The trick is to schedule the next session at the end of the current session. Everybody closes their books, opens their agenda and nobody leaves until you all write down a date and time and place.
@@garak55 The trick is to have a pool of people and a fixed schedule. This way, most players will treat it as a regular occurrence with more importance than if you schedule it again each time at the end of a session.
In my experience D&D is about having a group of 8-16 people who may or may not show up, and then spending 3 hours deciding what weapons we're bringing. I might be doing this wrong.
I love thinking about the meeting where Google decided that their motto should no longer be "Don't be evil!" Like someone had to say out loud, "We might have to be evil a little bit every once in a while."
They gave up the charade when their CEO said (regarding privacy of users): "If you have something that you don't want anyone to know, maybe you shouldn't be doing it in the first place." Of course they were gonna be evil. It's what corporate executives do.
The issue of course is the definition of evil. People define it differently and it’s probably a philosophical headache having to debate it all the time.
Douglas Adams was a genius. He predicted the notion of a black box algorithmic computer program giving the answer to life, the universe, and everything without supplying the thought process or derived question that led to the result of "42".
When my father was young, he and a friend made up the dumbest variant of stratego imaginable. You flip the pieces so you can only see those of your opponent and not your own (also when placing them, it's not about memorizing what your pieces are). You can move every piece, even if it's a bomb or your flag, because how are you to know you can't move it? The other rules remain largely unchanged. Good luck!
We played this at school. The flag has to be on the back row otherwise you can do a random set up or opponent places your board. It’s not as fun as the real thing though, but it does produce silly things like Bombs taking flags. You also play the landmine rule that bombs disappear when they explode else it’s just a game of who can find their bombs first
Stratego is decended from the French game 'L'Attaque'. The bomb is actually called a mine there, and it implies a mined field. The miner is called a mineur or sapeur, both imply a deminer unit.
And a descendant of the Chinese game named "Jungle". It is very similar, but with animals, being the elephant the strongest that can be beaten by the weakest, the mouse. But the banner is a fixed place on the board.
You made a 40 minute video expressing strong opinions about stratego. I dunno how to tell you this angela but you're member of the board game community now Also, Patchwork is a surprisingly fun and cheap 2-play game. It looks like it wouldn't be fun when you read it, but I was very pleased by it - this coming from a big board game person
I LOVE the direction of this channel. Came for the physics rants. Stayed for the general STEM rants. (But don't leave the physics stuff, please. You bring up conceptual nuances that many physics classes miss)
There's the silly humor too, .......alkaline water with lemon... I've come to believe ignorance is the most fundamental force of nature I could totally relate to that frustration.
Just a note about chess (38:20): it's actually completely fine to resign a game when you know you've lost. In fact it's proper chess etiquette to do so if you're totally lost rather than make your opponent go through the motions, though when a game is considered sufficiently lost that it's worth resigning depends heavily on the level of the players. Usually people would rather not have to play another thirty moves to mop up the game after you've blundered a queen.
My favorite is that clip of Carlsen and ... Hikaru maybe? Where one of them like moves a pawn in nowhere, they both stare at the board for a minute and then they go "Oh, I thought you could ..." "Right but then ..." "You're right." "Congratulations" And shake hands. They each just finished the game in their minds and agreed who won. Amazing.
@@davidgustavsson4000then there's how i play it, losing queen with ten moves and an endgame consisting of two pieces chasing a king around indefinitely but the opponent is also as bad as me so won't agree a draw.
@@davidgustavsson4000 That sounds you misunderstood that the game was over already. They often do a analyze afterwards (called a Post Mortem), and this analysis could go as you described because both still have the calculated lines in the heads. But they do not talk with each other while the game is ongoing, that would be a serious break of etiquette and rules.
30:30 Underrepresenting Vincent here, he is not just a stratego player, but also an AI researcher. (His sister Roseline, who was women's world champion, is a friend of mine.)
He’s the most important character in this video! Also please do not show Roseline de Boer this video. I will be very embarrassed of my fangirling all over this game.
@@idontwantahandlethough There's an International Stratego Federation and even a whole Mind Sports Olympiad (MSO) that includes Stratego. BTW, Demis Hassabis (DeepMind co-founder and a senior author of this paper) has won the MSO Pentamind (best all-round board games player) five times. So yeah, "these machine learning experts" may _actually_ be interested in Stratego.
Wait ... WOMEN'S world champion? The championship divisions are segregated by gender? Why? Do they use 100 lb. playing pieces that are harder for the average woman to lift?
You don't have to watch your friend clumsily win against you after you made a mistake. It is part of the culture and etiquette to simply resign a game that is unrecoverable (or at least, very unlikely to be recovered). It's a sign of respecting both their time and skill to not force them to prove they can win a one-sided board. :)
That's apparently the opposite of league of legends, wherein you apparently do not respect your opponent at all and in fact resent and hate them for playing the game at all.
I only have one solid memory of this game, though I remember loving it. I attacked with like a nine or an eight, my friend told me theirs was a ten, but I remembered their ten being on the other side of the board so I turned their piece over. I was right, he was lying. Then he got mad at me for checking and tossed the board. I had some odd friendships as a kid.
I had a friend that used to always play Sonic on his computer. But then when it was my turn to play, he'd always say the computer "needed a break". He also flipped board games a lot. What a guy 🙄
Of course, the most common usage of this in video games is "The AI is so [stupid, easy to game, inefffective]" so maybe *Intelligence* is a poor choice for this.
@@jfirth816 I keep wondering where people draw the line of intelligence if we look at animals rather than computers... If an algorithm is considered unintelligent because of criterias we would likely not apply to animals, is it that animals are not actually intelligent or that our algorithms are intelligent but we don't like to think about it?
@@Nathouuuutheone While I'm not sure why you're asking a guy who was riffing about computer games this, but the definition of "intelligent" would be the first thing to address to make any headway at all. Even my limited knowledge won't fit in a RUclips comment, but I believe that a lot of folks who think about these things would include "self awareness" of "intelligence" or more specifically "human intelligence" That's something we know humans can do and we can argue (or research, but mostly argue) whether or not animals can do it. But we wouldn't argue about computer algorithms having self awareness. Humans write these and don't put in self awareness, so they are not intelligent. They are algorithms. I think this captures the gist of why we can argue about it with animals but flatly state that computers are not intelligent.
38:15 Resigning from a losing position in chess isn't generally seen as bad manners, and shorter clocktimes can help out with this as well! My move-accuracy in losing positions is *way* higher than in winning ones so i'm a bit of a Clock Fiend
In hundreds of World Chess Championship matches, there has only been a single checkmate, and that happened in 1929. There were two stalemates in 1978 and 2007. Every other game had ended in either draws or resignations. Resigning is not only acceptable, it is expected. I've played lots of chess online. Every one of my games ended in resignations (because I'm a lousy player, at higher levels, draws are more common. Plus I play a lot of blitz 😊 ).
Since it’s impossible to get people to stop saying AI as shorthand for machine learning, maybe we can slowly get people to understand AI (in that context) & AGI (what they think AI is) are different things.
Robert Miles has an AI Safety channel here; he has a good handle on it all, and perhaps a more effective strategy for "I can't be bothered / you know what I meant" people is to just suggest his channel for more info. This gives them a chance to pick it up naturally.
AI has always been referred to as AI. The computer player in Civilization I had "AI", it was just limited. It was also heuristic rather than neural network, because this was 1990. Heck, computers used to be referred to "artificial thinking machines". "Artificial General Intelligence" is what all the hype lately about, and what we don't have. There is a qualitative difference between older AI technologies and more recent machine learning techniques, but we are still just talking about better Artificial Narrow Intelligence. The computer program can be trained in a particular task. Also, it's become a field in itself rather than just some automated rules for your CompSci tic-tac-toe exercise. It's all just semantics, but technically ANI and AGI are both AI.
I was in school when they called this stuff Expert Systems, and AI in coding was basically some blocks moving around the screen deciding whether they should run from or "eat" other blocks :)
I don't get the impression that most people who play video games mistake the "AI" in games for AGI. In most games, general opinion of AI's capability to play it is quite low, and players routinely find loopholes in algorithms to exploit...knowing full well the AI won't update with an expectation of the player attempting to repeat those exploits. To me, this presents at least intuitive understanding that "game AI =/= AGI" and that they're playing against an algorithm made by the devs, even if one doesn't know the term AGI. Machine learning has made some of these algorithms nearly insurmountable to humans though, even in games with incomplete information. APM-capped AlphaStar is a great example of that...despite being limited to "visible" information to players and APM far lower than pro players, it can still crush pro players. Even in this case, most random internet players don't mistake AlphaStar for something that can write a resume for you.
I suppose the authors felt the availability of the source warranted having only a surface-level description in the paper. But, I'm glad to hear someone who reads and navigates lots of papers share my complaint that machine learning papers tend to cite sources that are unreachable and/or papers all the way down. It's not just me! (or you!)
Scheduling D&D is definitely like herding cats. The best way is to incentivize them with food, whether it's players or cats. Sometimes it seems like I can't get a group together, but weblately I'm up to two alternating campaigns on Sundays and another game night on Tuesdays that I'm hoping to leverage into another campaign.
My granddad had the game generals, basically like stratego with no mines or lakes. All ranks move the same tho, and there’s a simple computer built into the side of the board that reads notches carved on the bottom of the pieces. The attacking and defending piece preserve their anonymity (to a degree). It’s a super fun game and a very early computer augmented board game!!!
One problem in the CS world is that things just move too fast - everyone cites conference papers because by the time a concept has been published in a full journal it's likely already been iterated on and 1-upped in a conference. That's why most citations are for conferences. Or that's what I've heard my colleagues say, I'm doing my phd in EE but with a heavy CS crossover
It's more than just that though - CS conferences are archival publications, just like journals in other fields. They aren't talks, they are papers, acompanied by talks, but what is being cited is the paper... I found free full text versions of the first 3 papers @acollierastro "cannot read" in 5 minutes. Many CS conferences have higher impact factors than all but the very best journals in CS. For someone to claim they read ML papers as part of their job not to know that conferences publish the full text of papers in CS is crazy...
