a
brute force programming. Butmodern ones I suspect use
machine learning like this.They're using it for go for
poker for pretty much all thethings you're doing now with
(19:28):
games, they're not trying toexplicitly program in the rules
of the game. They're just tryingto have the computer essentially
play against itself a wholebunch of times and learn the
rules.
Brian (19:36):
That's interesting, because in those systems, I've
seen people do things like thisto try to teach an AI how to
play Pokemon and you need to setthe rules up very carefully to
reward and punish appropriately.And I know the flags are the
objective, but how's it going toaccidentally find the flags?
train it to Oh, don't go off theboard or you can't stand still
(19:56):
or stuff like that.
Jason (19:58):
Yeah, basically, and again, And I don't know the
algorithmic details of this, Iknow some of the terminology.
But basically, when it doessomething that you want, which
the first many hundreds or 1000sof times will basically be by
random chance, it gets rewarded.So something about the way code
is executed, that time getsstrengthened, so it's more
likely to happen again. Whereasif something bad happens, you go
(20:19):
off the board, you fall down apit, whatever, then you get
punished. And we're literallytalking 1000s upon 1000s of
plays, just to get the firststep, and then you iteratively
go there. So these complexmachine learning algorithms that
can play Go and chess andpokimane, I've seen Minecraft
and StarCraft being worked on,they take probably millions to
(20:42):
billions of plays to learn therules, basically. But by the end
of it, they're actually reallygood. In fact, I remember when
AlphaGo beat the world Gochampion. The thing with it was
that because it had playedagainst itself, instead of
learning from past human ones,it came up with strategies that
humans don't do. Because Go istaught from essentially master
(21:02):
to student you learn from otherhumans. And so there's a bit of
culture in terms of like, Oh,these are the kinds of moves you
make, like chess has certainopening moves and such. The
computer didn't care. It justdid whatever it happened to
find. And so it found somesolutions that were way outside
the box, as far as human Goplaying was, someone described
it as Go from Mars, in someways, that was probably give it
(21:24):
an edge to beat the humans justbecause it did things that they
weren't expecting,
Brian (21:27):
Sort of developed its own culture. Here's my biology bias.
Again, this sounds an awful lotlike natural selection.
Jason (21:33):
best. And then you'd mutate them again, and so on. If
you look under the hood, whatthey're doing, they're actually
making many, many, manydifferent versions of these AIs,
randomly mutating them, keepingthe ones that do best mutating
those again, over and over andover again. So they're
(21:55):
iteratively, improving it. Andthey are essentially evolving
computers that can do thesetasks.
Brian (22:00):
I guess it's almost an extreme example of artificial
selection, because you've setthe task in front of it that you
wanted to do, but you could doit millions, billions of times.
Jason (22:10):
Yeah. And there's some really good YouTube videos on
this. So it's CPG. Gray has agood one on just general
artificial intelligencetraining. And then there's a
bunch of people that actuallyshow you what it looks like to
train an AI to do something liketo have a make a little AR
avatar walk, just giving itbasic instructions or play
various games are such they'reall over YouTube. So it's
(22:31):
interesting. I mean, it's veryfascinating. Watching the
computer learn to do things alsomay be a little bit scary as
people are realizing what's chatGTP as we're getting ones that
are good enough to mimic a humanand do things. I'm not worried
about computers taking over theworld yet, although that
actually kind of leads into thethird thing I wanted to talk
about. Because looking intothis, like I found stuff about
(22:53):
basic computer programming, Ifound stuff about CPUs and GPUs,
the last one I looked into wasautomated manufacturing. This is
sort of like the quintessentialend goal of replacing people
with robots in factories, whichis where you have a factory that
is essentially completelyrobotic, there are no humans
there. Or maybe there's like oneto make sure things don't break,
(23:14):
or maybe a few people doingquality control. But otherwise,
the factory runs itself. So thecompany Phillips, that makes
razors, they have a factory likethis in the Netherlands,
apparently there's got they'vegot some humans there that are
only for quality control. Andthen this next one actually made
me laugh. So there's a companycalled FANUC, F A N U C don't
know how to pronounce that. InJapan, they have a an automated
(23:37):
factory, where the robots aremaking more robots. And they can
do about 50 robots per day, theywork 24/7, they can go a full
month without any humanschecking in. And it has the
advantages that they don't haveto have lighting or heat or air
conditioning or anything likethat, that the humans need. But
as I read that, I had to thinkHave you not seen any science
fiction films about how robotsactually do take over the world?
(24:00):
The point at which you haverobots making more robots is the
point at which they start takingover the world.
Brian (24:06):
Oh, they have. That's why they did it. What sorts of
robots are they making?