Little note about chess puzzles: for most of them, it’s not about getting a win a higher percentage of the time. if you make the right moves for the puzzle, it’s a guaranteed win. They usually involve “forcing” moves where the other player has no choice but to defend or move their king in an unideal way. Of course there are other types of puzzles that have different goals that might just improve your position… those are annoying lol
I really enjoy cooperative board games, because they can still be tense, stressful, and challenging, without creating bad blood between players. My favorite is Spirit Island, where players take the role of guardian deities who use magic to protect an island from colonial invaders
Now I'm going to get out my vintage 1970's Stratego for a few sessions this weekend with the kids. The Blue and Red pieces, with their silver and gold embossing, are like visual candy.
I’m sure you have discovered by now that the AI/ML field primarily publishes at a series of headline conferences there isn’t really a journal culture beyond preprints. For anyone confused, all the papers highlighted at 20:55 that reference Q learning and R-NaD are of course available online and are not talks! I recommend the papers by Tuyls et al for the roots of the approach. You will become quickly annoyed by how hard it is to replicate many papers because of a lack of information. Many famous papers have quite large discrepancies between what the paper states they do and what their released code actually does! Much of the nitty gritty details are seemingly intentionally left out sometimes especially when it comes to exactly how the model architecture is initialised, pretrained, etc. it’s a problem that has been rampant for more than a decade and unfortunately seems only set to get worse as models get enormous and require enormous compute; why tell us how it works when nobody can economically replicate… /rant
2:20 i've literally never heard of this game before but i didn't even make it 5 minutes in without furiously googling it, asking my older brother if he's ever played it, and checking if there's a good online multiplayer version of it to play with friends overseas. So. you could say you have succeeded. :B
My fiancée and I started playing Stratego last year and it’s been a lot of fun. I began playing with my strategy from when I played as a kid (surround your flag with bombs), but she was new to the game. I won the first few rounds but then she figured out a much better setup that I just couldn’t match (place your flag randomly on the back row to make the opponent guess where it is). Eventually I had to adopt her setup and now we are very evenly matched. I didn’t know there was a competitive scene so I’ll have to check it out now
I'm on the bus going to a game night right now! We have a biweekly hobby circle with an average of 4 participants. A big reason for this to be possible is that some of us had to move back to our home town after college/university due to unemployment, and then we got jobs in the local area. So the silver lining of temporary joblessness turned out to be a sizable game group 🥳 Side note: try Fugitive, it is a very fun 2-player card game of bluffing and deduction.
This brings me back to my days in a "gifted program" during grade school where one day a week the gifted students would go to their own classes that were often about more fun things while still learning(like how to design a bridge or program a Lego robot or just doing research on a country). In my homeroom there was a copy of Stratego that we would always play before class started. It almost reminded me of Ender's Game where they have all of these smart students separated and playing wargames.
DeepNash's ability to trade off information vs pieces is mind boggling to me. It is so willing to sacrifice pieces just to keep the opponent in the dark, then come back and win.
The video is a year old so I doubt anyone will see this but, if I were to recommend a game I would recommend my favourite board game, BattleTech. I don't know much about Warhammer 40K but I have heard it compared to that. It's got all kinds of playable vehicles in it, but it's mainly about the mechs which it has a lot of. It's a wargame that is played on a hex grid (Although there are rules for playing in a 3d map without hexes if you wanna build your own homemade 3d map) where you control a force of mechs and or other vehicles like tanks and vtols and even aerospace fighters and space ships. There are a few box sets, I would recommend A Game of Armored Combat it gives you a lot of rules to use, and a decent amount of miniatures. The beginner box is cheaper, but has more basic and simplified rules, and 2 minis instead of 8. There is a large expansive lore, if you ever get interested in that I would recommend the youtube channel BPL or Black Pants Legion. They have a series called Tex Talks BattleTech. But I would specifically recommend their video BattleTech 101: The Great Houses of the Inner Sphere to start off.
The style of playing MTG you’re describing at the end of the video actually has a name! It’s called ‘Jumpstart’ and it’s a really popular way to play. Basically the only way I play MTG is on a game called ‘Tabletop simulator’, it’s less than $40 iirc and is just a really good and easy to understand platform for simulating a board game table. I usually load up a bunch of random booster packs Pull two random starters and play them blind, super fun.
As a chess player I love it when my opponent resigns, its not rude at all, in fact its often considered polite, though that depends on the position. I think a lot of the fun and beauty in chess comes from that idea of every game being almost always completely different. Of the vast amount of different moves and decisions possible for each player, you're almost always in a brand new position you've never seen before by move 15. Calculating these new positions in real time against an opponent doing the same can be quite exhilarating, especially in very complicated positions, and is one of many reasons I fell in love with the game. :)
Me and my sister created tons of variations for this game when we were kids. Our favorite version was "reverse mode", in which you would randomly set up your army, and you could only see your opponents army. So you could actually move your bombs around without knowing it (but they were removed afterwards)
Never played Stratego, but the hidden information aspect of it is very compelling to me. Sounds similar to a card game I absolutely adore - Android: Netrunner. If you want a 2 player game combing aspects of Magic: The Gathering style card games with hidden information on the board (with the added twist that its asymmetric and both players play by different rules) I'd suggest checking it out. It is out of print, though, but I think you can still pick up some versions of the core set for reasonable prices. Its a game about hackers (runners) stealing corporate secrets in a cyberpunk future. Corp players play cards secretly where positioning matters while the runners need to decide what avenues of attack to pursue with initially limited information.
The music in between sections is an old royalty free sample that is most famously used in GarageBand / Logic as an Apple Loop. For anyone who was looking for it. (It was also used in a well-known indie hit where most of the soundtrack was made of Apple Loops. It's messed up that the soundtrack "composer" accepted an award for it, considering how the majority of it was written by other people…)
Suggestion for the stratego bomb exploding multiple times issue: Imagine the bomb as a minefield. It can kill as many troops as attempt to cross it, unless you have an engineer demine the entire field.
Stratego is one of my favorite two-player games for sure. I saw the notification and I was so happy that one of my favorite channels was doing something related to board games, one of my main hobbies!
33:59 - I played D&D via webcam with friends in the Netherlands ... for a few weeks. It was great while it lasted! They had me on a TV screen and I had them on my computer screen. It was cool ... until someone took a raincheck. Then the next person did. Then the DM had an irl fight with the Monk. Then the Druid had a falling out with practically everyone except the Wizard (his gf), so they both got kicked out of the game by the DM. Looking back, I'm starting to wonder if those game sessions truly took place or whether I hallucinated the whole thing 😩
There's a big overlap between B. F. Skinner's behavior reinforcement and how ML neural networks are structured and trained, thus the use of 'reward' and 'reinforcement'. It sounds like DeepNash is learning in much the same way anything with a nervous system learns, through experience. In the end, DeepNash isn't making decisions based on deduction - it's making the move that 'feels' best. Thank you for the links to the paper and the GitHub project!
Side note: I'm fairly certain that not that long ago Google proved this same thing with chess. Their first (for lack of a better term) ai made use of this neural network stuff (which I only kinda understand for the record), but was manually tuned similar to how other chess engines like Lc0 and Stockfish NNUE are tuned. However, they managed to replicate their previous ai by only giving it the rules, and the've now extended that tech to Go and now (apparently) Stratego. It's been interesting how they've been slowly giving more and more complex and varied challenges, and assuming it's all real paints an interesting picture since last I checked they hadn't worked with a non-determenistic game until this point.
I wrote a Stratego game to play against a Tandy Color Computer in 1980. It used both Milton Bradley and Jumbo rules for the game. The player chose which ruleset before setting up their pieces on the board. I have no idea if the game was saved on a server or not. It could be lost to the ages. But I still remember how I programmed the computer to play the game.
I play D&D online these days cuz my friends are scattered too. It doesn't make it easy to schedule, but it's at least possible. Try Diplomacy if you want a game to play with your 6 adult friends who live in different states. You don't have to schedule a time to get together. Play online with 1 turn = 1 day so you each have all day to submit your moves and then they're all resolved by the computer. The day is spent idly thinking about your next move and texting the players to negotiate before submitting.
The bomb remaining on the field makes sense to me. If what a bomb represents is a minefield then it wouldn't be "resolved" just by going off once. You need a sapper to actually go and clear the area. ...I have no idea why I'm rationalizing the rules of a game I don't play.
Mtg draft us a lot of fun because when you’re building your deck you kind of have a silent negotiation with all the players around you just by looking at what cards they don’t pick. You can also take cards that aren’t good in your deck to deny them from your opponents, though this is usually a bad idea. Most local game stores run drafts so you don’t need 8 free friends
It's great, but it's still a lot more expensive than buying duel decks since you have to buy in every time. Obviously you get to keep the cards, but that's only useful if you're into constructed...
Based on how you described your favorite way to play Magic, I think you would really enjoy KeyForge. It's a newer game made by the creator of Magic, and it's like a TCG without the deckbuilding. You buy premade decks (which are also unique to all others in circulation) and the game is all about learning how to play that deck the best. It also has a built in handicap system for when one deck is advantaged against another, to make it a little more fair.
For my money - Netrunner is Richard Garfield best design. The asymmetry of the two players - the 'consealed' corp player, vs the 'revealed' runner player. The perfect balance between pre-planning strategy of the deckbuilding, the luck of the draw, and poker style bluffing/mind games. And also, the emergent narrative of what actually happens during a game. Hot damn I love Netrunner.
The fact that authors don't clearly spell out what they do in papers sounds like a nightmare. How are you supposed to learn anything as an up-and-coming researcher if everyone hides their work?
On the getting gaming groups together, the main thing is to have a group of people who will block that day off to get together. Usually that means not playing every week. And it means not playing with new parents.
Been sitting on a copy of stratego my uncle gave me for over a decade, guess it's time to finally try it! Edit: Yeah that was fun, definitely putting that in my rotation.
The friend who taught me Stratego has an interesting house rule where the scout can move one or two spaces at a time but it can also jump over other units including bombs
The only way I can get my RPG group together was by setting a date and time that it would be every week and it would be totally online. I then posted and gathered players who would fit that time. With kids and other obligations, there's no way I could get people into my house currently once a week consistently.
Conference proceedings aren't talks, they're the papers about the talks (or I guess the talks are about the papers) Conferences require you to write a 4-15 page paper (depending on the conference) about what you're presenting, which they then publish in the conference proceedings. I only checked a handful of the references at 20:57, but the ones I checked were all papers Edit: Imma be real, I tend to get a bit triggered on the DL videos, Google isn't doing all of the AI. If anything that would be OpenAI, they're the ones that beat the Dota 2 world champs, they did ChatGPT, they did Dall-E
You describe precisely why I don't want to play MTG with my friends despite having fun with optimizing deck-based games. If you like worker placement, Roll for The Galaxy and Caverna are both great 2+-player-games. Also Tabletop Simulator is a great way to play boardgames with friends
I found people to play board games with through Meetup. I attended various board game meetups in my city for a few months and eventually found my kind of board gamers. Now we have a group of 8-12 regulars who meet once week outside of a formal meetup. The trick is showing up consistently for a few months and trying different groups out. The first group I went to really wasn’t my jam.
@@w花b You are correct, the classification of AI is very broad and the learning aspect of more complex algorithms is not a requirement. It ranges from DL and ML to something as simple as a tic tac toe opponent with hardcoded responses in every state of the board.
My favorite casual Magic format is Cube Sealed League. Pick 360 or more cards you like, roughly equally distributed among the colors, and put them in a box. When you want to play, take out the box, shuffle the cards, hand out 80 randomly-chosen cards to each player, and they have to build a 40-card deck from those cards and any number of basic lands. After each game, the loser gets to take an additional 5 random cards from the cube to add to their card pool to improve their deck. Keep going until you decide to start over. People spend a lot of time designing the perfect cubes (I'm guilty), and for people who enjoy that, it's part of the fun. But honestly it's almost as fun to just use a pile of random cards that you think are cool.
I'm so glad you mentioned Magic! Not only because it's one of my favorite games, but because it's one that seems really resistant to being cracked by so-called AI - to my knowledge computers still can't reliably beat human players with a decent grasp of the game. I feel like it makes sense intuitively that this is the case (imperfect information Turing complete game, plus an astronomical number of legal starting configurations and possible legal moves) but it's really cool to feel like I can probably beat even today's computers at something. For what it's worth, if you enjoy the game but don't like the possibility of playing with imbalanced decks, definitely give booster draft a try. Maybe you already have, not sure how into the game you are, but everyone is building out of three fresh packs - no outside cards except basic lands - so it levels that aspect of the playing field. It's a similar experience to your "play new decks with people until we get good at them" thing, but with the fun of getting to customize your deck, and it costs about the same $20 ($15 a person seems to be the number at most game stores).
@@jonathanshor2893 dunno if links work on youtube but paper is just called "magic: the gathering is turing complete" and it's on arxiv, by churchill biderman and herrick
Part of the challenge of MtG is that there are a whole bunch of tasks that are each very hard to accomplish. First the software needs to be able to read a card and understand its rules, something that can be hard even for humans. If we want to it play "like a human" then it would also need to be able to build a deck which is several tasks but we can probably skip this. In game it needs to understand the overall strategy of its own deck and then play an imperfect information game that includes not only understanding all the interactions but also infering both the strategy and likely specific cards an opponent will have based on the cards seen. I would expect TCGs to be one of the last games to fall to computers, I don't think more than a few of these tasks could be accomplished in my lifetime.
I have fond memories of playing Stratego with my dad as a kid :) We had a version with boards where you could play with three or four players!! Fluxx is another game I love too. Some 2-player games you didn't mention that I've enjoyed are: Santorini - you build white houses on a grid, and you've got to get your guy on top of a building before the other person. On top of that, you each draw a Greek god card at the beginning that gives you a special ability, making each game a little different Tak - this is a more abstract game, inspired by a game in the Name of the Wind novel. You aim to connect opposite sides of the board with a road of your slabs (you can stack on top of your own or your opponent's pieces), but you can drastically alter the board state by picking up a stack of slabs and spreading them out along a row. Fun strategy game!
You asked, "How do you get 4 adults together in the same location once a week?", My immediate reaction is, "Go meet people at Friday Night Magic, those people know how to do a thing once a week." You proceed to immediately bring up MTG. Good times.
I think you might like the game coup. It's really low barrier to entry (both financially and emotionally), 2-6 players, and rounds are fast and addictive which makes learning really fun. Sort of like poker or stratego in terms of bluffing and/or social engineering. (I am also not deeply entrenched in the tabletop gaming community so I apologize in advance if I have misrepresented coup or it's no longer cool lol)
Don't discard sources just because they are conference proceedings! :) I'm doing my PhD in Automatic Control and conferences are the preferred way of publishing in my field. Usually it's a full (but limited in scope) paper. Edit: agree with your point about poor sourcing of many machine learning papers though!
@@andrewfleenor7459 True, and I'm not saying these specific ones are good. I don't know. But most conference proceedings are available if you have institutional access, just like journal papers. I mostly reacted to the comment calling them out on being conference proceedings and therefore a summary of an unrecorded talk. Like I said above, conference proceedings in some areas (like this one) are usually standalone papers written to be read without the context of a specific presentation. TLDR: Read as "Don't discard sources JUST because they are conference proceedings." Might still be bad sources though :)
Yeah, I'm a bit confused. For the screenshot at 21:00, I looked up the first three papers simply by the author names and years and had absolutely no issues finding them, both "official" references like Springer and fulltext open access via Researchgate. Sure, it might be slightly irritating to not have a paper title directly from the reference, but in a world where everything is a single Scholar search away, I really don't see the issue. @acollierastro am I misunderstanding the point you're trying to make here? Your subsequent criticisms about reproducibility and reference quality are completely valid, but I have never had problems with the accessibility of conference proceedings as you described.
D&D IS a thing. But, one of the greatest challenges of D&D isn't the rules, or the strategy ... it's the scheduling. I'm currently in a D&D game, but it only meets every other week. We've all sort of decided that Sunday is our "game day". And that seems to work for us. It's not perfect. Sometimes one or two people aren't able to make it, and sometimes we have to cancel. But overall, it works. There are ways to play online. We did it a lot during lockdown, but it's never quite as good. I imagine the same is true of Stratego. A big part of it is the interaction.
For Fluxx I always tell people that the expected duration of the game is about 2-3 more turns from now. The past does not matter in this evaluation, no matter what the board state currently is, you can expect on average that the game will go for another 2-3 turns... or maybe that's the mean or mode...
Speaking of MTG and other TCGs, something very fun (and easily very affordable) is to build what’s called a Cube. Cubes are preconstructed (typically by you, but are also available online) draft sets for two or more players. In a well-designed cube, the games are interesting and fun, but so is the draft. Building towards and playing a specific strategy and trying to guess by the holes in the cards available what people are brewing are both fantastic fun. And as the owner, you can slowly (or dramatically) try and adjust your game experience, makes you feel like your own little game designer at home. Such a great experience, I highly recommend it!
Anything that can be reduced to discrete numbers can be replicated in code. The human element in stratego can be kind of replicated by a decision tree considering optimal moves based on your own deployment. Then you make assumptions like bombs are always around the flag. So you start getting a picture of the other side like a human would. It is an interesting paper but I think stratego is a prime game to be computerized. Again, all the games that can be reduced to discrete arrays of numbers can be easily computerized.
I love your take on things in general, and Machine Learning in particular. The "actual understanding" vs. "how the majority of people (or popular media) like to (or have the capacity to) understand and talk about things" seems to be a perpetually losing one, of course, but those of us who think more like you very much appreciate our perspective being represented. edit: lol it makes me so happy to see you pull out Magic: The Gathering cards at the end. Definitely something I've loved since being into it when it originally came out when I was 12-13... I've gone through long periods of ignoring it, but also always come back to it and buy some packs for a set that looks especially neat (loved Lorwyn/Shadowmoor the most probably of any post-90's sets...) or watch set reviews, watch a Spice8Rack video, boot up Arena for a few days randomly, that kind of thing. It makes me happy that it is still going so strong, but also a little annoyed at the rate at which WotC releases new sets lol. You're so awesome : ) And you have an excellent sense of what true fun is, also
Oh my GOD. I played this with my therapist when I was 9~10yo. I must've played at most 10 total matches over the course of a few months. I _completely_ forgot this game existed until you showed the board, wtf. It's so good!
Hmm, its almost like its got a form of intelligence. I mean, not real intelligence, not human like intelligence, but a sort of... artificial intelligence, an AI you could say... quite accurately... and not need to complain about it... cause its actually a perfectly reasonable term to use... and has been for decades with zero confusion... and the only problem seems to be recent media distortion of the facts which would happen regardless of the terminology... My favourite board game is Pandemic, its not too bad with just two players and has a solo mode of play. Also, if you want to play more MTG without spending a lot you can play the Pauper format. Same basic deck rules but all cards have to be common (Overwhelmingly just a few cents per card.) and can be taken from any set. Deck building can be overwhelming but the internet has plenty of interesting ones.
The paper is also calling it AI, and acknowledges that "artificial intelligence (AI)" has mastered other board games in the past. Deepmind themselves call DeepNash a "Game-playing artificial intelligence (AI)" on their own website. The whole "AI doesn't exist" thing is weird to me because it seems to assume that AGI is the only kind of AI. Google seems to think it's any machine that can reason, learn, and act in such a way that would normally require human intelligence or that involves data whose scale exceeds what humans can analyze.
I made a comment on her other AI video pointing out that she's conflating "intelligence" with "human intelligence". Something can be artificially intelligent without having any kind of self-reflection or theory of mind.
@Falcrist She is knowledgeable in other areas and it's exhausting to constantly make that differentiation instead of leaning back and trust the information. Hope she'll stop talking about ai, that is not what I am here for.
You're confusing performance with intelligence. This AI is a calculator with a fancy spreadsheet - it's a machine model that runs in a rut. If you told it to now play Go it wont be able to do it, you'd get garbage out. AGI might create models for it to use as a tool but it is the hand not the tool. True 'general' artificial intelligence is able to learn new skills: while this model may be able to mime strategy in Stratego, it can't apply what it's learned to another skill such as a different game, a new model would need to be made specifically to play Go. This requires flexibility to take patterns and reapply them to new contexts by attempting to adapt. It's autonomous, so a human wouldn't need to train it for every aspect and determine for it good vs bad actions - it would figure out what the point is and read the rules on its own in order to play and may even understand to purposefully throw a game to let someone win. Its directive would be to amuse someone, not eliminating a particular piece. This requires complex awareness of its environment to gauge purpose. And agency, self-determinism to take actions beyond its initial directives. This requires self-awareness on top of environmental awareness. The machine learning model has to be told what winning is to play, it didn't know what a game or winning is from its understanding of human constructing games for entertainment. This model is a train on tracks that can safely figure how fast it needs to go. True AI can book a vacation and plan your drive along a route that has a good burger joint that you'd enjoy.