Jason (24:10):
I don't know. I mean, they could just be other
manufacturing robots and such.The thing is like, I'm actually
not concerned about robotstaking over the world in terms
like, oh, they suddenly developsentience and want to command
themselves and be autonomous andget rid of their human over
masters. I don't think we canmake AI that good yet. I'm more
(24:30):
worried about what someonecalled, I think it was termed
the paperclip problem. All youneed is for a sufficiently
powerful AI whose job it is tomake paperclips. decide the best
way to do that is to convert allother mass on the planet into
paperclips. And that's not beingable to stop it. It has no
intelligence as far as we wouldunderstand it. It has no
morality. It's not evil. It'sjust doing its job in a very
(24:54):
efficient and kind ofunfortunate way. That's the kind
of AI I'm worried about is whereit will do what we have
programmed it to do so well thatwe suffer unintended
consequences from it. Probablynot from paperclips. But well,
this is not the time to get intoa spiral off tangent in terms of
what social media and all thatsort of stuff is doing with AI.
(25:14):
That's where it concerns me. Butthankfully, Robo rally is just
cute little robots playing. Whenthat laser tag gets actually
they're shooting each othertrying to blow each other up. So
cute little robots playingbattle bot, capture the flag in
a factory at night when thehumans have gone home.
Brian (25:28):
It's full contact laser tag.
Jason (25:30):
Yes. Oh, definitely pushing is a big part of this.
There's nothing better thanbeing able to push someone's
robot one space to the side andthrow off their entire plans.
Brian (25:39):
Yeah, we went pretty pretty far away from I can't
remember which direction aconveyor belt goes to AI is
making paper clips that convertthe entire planet into paper
clips.
Jason (25:48):
Yes, well, I mean, maybe we'll be better off and we'll
just have the AI is will convertthe entire planet into
computational infrastructure forthem to play Go against each
other. That may be more likewhere we're heading now. But
yes, we did have that issuewhere you cannot remember which
way conveyor belts go. So I knowany game in the future, I just
need to introduce conveyorbelts, so I can win.
Brian (26:08):
But how well do you think the aim of Robo rally sort of
represents the science of themetaphor? Is it doing a good
job?
Jason (26:16):
So this was tricky for me. And I was thinking about
this because we wanted to giveletter grades like how well does
this actually represent thescience of running a robot. And
on the one hand, there's notthat much science here, I mean,
I did have to go looking alittle bit to try to find
something because it really isjust Battle Bot Capture the
Flag. That's what the game istrying to be. It's not trying to
encapsulate a scientificproject. But on the other hand,
(26:37):
playing the game feels likewriting computer code, it
actually feels very similar tome. And I can see it being a
good introductory thing forlike, middle schoolers or such
to teach them the very basicsof, hey, this is how programming
goes. And such. And so I think,for that point, in terms of
capturing the the feel, and theessence of writing code of
programming a computer, I thinkit does pretty well. I mean, if
(26:59):
I were to give it a grade, I'dprobably give it. Well, here's
the thing as just pure scienceportrayal, probably like a B,
B+. But if you take it like howmuch science is actually trying
to convey, I'd bump it up to anA or an A-, because it's not
trying to convey a lot ofscience. It's just trying to be
fun. And using a little bit ofcomputer science to do that. And
(27:19):
it does that little bit quitewell.
Brian (27:21):
Okay, well if we're going to look at it just purely from
the science perspective, youthink maybe a B+ then?
Jason (27:26):
something like that. And that's mostly just because it
doesn't have that much in it.
Brian (27:29):
Yeah, this is not an inherent objective of the game.
It's there, but you kind ofgotta go looking for it.
Jason (27:35):
Yeah, which is not a problem. Like not all games need
to have something in thescience. So
Brian (27:40):
Well, that's true. But our games do you have to have at
least a little bit. So what doesthis game feel like to play? So
let's see. Not facts, butfeelings on this. For me, it
makes me feel like I'm crazy.
Jason (27:54):
How so? like, like, I can see frustration. But what do you
mean crazy?
Brian (27:58):
It makes me feel like I am five years old and can't
remember left from right.
Jason (28:02):
Okay yes, that happens. There have definitely been times
I turned left when I meant toturn right. Yeah, I think one of
our games that happened at leastonce, possibly twice.
Brian (28:09):
It's interesting to me that the metaphor of the game is
no longer I am an advanced AIbecause if I am an advanced AI,
I evidently am one that cannotsolve basic CAPTCHAs of what is
a left and what is it a right,so maybe in that way, sure. I
don't mind playing Robo rally,it's fine. I'm not good at the
game. So it's really aboutfeeling that I am offering very
little competition for someoneI'm playing with. But as long as
(28:31):
they don't mind, I don't mindbeing a bad player at the game.