@@youtubelisk I don't think she's unknowledgeable in this field, nor do I think she's memeing. In this video she says she's switched her career from physics to machine learning (unless I misunderstood her). She obviously thinks its important to point out that these programs don't have human-level intelligence, but she does it in this way that makes her look extremely biased and unreasonable.
As one of those board game people - I mostly play 2 player games with my spouse. Big 6+ player games don't happen nearly as often as when I was in college. There are 2 ways that I get my group of 4 meeting regularly: 1) I practically drag my friends kicking and screaming into the hobby lol, 2) I was changing careers, and was looking at various job offers. I took a lower paying position because it was in my hometown where my family and most of my friends are. There are a couple other methods I've heard of but don't employ: 3) find a local game store and play with randos there??? Weird. Talk to people I don't know? Not gonna happen. 4) have kids, wait for them to grow up for a while, then make them play with you lol
The "defender doesn't reveal their number" rule isn't just a family one. In my box set the rulebook mentioned it as a potential way to spice the game up.
One of my favorite two-player games is Quarto, which is a kind of tic-tac-toe/connect-4 with a pool of pieces where each piece has four attributes. Each player takes turns giving the other a piece to play, and the goal is to connect a line of four pieces that share any one attribute
Watching the video, becoming interested in playing the game, wandering why I have never played it with friends and coming to the realization that this is what the friends in question would call a "Paulina not sueded"/"ADHD incomparable" game was an experance. So I texted a friend to ask him if we could play this game and his answer was... "I mean we could try, but I do not think that you would enjoy the experience".
Great video! A small remark about conferences in computer science: they are quite different from conferences in other fields. Talks during the conference itself are borderline very low-level advertisements to get people to read the conference paper. Since conference papers have a limited number of pages (and a very short time between submission and publication), this also contributes to having very incremental papers, with usually only a few central new ideas or results, and many things that rely on the reader knowing previous papers because there isn't enough space to present them. Code is considered to be part of the supplementary material, and should be reviewed, usually to check that new ideas are indeed implemented (there is even some events specialized in code). (also there should be a space between the arxiv link of the paper and "DeepNash")
Wingspan! It's an engine building game focused around the birds of the world, beautifully illustrated on cards. You can play with like 1-7 people depending on which of a few modes you play in. The default mode I find most fun with 3-4 people but 2 definitely works and there's a duet mode specifically for 2 people as well though I have not played that. Also there's an online version and the degree of competitive angst is relatively low. Like in the base game there aren't many opportunities to mess someone else up but there's a goal board with sort of casual/sweaty sides and there are small things you can do to negatively influence someones game.
I've heard so many good things about Wingspan, sadly never had the chance to play it. Apparently there's a really cute companion app that plays the bird song of whatever bird card you scan with it
I love Wingspan. I have two of the expansions and the PC game. Its the only "complicated" board game I've gotten my family to play and really enjoy. 99/100
Re: Wingspan and few oppurtunities to "mess someone else up" This aspect of engine-building games is unappealing because it makes the players too focussed on their own games. It gets in the way of discussion and table-talk, which in my opinion is the whole reason to play board games. As if everyone is playing the same game at the same time, but they could be in separate rooms and it wouldn't change anything. For however cliched it's become as one of the first Euro-style board games to gain popularity in Canada/USA, Settlers of Catan is an engine-building game that is very much focussed on the competitive aspect where everyone's engines are intertwined, and it's why I like it so much. It's also relatively simple and somewhat balanced, which makes it good for less experienced players.
@@cmmartti On the other hand, that's kind of the goal of the game. It's not like it's an accidental defect, it's more of a design goal. It's ok if that's not a feature you personally enjoy. I play Catan as well but often I'd rather be in my own head about the game while having a chat not relevant to the game. My friends and I are birders as well so we get a kick out of talking about birds, and my mom won't play a competitive game so it's awesome to play anything with her at all. It fills a niche for a lot of people.
@@semanticboat Very true, I know a lot of people who like it and I will definitely play it if that's what they want to play and I'll have a good time, but it will never be my first pick. I also admire the bird cards, but the details are irrelevant to the gameplay and trying to talk about the cards inevitably pauses the game-so they may as well not be present. Settlers is not my favourite game either but it naturally directs all discussion towards the game currently being played, which I really like.
9:08 still watching the video and this feels woefully unrelated to the important part but I think this segment has convinced me to try this game with some friends despite not knowing what it was 10 minutes ago, my problem with Chess is how everything feels futile since I know I could mathematically be making better plays and it feels like more of a numbers game than strategy to me in that, but the way you describe it as "playing it for the game" rather than playing it for the best move sounds like it fixes all my issues with that set up and sounds really awesome to try out, so, thanks for that lol.
"AI" is certainly the right term in the context of computer games. In that context AI means _whatever_ logic controls the computer's units, nothing more. And as a wise game designer once said: it's not about making the opponents smart, the player is supposed to beat them eventually, it's about avoiding Artificial Stupidity, avoiding them behaving in a jarringly bad way. The ghosts in Pac Man were a great early example of good AI: they seem to work together cleverly to surround the player, but are each just a few simple lines of code.
Yeah, the field as a whole has been called AI since the 60s or so. It is and always has been a very broad term. Anyone who insists on confusing it with AGI cannot be saved by terminology, and creating yet another disconnect between academic and public language wouldn't be worth it anyway.
I highly recommend Tabletop Simulator on Steam. Not only can you play with your out of town friends: you can also play expensive or out of print games (like LOTR Stratego) without having to scrounge up a copy. Sometimes my local friends are so lazy that we play on Tabletop Simulator anyway... but it is a lot easier sometimes to not have to all meet up.
If you're looking for good two-player games, I recommend Hannibal (full name Hannibal & Hamilcar: Rome vs Carthage). It's an asymmetrical two-player wargame which is really fun if you like history. I have stayed up until 4 AM desperately trying to get Scipio Africanus off the board before
Hi, to answer the question: "where do you people even find the time to play boardgames with your friends?" we don't, we go to boardgame clubs, where all the people there LOVE playing 3-hour 4-player games, and we learn new games together, and we make new friends together. I would totally recomend that.
Back in the day my crew had board game parties. We (wife, friends, kid) were bug into a bunch of different ones like Catan, Munchkin (Adventure Time version is the best!) and Stratego! My daughter was young and wanted to play, so Catan became a matching and adding game for her, and it was fun for us having this force of randomness added. For Stratego I made a list of maybe 8 rules for her to set up the board: 1 place flag, 2 protect flag with bombs or strong pieces, 3 place bombs, 4 place miners, 5 protect miners, 6 place scouts, 7 place spy commander and general,8 place all other pieces. We had some really great games! I usually won but she displayed some great strategies based off those rules like flag in corner, bombs on all the front lines and she learned to defend against scouts flying thru the gaps to steal the flag :)
In the original french version of the game, they actually were called "mines" to represent a mine field, and the defusers were actually mine sweepers. That also explains why they don't go away when they explode. Even if a single mine explodes, the mine field is still there.
One variation in Stratego is that in every battle, the spot of the losing piece is occupied by the winning piece, unless the winning piece is a minefield, in which case the minefield stays put. In case of a tie in level number, level 3 (colonel) versus level 3 (colonel) for example, both pieces are destroyed in a battle. In that variation a good initial set up is to have one lakeside path blocked by two minefields, a second lakeside path partially blocked by using one minefield, and the third lakeside path unblocked by mines. This approach leaves 3 mines for protecting your flag, which is good enough if your flag is on an edge, or leaves a spare mine if your flag is in a corner. Wherever your frontline minefields are, you want to have a rank 7 (sergeant) or rank 8 (miner) piece adjacent in a rear area ready to destroy an enemy miner who destroys your minefield, since if you can succeed in destroying all enemy miners, your flag is safe if surrounded by mines, and then the only way for the enemy to win is to defeat all of your pieces with the ability to move. A spy should ordinary be very close to the general, so that if the general is defeated by the enemy field marshal, then the spy can defeat the enemy general on the next turn. In the gaps usually near the frontal minefields, a level 1 or 2 or 3 piece should be nearby in order to be able to stop most strong assaults. An series of level 9 (scout) pieces might be initially laid out in a left to right manner in the second rank, to be intentionally quickly gotten rid of in order to create a strait line lateral path for higher ranking pieces to quickly shift leftward or rightward. Being the first to take a level 3 (colonel) or higher piece usually results in a winning game if you can also strongly deplete the enemy's initial supply of rank 8 (miner) pieces, since you can usually trade down until you have the highest ranking moving piece on the board once you have reasonably good flag safety.
At 28:54 that second piece of information given to the algorithm that a Stratego spy will always attack a known enemy field marshal is not always true. Aside from a blunder, there can be tactical situations such as a perceived "flag rush", during which the delay of an attack toward the opponents perceived flag position is not worth taking the time out from the advance toward that perceived flag position in order to capture a field marshal that is not essential to either stopping one's own flag rush or in speeding up the enemy's flag rush. A better approach therefore is to give the mystery non-attacking piece a hard-coded multiplier of around 0.03 on the possibility of that piece being a spy (unless all other opposing pieces have been absolutely ruled out), and then making mathematical adjustments accordingly on the possibility of other opposing pieces being the spy piece, so that all the individual piece probabilities add up to approximately 1.000 of a spy, in accordance with the game start setting of 1 spy per side.
@@anita.b no, unfortunately i taught her Stratego to use to decide how the divorce settlement would play out but she was too good at the game. She beat me so thoroughly she got all of my stuff and even my google account. this will probably be my last comment and then she'll own this account too.
I used to play Star War Stratego with my dad, so there has always been this, faint, almost whisper of a voice in the back of my head, whenever anyone brings up chess and checkers and it just goes, "I wonder if Stratego's fun?" As I've last played it when I was six. So, thank you so much for this video.
Does it have to be a 50% win rate to be considered a Nash equilibrium? My understanding of it was a Nash equilibrium is reached when both players in the zero-sum game have no incentive to unilaterally change their strategy.
It does not, because games can be asymmetric. But for games like stratego that are symmetric (assuming you flip a coin to decide who goes first), every Nash equilibrium is guaranteed to be 50/50. If you start talking about asymmetric games -- including stratego where the same player always goes first -- then the Nash equilibrium can be biased, with one player only able to win 40% of the time or whatever.
The trick to scheduling D&D is to have a pool of 6-8 people and if you can get at least 4 people you get to play a session
The trick is to schedule the next session at the end of the current session. Everybody closes their books, opens their agenda and nobody leaves until you all write down a date and time and place.
@@garak55 we absolutely do that, but if we didn't run sessions because one person had a multi-week hiatus then we'd play once a year
*at least 3 people in my experience but yea lol
@@garak55 The trick is to have a pool of people and a fixed schedule. This way, most players will treat it as a regular occurrence with more importance than if you schedule it again each time at the end of a session.
In my experience D&D is about having a group of 8-16 people who may or may not show up, and then spending 3 hours deciding what weapons we're bringing. I might be doing this wrong.
I love thinking about the meeting where Google decided that their motto should no longer be "Don't be evil!" Like someone had to say out loud, "We might have to be evil a little bit every once in a while."
I think it was more like, "That's gonna come back and kick us in the butt. We better change it."
They gave up the charade when their CEO said (regarding privacy of users): "If you have something that you don't want anyone to know, maybe you shouldn't be doing it in the first place."
Of course they were gonna be evil. It's what corporate executives do.
The issue of course is the definition of evil. People define it differently and it’s probably a philosophical headache having to debate it all the time.
I mean at that point they'd already dropped the first word from their unwritten motto, going all the way probably wasn't a big step to take.
@@adamstevens5518 Only evil people define it differently
Douglas Adams was a genius. He predicted the notion of a black box algorithmic computer program giving the answer to life, the universe, and everything without supplying the thought process or derived question that led to the result of "42".
When my father was young, he and a friend made up the dumbest variant of stratego imaginable. You flip the pieces so you can only see those of your opponent and not your own (also when placing them, it's not about memorizing what your pieces are). You can move every piece, even if it's a bomb or your flag, because how are you to know you can't move it? The other rules remain largely unchanged. Good luck!
wow
We played this at school. The flag has to be on the back row otherwise you can do a random set up or opponent places your board. It’s not as fun as the real thing though, but it does produce silly things like Bombs taking flags. You also play the landmine rule that bombs disappear when they explode else it’s just a game of who can find their bombs first
Pure chaos
Stratego is decended from the French game 'L'Attaque'. The bomb is actually called a mine there, and it implies a mined field. The miner is called a mineur or sapeur, both imply a deminer unit.
That explains that rule about bombs not getting destroyed after detonation, even if a single piece gets destroyed, the mine field is still there.
Also explains why it can't move, lol
And a descendant of the Chinese game named "Jungle". It is very similar, but with animals, being the elephant the strongest that can be beaten by the weakest, the mouse. But the banner is a fixed place on the board.
a bit "lost in translation" there. that makes way more sense.
miner i hardly know her
You made a 40 minute video expressing strong opinions about stratego. I dunno how to tell you this angela but you're member of the board game community now
Also, Patchwork is a surprisingly fun and cheap 2-play game. It looks like it wouldn't be fun when you read it, but I was very pleased by it - this coming from a big board game person
Patchwork and its variants are awesome
I LOVE the direction of this channel. Came for the physics rants. Stayed for the general STEM rants. (But don't leave the physics stuff, please. You bring up conceptual nuances that many physics classes miss)
… I know physics, sometimes I go gym
i like physics at all because of what she focuses on
There's the silly humor too, .......alkaline water with lemon...
I've come to believe ignorance is the most fundamental force of nature I could totally relate to that frustration.
That was a fantastic video @@petevenuti7355
Just a note about chess (38:20): it's actually completely fine to resign a game when you know you've lost. In fact it's proper chess etiquette to do so if you're totally lost rather than make your opponent go through the motions, though when a game is considered sufficiently lost that it's worth resigning depends heavily on the level of the players. Usually people would rather not have to play another thirty moves to mop up the game after you've blundered a queen.
My favorite is that clip of Carlsen and ... Hikaru maybe? Where one of them like moves a pawn in nowhere, they both stare at the board for a minute and then they go
"Oh, I thought you could ..."
"Right but then ..."
"You're right."
"Congratulations"
And shake hands. They each just finished the game in their minds and agreed who won. Amazing.
@@davidgustavsson4000 that's hilarious. do you have a link to the clip, by chance?
@@WetRatGaming I'm trying really hard to find it, but it had some ungoogleable title like "They don't even need to play!" or something.
@@davidgustavsson4000then there's how i play it, losing queen with ten moves and an endgame consisting of two pieces chasing a king around indefinitely but the opponent is also as bad as me so won't agree a draw.
@@davidgustavsson4000 That sounds you misunderstood that the game was over already. They often do a analyze afterwards (called a Post Mortem), and this analysis could go as you described because both still have the calculated lines in the heads. But they do not talk with each other while the game is ongoing, that would be a serious break of etiquette and rules.
30:30 Underrepresenting Vincent here, he is not just a stratego player, but also an AI researcher. (His sister Roseline, who was women's world champion, is a friend of mine.)
God damn it, there are Stratego world championships?!
Welp, turns out I've wasted my life on frivilous, non-Stratego related pursuits.. I AM A FOOL 😭
That’s so cool!
He’s the most important character in this video! Also please do not show Roseline de Boer this video. I will be very embarrassed of my fangirling all over this game.
@@idontwantahandlethough There's an International Stratego Federation and even a whole Mind Sports Olympiad (MSO) that includes Stratego. BTW, Demis Hassabis (DeepMind co-founder and a senior author of this paper) has won the MSO Pentamind (best all-round board games player) five times. So yeah, "these machine learning experts" may _actually_ be interested in Stratego.
Wait ... WOMEN'S world champion? The championship divisions are segregated by gender?
Why? Do they use 100 lb. playing pieces that are harder for the average woman to lift?
You don't have to watch your friend clumsily win against you after you made a mistake. It is part of the culture and etiquette to simply resign a game that is unrecoverable (or at least, very unlikely to be recovered). It's a sign of respecting both their time and skill to not force them to prove they can win a one-sided board. :)
That's apparently the opposite of league of legends, wherein you apparently do not respect your opponent at all and in fact resent and hate them for playing the game at all.
absolutely not, you will respect me by sitting there and suffering through your slow and agonizing defeat 😂
@@stuartp2006 FFs are for losers
@@stuartp2006 Only your opponent? I thought you were supposed to hate everybody else who's ever touched the game before.
@@Norade True. League players don't hate their opponents. They hate their team mates with a passion unmatched. Ima go play some TFT now.
I only have one solid memory of this game, though I remember loving it. I attacked with like a nine or an eight, my friend told me theirs was a ten, but I remembered their ten being on the other side of the board so I turned their piece over. I was right, he was lying. Then he got mad at me for checking and tossed the board. I had some odd friendships as a kid.
I had a friend that used to always play Sonic on his computer. But then when it was my turn to play, he'd always say the computer "needed a break". He also flipped board games a lot. What a guy 🙄
People being bad at a form of entertainment, cheating badly and then getting upset when caught is baffling.
@@hayuseen6683
Same as a criminal mindset. They want to get away with things illegally and when they get caught, they get mad and want a lawyer. 😂
@@Bassotronics There is so much wrong with your comment, I don't even know where to begin.
@@Huntracony
E=MC Hammer
We've been calling the computer side of a video game AI for a long time.
this is the only case where I will call machine learning AI. it is more similar to the little man in nintendo 64 mario kart than the terminator
Loterally tho. I wonder why some people completely avoid that use of the word.
Of course, the most common usage of this in video games is "The AI is so [stupid, easy to game, inefffective]" so maybe *Intelligence* is a poor choice for this.
@@jfirth816 I keep wondering where people draw the line of intelligence if we look at animals rather than computers... If an algorithm is considered unintelligent because of criterias we would likely not apply to animals, is it that animals are not actually intelligent or that our algorithms are intelligent but we don't like to think about it?
@@Nathouuuutheone While I'm not sure why you're asking a guy who was riffing about computer games this, but the definition of "intelligent" would be the first thing to address to make any headway at all. Even my limited knowledge won't fit in a RUclips comment, but I believe that a lot of folks who think about these things would include "self awareness" of "intelligence" or more specifically "human intelligence" That's something we know humans can do and we can argue (or research, but mostly argue) whether or not animals can do it. But we wouldn't argue about computer algorithms having self awareness. Humans write these and don't put in self awareness, so they are not intelligent. They are algorithms. I think this captures the gist of why we can argue about it with animals but flatly state that computers are not intelligent.
38:15 Resigning from a losing position in chess isn't generally seen as bad manners, and shorter clocktimes can help out with this as well! My move-accuracy in losing positions is *way* higher than in winning ones so i'm a bit of a Clock Fiend
In hundreds of World Chess Championship matches, there has only been a single checkmate, and that happened in 1929. There were two stalemates in 1978 and 2007. Every other game had ended in either draws or resignations.
Resigning is not only acceptable, it is expected. I've played lots of chess online. Every one of my games ended in resignations (because I'm a lousy player, at higher levels, draws are more common. Plus I play a lot of blitz 😊 ).
Since it’s impossible to get people to stop saying AI as shorthand for machine learning, maybe we can slowly get people to understand AI (in that context) & AGI (what they think AI is) are different things.
Robert Miles has an AI Safety channel here; he has a good handle on it all, and perhaps a more effective strategy for "I can't be bothered / you know what I meant" people is to just suggest his channel for more info. This gives them a chance to pick it up naturally.
AI has always been referred to as AI. The computer player in Civilization I had "AI", it was just limited. It was also heuristic rather than neural network, because this was 1990. Heck, computers used to be referred to "artificial thinking machines".
"Artificial General Intelligence" is what all the hype lately about, and what we don't have. There is a qualitative difference between older AI technologies and more recent machine learning techniques, but we are still just talking about better Artificial Narrow Intelligence. The computer program can be trained in a particular task. Also, it's become a field in itself rather than just some automated rules for your CompSci tic-tac-toe exercise.