It's enjoyable to watch yourrobot get pushed in unexpected
ways.
Jason (28:38):
I totally agree. In fact, it was infamous in my family
that we owned this game. And itwas my favorite game for like
five or six years before Iactually won a game. But I still
loved it. It's one of thosegames where I don't care if I
win. It's just fun to play. Andsometimes it's even more fun to
lose spectacularly.
Brian (28:56):
So for those of us might be more videogame inclined for
anybody who played Portal 2 theend of the game involves sort of
a collaborative work of tworobots trying to solve a puzzle
and get through a complexfactory. That's a collaborative
game. In a way Robo rally feelsa little bit like that. But you
are not working together. Youare explicitly working against
each other. But it would beinteresting to see what a
(29:18):
collaborative form of Robo rallywould look like.
Jason (29:21):
I bet people could hack that and now you have me wanting
to make the portal gun upgradefor you just be insane. Although
there are teleporters and one ofthe expansions so actually not
that crazy.
Brian (29:31):
That can be one of the upgrade cards. Yeah, your laser
creates a portal on a flatsurface.
Jason (29:36):
Yeah, Okay, so how about you? If you had to grade the
gameplay? How would this go?
Brian (29:41):
Oh, that's difficult for me. Because again, it's like, I
know this is one of yourfavorite games. It's one that
I'm happy to play, but it's notone that I'm super enthusiac.
Yeah, it's not what I'm gonnaget off the shelf. So if it just
my own pure grade, I'm gonnahave to give it a B, B- because
it's not going to be one that'sgoing to be a go to.
Jason (29:57):
Okay, and obviously, you can probably guess I'm gonna
give it an A or an A+, justbecause I think it is a blast to
play, especially if you can getfour or five people so that the
robots are all running into eachother a lot. We played it first
with just two people. And it's,it's okay with two people. But
you don't get that muchinteraction, when you have four
or five, and you're all runninginto each other and shooting
each other, it becomes a lotmore fun, at least from my
(30:18):
definition of fun.
Brian (30:19):
And we've done some of those games with more people.
Luckily, it's not just the twoof us, we do get to test these
games out with a larger playercount. And so we do kind of know
what that's like as well. So youwould recommend it clearly?
Jason (30:31):
I would clearly recommend this. I love this game. And I
actually really liked the rulesupgrade. So I think they did a
lot of good improvements for it.And I think I now prefer the
newest version over the one Ioriginally bought just because
it's a little bit slicker andsmoother. And the good news is
that most of the pieces,especially the boards are
actually still compatible, youjust slap the board down, maybe
figure out how to put a few ofthe new, the new elements on
(30:54):
what stickers are just print offlittle things you can just place
on as temporary tokens orsomething. But otherwise, it's
still completely compatible.
Brian (31:02):
I don't think we talked about this last time, what's the
price point on this.
Jason (31:05):
So when I got this, the MSRP was $50. Obviously, you can
get it for less at Big box stufffor Amazon, we always encourage
people to support your localgame stores, which are probably
selling it at full price. So Ijust consider that to be the tax
for keeping my friendly localgame store in business. But I
would rather pay a little bitextra and make sure it's going
in the pocket of someone who ishere and local and who loves
(31:26):
board games then to, Well, let'sbe blunt, Amazon technically has
humans running it. But mostlyit's run by an AI.
Brian (31:33):
So we don't want to support robots?
Jason (31:36):
They're doing just fine on their own. I can go to my
local game store and Amazon willnot care.
Brian (31:42):
$50 actually doesn't seem that bad for a game that you're
gonna get this much replay outof. And with that this was sort
of intrinsic resources availableso many ways to support it. So
many different ways to play itif you want to hack it if you
like it $50 seems like a goodvalue.
Jason (31:56):
Yeah, you get a few replays out of it. It's
definitely worth it. And there'sdefinitely a very devoted fan
base that you can find on theinternet with all sorts of
stuff. All right. Well, I thinkthat's where we're going to wrap
it up. Thank you very mucheveryone for listening. Until
next time, have fun, have goodgames, and we will see you next
time. See ya. This has been thegaming with Science Podcast
copyright 2024. listeners arefree to reuse this recording for
(32:17):
any non commercial purpose aslong as credit is given to
Gaming with Science. Thispodcast is produced with the
support from the University ofGeorgia. All opinions are those
of the hosts and do not implyendorsement by the sponsors. If
you wish to purchase any of thegames we talked about, we
encourage you to do so throughyour friendly local game store.
Thank you and have fun playingdice with the universe
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.