It's all just semantics, but technically ANI and AGI are both AI.
I was in school when they called this stuff Expert Systems, and AI in coding was basically some blocks moving around the screen deciding whether they should run from or "eat" other blocks :)
@@TevelDrinkwater This. Just because some folks can't distinguish between AI and AGI doesn't mean that AI doesn't exist.
I don't get the impression that most people who play video games mistake the "AI" in games for AGI. In most games, general opinion of AI's capability to play it is quite low, and players routinely find loopholes in algorithms to exploit...knowing full well the AI won't update with an expectation of the player attempting to repeat those exploits. To me, this presents at least intuitive understanding that "game AI =/= AGI" and that they're playing against an algorithm made by the devs, even if one doesn't know the term AGI.
Machine learning has made some of these algorithms nearly insurmountable to humans though, even in games with incomplete information. APM-capped AlphaStar is a great example of that...despite being limited to "visible" information to players and APM far lower than pro players, it can still crush pro players. Even in this case, most random internet players don't mistake AlphaStar for something that can write a resume for you.
I suppose the authors felt the availability of the source warranted having only a surface-level description in the paper. But, I'm glad to hear someone who reads and navigates lots of papers share my complaint that machine learning papers tend to cite sources that are unreachable and/or papers all the way down. It's not just me! (or you!)
14:46 "I've been playing this game since I was a kid"
(fifteen minutes later)
_"Oh my god, strategy. Stratego. I get it"._
Actually, 'strategos' (στρατηγός) was Greek for a general.
Scheduling D&D is definitely like herding cats.
The best way is to incentivize them with food, whether it's players or cats.
Sometimes it seems like I can't get a group together, but weblately I'm up to two alternating campaigns on Sundays and another game night on Tuesdays that I'm hoping to leverage into another campaign.
So my choice is between eating players or eating cats?
@@williamfrederickiii1683 ideally you'd feed the players to the cats. 😂
My granddad had the game generals, basically like stratego with no mines or lakes. All ranks move the same tho, and there’s a simple computer built into the side of the board that reads notches carved on the bottom of the pieces. The attacking and defending piece preserve their anonymity (to a degree). It’s a super fun game and a very early computer augmented board game!!!
"It's papers all the way down" sent a spooky feeling down my spine.
One problem in the CS world is that things just move too fast - everyone cites conference papers because by the time a concept has been published in a full journal it's likely already been iterated on and 1-upped in a conference. That's why most citations are for conferences.
Or that's what I've heard my colleagues say, I'm doing my phd in EE but with a heavy CS crossover
It's more than just that though - CS conferences are archival publications, just like journals in other fields. They aren't talks, they are papers, acompanied by talks, but what is being cited is the paper... I found free full text versions of the first 3 papers @acollierastro "cannot read" in 5 minutes. Many CS conferences have higher impact factors than all but the very best journals in CS. For someone to claim they read ML papers as part of their job not to know that conferences publish the full text of papers in CS is crazy...
Little note about chess puzzles: for most of them, it’s not about getting a win a higher percentage of the time. if you make the right moves for the puzzle, it’s a guaranteed win. They usually involve “forcing” moves where the other player has no choice but to defend or move their king in an unideal way.
Of course there are other types of puzzles that have different goals that might just improve your position… those are annoying lol
I really enjoy cooperative board games, because they can still be tense, stressful, and challenging, without creating bad blood between players. My favorite is Spirit Island, where players take the role of guardian deities who use magic to protect an island from colonial invaders
"I'M THE ARBITER OF FUN" the moment when she realizes people are having the wrong kind of fun with computers instead of mind games cracked me up
I love this channel.
She does not post very often but when she does, I’m all ears. Always interesting information and points of view on the subjects.
I think 2 posts per month seems like a lot
Now I'm going to get out my vintage 1970's Stratego for a few sessions this weekend with the kids. The Blue and Red pieces, with their silver and gold embossing, are like visual candy.
I’m sure you have discovered by now that the AI/ML field primarily publishes at a series of headline conferences there isn’t really a journal culture beyond preprints.
For anyone confused, all the papers highlighted at 20:55 that reference Q learning and R-NaD are of course available online and are not talks! I recommend the papers by Tuyls et al for the roots of the approach.
You will become quickly annoyed by how hard it is to replicate many papers because of a lack of information. Many famous papers have quite large discrepancies between what the paper states they do and what their released code actually does! Much of the nitty gritty details are seemingly intentionally left out sometimes especially when it comes to exactly how the model architecture is initialised, pretrained, etc. it’s a problem that has been rampant for more than a decade and unfortunately seems only set to get worse as models get enormous and require enormous compute; why tell us how it works when nobody can economically replicate… /rant
2:20 i've literally never heard of this game before but i didn't even make it 5 minutes in without furiously googling it, asking my older brother if he's ever played it, and checking if there's a good online multiplayer version of it to play with friends overseas. So. you could say you have succeeded. :B
Have you tried mindfulness? To calm down the furious Googling.
My fiancée and I started playing Stratego last year and it’s been a lot of fun. I began playing with my strategy from when I played as a kid (surround your flag with bombs), but she was new to the game. I won the first few rounds but then she figured out a much better setup that I just couldn’t match (place your flag randomly on the back row to make the opponent guess where it is). Eventually I had to adopt her setup and now we are very evenly matched. I didn’t know there was a competitive scene so I’ll have to check it out now
I'm on the bus going to a game night right now! We have a biweekly hobby circle with an average of 4 participants. A big reason for this to be possible is that some of us had to move back to our home town after college/university due to unemployment, and then we got jobs in the local area. So the silver lining of temporary joblessness turned out to be a sizable game group 🥳
Side note: try Fugitive, it is a very fun 2-player card game of bluffing and deduction.
upvote for Fugitive! and Firefly Fluxx is the Fluxxiest Fluxx.
This brings me back to my days in a "gifted program" during grade school where one day a week the gifted students would go to their own classes that were often about more fun things while still learning(like how to design a bridge or program a Lego robot or just doing research on a country). In my homeroom there was a copy of Stratego that we would always play before class started. It almost reminded me of Ender's Game where they have all of these smart students separated and playing wargames.
DeepNash's ability to trade off information vs pieces is mind boggling to me. It is so willing to sacrifice pieces just to keep the opponent in the dark, then come back and win.
Sounds reminiscent of how Alpha Zero would sacrifice pawns/pieces to give itself more options in chess
The video is a year old so I doubt anyone will see this but,
if I were to recommend a game I would recommend my favourite board game, BattleTech.
I don't know much about Warhammer 40K but I have heard it compared to that.
It's got all kinds of playable vehicles in it, but it's mainly about the mechs which it has a lot of.
It's a wargame that is played on a hex grid (Although there are rules for playing in a 3d map without hexes if you wanna build your own homemade 3d map) where you control a force of mechs and or other vehicles like tanks and vtols and even aerospace fighters and space ships.
There are a few box sets, I would recommend A Game of Armored Combat it gives you a lot of rules to use, and a decent amount of miniatures.
The beginner box is cheaper, but has more basic and simplified rules, and 2 minis instead of 8.
There is a large expansive lore, if you ever get interested in that I would recommend the youtube channel BPL or Black Pants Legion.
They have a series called Tex Talks BattleTech. But I would specifically recommend their video BattleTech 101: The Great Houses of the Inner Sphere to start off.
The style of playing MTG you’re describing at the end of the video actually has a name! It’s called ‘Jumpstart’ and it’s a really popular way to play. Basically the only way I play MTG is on a game called ‘Tabletop simulator’, it’s less than $40 iirc and is just a really good and easy to understand platform for simulating a board game table. I usually load up a bunch of random booster packs Pull two random starters and play them blind, super fun.
As a chess player I love it when my opponent resigns, its not rude at all, in fact its often considered polite, though that depends on the position. I think a lot of the fun and beauty in chess comes from that idea of every game being almost always completely different. Of the vast amount of different moves and decisions possible for each player, you're almost always in a brand new position you've never seen before by move 15. Calculating these new positions in real time against an opponent doing the same can be quite exhilarating, especially in very complicated positions, and is one of many reasons I fell in love with the game. :)
Me and my sister created tons of variations for this game when we were kids. Our favorite version was "reverse mode", in which you would randomly set up your army, and you could only see your opponents army. So you could actually move your bombs around without knowing it (but they were removed afterwards)
Never played Stratego, but the hidden information aspect of it is very compelling to me. Sounds similar to a card game I absolutely adore - Android: Netrunner. If you want a 2 player game combing aspects of Magic: The Gathering style card games with hidden information on the board (with the added twist that its asymmetric and both players play by different rules) I'd suggest checking it out. It is out of print, though, but I think you can still pick up some versions of the core set for reasonable prices.
Its a game about hackers (runners) stealing corporate secrets in a cyberpunk future. Corp players play cards secretly where positioning matters while the runners need to decide what avenues of attack to pursue with initially limited information.
Just when you think this girl can't get much cooler and more informative, you find out she's into Stratego as well. 👏👏👏
The music in between sections is an old royalty free sample that is most famously used in GarageBand / Logic as an Apple Loop. For anyone who was looking for it.
(It was also used in a well-known indie hit where most of the soundtrack was made of Apple Loops. It's messed up that the soundtrack "composer" accepted an award for it, considering how the majority of it was written by other people…)
Suggestion for the stratego bomb exploding multiple times issue: Imagine the bomb as a minefield. It can kill as many troops as attempt to cross it, unless you have an engineer demine the entire field.
Stratego is one of my favorite two-player games for sure. I saw the notification and I was so happy that one of my favorite channels was doing something related to board games, one of my main hobbies!
33:59 - I played D&D via webcam with friends in the Netherlands ... for a few weeks. It was great while it lasted! They had me on a TV screen and I had them on my computer screen. It was cool ... until someone took a raincheck. Then the next person did. Then the DM had an irl fight with the Monk. Then the Druid had a falling out with practically everyone except the Wizard (his gf), so they both got kicked out of the game by the DM.
Looking back, I'm starting to wonder if those game sessions truly took place or whether I hallucinated the whole thing 😩
There's a big overlap between B. F. Skinner's behavior reinforcement and how ML neural networks are structured and trained, thus the use of 'reward' and 'reinforcement'. It sounds like DeepNash is learning in much the same way anything with a nervous system learns, through experience. In the end, DeepNash isn't making decisions based on deduction - it's making the move that 'feels' best.
Thank you for the links to the paper and the GitHub project!
Oh no, not Stratego too
-A chess player
Side note: I'm fairly certain that not that long ago Google proved this same thing with chess. Their first (for lack of a better term) ai made use of this neural network stuff (which I only kinda understand for the record), but was manually tuned similar to how other chess engines like Lc0 and Stockfish NNUE are tuned. However, they managed to replicate their previous ai by only giving it the rules, and the've now extended that tech to Go and now (apparently) Stratego. It's been interesting how they've been slowly giving more and more complex and varied challenges, and assuming it's all real paints an interesting picture since last I checked they hadn't worked with a non-determenistic game until this point.
@@LazarNaskov yes, chess definitely also has this type of engine. Seems to have plateau'd somewhat in ability though after an initial rise.
I wrote a Stratego game to play against a Tandy Color Computer in 1980. It used both Milton Bradley and Jumbo rules for the game. The player chose which ruleset before setting up their pieces on the board. I have no idea if the game was saved on a server or not. It could be lost to the ages. But I still remember how I programmed the computer to play the game.
I play D&D online these days cuz my friends are scattered too. It doesn't make it easy to schedule, but it's at least possible.
Try Diplomacy if you want a game to play with your 6 adult friends who live in different states. You don't have to schedule a time to get together. Play online with 1 turn = 1 day so you each have all day to submit your moves and then they're all resolved by the computer. The day is spent idly thinking about your next move and texting the players to negotiate before submitting.
The bomb remaining on the field makes sense to me. If what a bomb represents is a minefield then it wouldn't be "resolved" just by going off once. You need a sapper to actually go and clear the area.
...I have no idea why I'm rationalizing the rules of a game I don't play.
Mtg draft us a lot of fun because when you’re building your deck you kind of have a silent negotiation with all the players around you just by looking at what cards they don’t pick. You can also take cards that aren’t good in your deck to deny them from your opponents, though this is usually a bad idea. Most local game stores run drafts so you don’t need 8 free friends
Finding lanes, reading signals, planting flags, etc.
Gumption
If one can't afford lotr board game, one def can't afford mtg. Ha
It's great, but it's still a lot more expensive than buying duel decks since you have to buy in every time. Obviously you get to keep the cards, but that's only useful if you're into constructed...
@@chrisprice8112 Or you build a cube.
Based on how you described your favorite way to play Magic, I think you would really enjoy KeyForge. It's a newer game made by the creator of Magic, and it's like a TCG without the deckbuilding. You buy premade decks (which are also unique to all others in circulation) and the game is all about learning how to play that deck the best. It also has a built in handicap system for when one deck is advantaged against another, to make it a little more fair.
For my money - Netrunner is Richard Garfield best design.
The asymmetry of the two players - the 'consealed' corp player, vs the 'revealed' runner player. The perfect balance between pre-planning strategy of the deckbuilding, the luck of the draw, and poker style bluffing/mind games. And also, the emergent narrative of what actually happens during a game.
Hot damn I love Netrunner.
The fact that authors don't clearly spell out what they do in papers sounds like a nightmare. How are you supposed to learn anything as an up-and-coming researcher if everyone hides their work?
On the getting gaming groups together, the main thing is to have a group of people who will block that day off to get together. Usually that means not playing every week. And it means not playing with new parents.
Been sitting on a copy of stratego my uncle gave me for over a decade, guess it's time to finally try it!
Edit: Yeah that was fun, definitely putting that in my rotation.
The friend who taught me Stratego has an interesting house rule where the scout can move one or two spaces at a time but it can also jump over other units including bombs
The only way I can get my RPG group together was by setting a date and time that it would be every week and it would be totally online. I then posted and gathered players who would fit that time. With kids and other obligations, there's no way I could get people into my house currently once a week consistently.
Conference proceedings aren't talks, they're the papers about the talks (or I guess the talks are about the papers)
Conferences require you to write a 4-15 page paper (depending on the conference) about what you're presenting, which they then publish in the conference proceedings.
I only checked a handful of the references at 20:57, but the ones I checked were all papers
Edit: Imma be real, I tend to get a bit triggered on the DL videos, Google isn't doing all of the AI. If anything that would be OpenAI, they're the ones that beat the Dota 2 world champs, they did ChatGPT, they did Dall-E
You describe precisely why I don't want to play MTG with my friends despite having fun with optimizing deck-based games. If you like worker placement, Roll for The Galaxy and Caverna are both great 2+-player-games. Also Tabletop Simulator is a great way to play boardgames with friends
I found people to play board games with through Meetup. I attended various board game meetups in my city for a few months and eventually found my kind of board gamers. Now we have a group of 8-12 regulars who meet once week outside of a formal meetup.
The trick is showing up consistently for a few months and trying different groups out. The first group I went to really wasn’t my jam.
As someone in the AI field, I appreciate your humility in admitting the battle to stop cashing it AI is lost, and your continued resolve to fight on
the ML field* ;)
@@jinxedpenguin Both you and Dr. Collier need to learn how to google simple definitions before planting a flag. It'll serve you well in life.
@@w花b You are correct, the classification of AI is very broad and the learning aspect of more complex algorithms is not a requirement. It ranges from DL and ML to something as simple as a tic tac toe opponent with hardcoded responses in every state of the board.
My three most frequently played two player games (that are specifically only for two players):
Four in a Row
Chess
One Deck Dungeon
Im buying an alcoholic beverage then watching this video. The end.
Total vibes from this channel.
I found this comment while drinking an alcoholic beverage
My favorite casual Magic format is Cube Sealed League. Pick 360 or more cards you like, roughly equally distributed among the colors, and put them in a box. When you want to play, take out the box, shuffle the cards, hand out 80 randomly-chosen cards to each player, and they have to build a 40-card deck from those cards and any number of basic lands. After each game, the loser gets to take an additional 5 random cards from the cube to add to their card pool to improve their deck. Keep going until you decide to start over.
People spend a lot of time designing the perfect cubes (I'm guilty), and for people who enjoy that, it's part of the fun. But honestly it's almost as fun to just use a pile of random cards that you think are cool.
This video made me think of magic as well. I love it very much so.
I'm so glad you mentioned Magic! Not only because it's one of my favorite games, but because it's one that seems really resistant to being cracked by so-called AI - to my knowledge computers still can't reliably beat human players with a decent grasp of the game. I feel like it makes sense intuitively that this is the case (imperfect information Turing complete game, plus an astronomical number of legal starting configurations and possible legal moves) but it's really cool to feel like I can probably beat even today's computers at something.
For what it's worth, if you enjoy the game but don't like the possibility of playing with imbalanced decks, definitely give booster draft a try. Maybe you already have, not sure how into the game you are, but everyone is building out of three fresh packs - no outside cards except basic lands - so it levels that aspect of the playing field. It's a similar experience to your "play new decks with people until we get good at them" thing, but with the fun of getting to customize your deck, and it costs about the same $20 ($15 a person seems to be the number at most game stores).
Have a ref for MTG being Turing complete? I'm not seeing how that could fit.
@@jonathanshor2893 dunno if links work on youtube but paper is just called "magic: the gathering is turing complete" and it's on arxiv, by churchill biderman and herrick
Part of the challenge of MtG is that there are a whole bunch of tasks that are each very hard to accomplish. First the software needs to be able to read a card and understand its rules, something that can be hard even for humans. If we want to it play "like a human" then it would also need to be able to build a deck which is several tasks but we can probably skip this. In game it needs to understand the overall strategy of its own deck and then play an imperfect information game that includes not only understanding all the interactions but also infering both the strategy and likely specific cards an opponent will have based on the cards seen. I would expect TCGs to be one of the last games to fall to computers, I don't think more than a few of these tasks could be accomplished in my lifetime.
I have fond memories of playing Stratego with my dad as a kid :) We had a version with boards where you could play with three or four players!! Fluxx is another game I love too.
Some 2-player games you didn't mention that I've enjoyed are:
Santorini - you build white houses on a grid, and you've got to get your guy on top of a building before the other person. On top of that, you each draw a Greek god card at the beginning that gives you a special ability, making each game a little different
Tak - this is a more abstract game, inspired by a game in the Name of the Wind novel. You aim to connect opposite sides of the board with a road of your slabs (you can stack on top of your own or your opponent's pieces), but you can drastically alter the board state by picking up a stack of slabs and spreading them out along a row. Fun strategy game!
You asked, "How do you get 4 adults together in the same location once a week?", My immediate reaction is, "Go meet people at Friday Night Magic, those people know how to do a thing once a week."
You proceed to immediately bring up MTG. Good times.
I think you might like the game coup. It's really low barrier to entry (both financially and emotionally), 2-6 players, and rounds are fast and addictive which makes learning really fun. Sort of like poker or stratego in terms of bluffing and/or social engineering.
(I am also not deeply entrenched in the tabletop gaming community so I apologize in advance if I have misrepresented coup or it's no longer cool lol)
Don't discard sources just because they are conference proceedings! :) I'm doing my PhD in Automatic Control and conferences are the preferred way of publishing in my field. Usually it's a full (but limited in scope) paper.
Edit: agree with your point about poor sourcing of many machine learning papers though!
You can't really discard a source if you can't access it at all.
@@andrewfleenor7459 True, and I'm not saying these specific ones are good. I don't know. But most conference proceedings are available if you have institutional access, just like journal papers.
I mostly reacted to the comment calling them out on being conference proceedings and therefore a summary of an unrecorded talk. Like I said above, conference proceedings in some areas (like this one) are usually standalone papers written to be read without the context of a specific presentation.
TLDR: Read as "Don't discard sources JUST because they are conference proceedings." Might still be bad sources though :)
Yeah, I'm a bit confused. For the screenshot at 21:00, I looked up the first three papers simply by the author names and years and had absolutely no issues finding them, both "official" references like Springer and fulltext open access via Researchgate.
Sure, it might be slightly irritating to not have a paper title directly from the reference, but in a world where everything is a single Scholar search away, I really don't see the issue.
@acollierastro am I misunderstanding the point you're trying to make here? Your subsequent criticisms about reproducibility and reference quality are completely valid, but I have never had problems with the accessibility of conference proceedings as you described.
D&D IS a thing. But, one of the greatest challenges of D&D isn't the rules, or the strategy ... it's the scheduling. I'm currently in a D&D game, but it only meets every other week. We've all sort of decided that Sunday is our "game day". And that seems to work for us. It's not perfect. Sometimes one or two people aren't able to make it, and sometimes we have to cancel. But overall, it works.
There are ways to play online. We did it a lot during lockdown, but it's never quite as good. I imagine the same is true of Stratego. A big part of it is the interaction.
For Fluxx I always tell people that the expected duration of the game is about 2-3 more turns from now. The past does not matter in this evaluation, no matter what the board state currently is, you can expect on average that the game will go for another 2-3 turns... or maybe that's the mean or mode...
"i know i've lost the terminology war but i still won't shut up about it" "ok let's get you to bed grandma" is SO relatable
Speaking of MTG and other TCGs, something very fun (and easily very affordable) is to build what’s called a Cube. Cubes are preconstructed (typically by you, but are also available online) draft sets for two or more players. In a well-designed cube, the games are interesting and fun, but so is the draft. Building towards and playing a specific strategy and trying to guess by the holes in the cards available what people are brewing are both fantastic fun. And as the owner, you can slowly (or dramatically) try and adjust your game experience, makes you feel like your own little game designer at home. Such a great experience, I highly recommend it!
The realization when saying strategy at loud, pure gold!!
I really like the real time story time, I really like your mind 👏👏
I love your videos so much!
Can't wait to have the time later today to watch it.
Thank you Dr Callier for making great content.
Anything that can be reduced to discrete numbers can be replicated in code. The human element in stratego can be kind of replicated by a decision tree considering optimal moves based on your own deployment. Then you make assumptions like bombs are always around the flag. So you start getting a picture of the other side like a human would. It is an interesting paper but I think stratego is a prime game to be computerized. Again, all the games that can be reduced to discrete arrays of numbers can be easily computerized.
I love your take on things in general, and Machine Learning in particular. The "actual understanding" vs. "how the majority of people (or popular media) like to (or have the capacity to) understand and talk about things" seems to be a perpetually losing one, of course, but those of us who think more like you very much appreciate our perspective being represented.
edit: lol it makes me so happy to see you pull out Magic: The Gathering cards at the end. Definitely something I've loved since being into it when it originally came out when I was 12-13... I've gone through long periods of ignoring it, but also always come back to it and buy some packs for a set that looks especially neat (loved Lorwyn/Shadowmoor the most probably of any post-90's sets...) or watch set reviews, watch a Spice8Rack video, boot up Arena for a few days randomly, that kind of thing. It makes me happy that it is still going so strong, but also a little annoyed at the rate at which WotC releases new sets lol. You're so awesome : )
And you have an excellent sense of what true fun is, also
Oh my GOD. I played this with my therapist when I was 9~10yo. I must've played at most 10 total matches over the course of a few months. I _completely_ forgot this game existed until you showed the board, wtf. It's so good!
This channel is quickly becoming one of my all time favorites. These videos just keep getting better and better no matter what the topic is.
I'm so glad you started a RUclips channel. I always learn things.
Hmm, its almost like its got a form of intelligence. I mean, not real intelligence, not human like intelligence, but a sort of... artificial intelligence, an AI you could say... quite accurately... and not need to complain about it... cause its actually a perfectly reasonable term to use... and has been for decades with zero confusion... and the only problem seems to be recent media distortion of the facts which would happen regardless of the terminology...
My favourite board game is Pandemic, its not too bad with just two players and has a solo mode of play. Also, if you want to play more MTG without spending a lot you can play the Pauper format. Same basic deck rules but all cards have to be common (Overwhelmingly just a few cents per card.) and can be taken from any set.
Deck building can be overwhelming but the internet has plenty of interesting ones.
I’ve been loving how often you put out videos! But don’t get burnt out!
The paper is also calling it AI, and acknowledges that "artificial intelligence (AI)" has mastered other board games in the past. Deepmind themselves call DeepNash a "Game-playing artificial intelligence (AI)" on their own website.
The whole "AI doesn't exist" thing is weird to me because it seems to assume that AGI is the only kind of AI. Google seems to think it's any machine that can reason, learn, and act in such a way that would normally require human intelligence or that involves data whose scale exceeds what humans can analyze.
I made a comment on her other AI video pointing out that she's conflating "intelligence" with "human intelligence". Something can be artificially intelligent without having any kind of self-reflection or theory of mind.
@@APaleDot I think she's meming. I'm not sure. It just bothers me irrationally.
@Falcrist She is knowledgeable in other areas and it's exhausting to constantly make that differentiation instead of leaning back and trust the information. Hope she'll stop talking about ai, that is not what I am here for.
You're confusing performance with intelligence. This AI is a calculator with a fancy spreadsheet - it's a machine model that runs in a rut. If you told it to now play Go it wont be able to do it, you'd get garbage out. AGI might create models for it to use as a tool but it is the hand not the tool.
True 'general' artificial intelligence is able to learn new skills: while this model may be able to mime strategy in Stratego, it can't apply what it's learned to another skill such as a different game, a new model would need to be made specifically to play Go. This requires flexibility to take patterns and reapply them to new contexts by attempting to adapt.
It's autonomous, so a human wouldn't need to train it for every aspect and determine for it good vs bad actions - it would figure out what the point is and read the rules on its own in order to play and may even understand to purposefully throw a game to let someone win. Its directive would be to amuse someone, not eliminating a particular piece. This requires complex awareness of its environment to gauge purpose.
And agency, self-determinism to take actions beyond its initial directives. This requires self-awareness on top of environmental awareness.
The machine learning model has to be told what winning is to play, it didn't know what a game or winning is from its understanding of human constructing games for entertainment.
This model is a train on tracks that can safely figure how fast it needs to go. True AI can book a vacation and plan your drive along a route that has a good burger joint that you'd enjoy.
@@youtubelisk
I don't think she's unknowledgeable in this field, nor do I think she's memeing. In this video she says she's switched her career from physics to machine learning (unless I misunderstood her).
She obviously thinks its important to point out that these programs don't have human-level intelligence, but she does it in this way that makes her look extremely biased and unreasonable.
As one of those board game people - I mostly play 2 player games with my spouse. Big 6+ player games don't happen nearly as often as when I was in college. There are 2 ways that I get my group of 4 meeting regularly: 1) I practically drag my friends kicking and screaming into the hobby lol, 2) I was changing careers, and was looking at various job offers. I took a lower paying position because it was in my hometown where my family and most of my friends are. There are a couple other methods I've heard of but don't employ: 3) find a local game store and play with randos there??? Weird. Talk to people I don't know? Not gonna happen. 4) have kids, wait for them to grow up for a while, then make them play with you lol
The "defender doesn't reveal their number" rule isn't just a family one. In my box set the rulebook mentioned it as a potential way to spice the game up.
One of my favorite two-player games is Quarto, which is a kind of tic-tac-toe/connect-4 with a pool of pieces where each piece has four attributes. Each player takes turns giving the other a piece to play, and the goal is to connect a line of four pieces that share any one attribute
Watching the video, becoming interested in playing the game, wandering why I have never played it with friends and coming to the realization that this is what the friends in question would call a "Paulina not sueded"/"ADHD incomparable" game was an experance. So I texted a friend to ask him if we could play this game and his answer was... "I mean we could try, but I do not think that you would enjoy the experience".
I think you'd like it honestly, you might be surprised! I'm pretty ADHD too, but Stratego is one of my fav board games of all time :D
Great video!
A small remark about conferences in computer science: they are quite different from conferences in other fields. Talks during the conference itself are borderline very low-level advertisements to get people to read the conference paper.
Since conference papers have a limited number of pages (and a very short time between submission and publication), this also contributes to having very incremental papers, with usually only a few central new ideas or results, and many things that rely on the reader knowing previous papers because there isn't enough space to present them.
Code is considered to be part of the supplementary material, and should be reviewed, usually to check that new ideas are indeed implemented (there is even some events specialized in code).
(also there should be a space between the arxiv link of the paper and "DeepNash")
Wingspan! It's an engine building game focused around the birds of the world, beautifully illustrated on cards. You can play with like 1-7 people depending on which of a few modes you play in. The default mode I find most fun with 3-4 people but 2 definitely works and there's a duet mode specifically for 2 people as well though I have not played that.
Also there's an online version and the degree of competitive angst is relatively low. Like in the base game there aren't many opportunities to mess someone else up but there's a goal board with sort of casual/sweaty sides and there are small things you can do to negatively influence someones game.
I've heard so many good things about Wingspan, sadly never had the chance to play it. Apparently there's a really cute companion app that plays the bird song of whatever bird card you scan with it
I love Wingspan. I have two of the expansions and the PC game. Its the only "complicated" board game I've gotten my family to play and really enjoy. 99/100
Re: Wingspan and few oppurtunities to "mess someone else up"
This aspect of engine-building games is unappealing because it makes the players too focussed on their own games. It gets in the way of discussion and table-talk, which in my opinion is the whole reason to play board games. As if everyone is playing the same game at the same time, but they could be in separate rooms and it wouldn't change anything.
For however cliched it's become as one of the first Euro-style board games to gain popularity in Canada/USA, Settlers of Catan is an engine-building game that is very much focussed on the competitive aspect where everyone's engines are intertwined, and it's why I like it so much. It's also relatively simple and somewhat balanced, which makes it good for less experienced players.
@@cmmartti On the other hand, that's kind of the goal of the game. It's not like it's an accidental defect, it's more of a design goal. It's ok if that's not a feature you personally enjoy. I play Catan as well but often I'd rather be in my own head about the game while having a chat not relevant to the game. My friends and I are birders as well so we get a kick out of talking about birds, and my mom won't play a competitive game so it's awesome to play anything with her at all. It fills a niche for a lot of people.
@@semanticboat Very true, I know a lot of people who like it and I will definitely play it if that's what they want to play and I'll have a good time, but it will never be my first pick. I also admire the bird cards, but the details are irrelevant to the gameplay and trying to talk about the cards inevitably pauses the game-so they may as well not be present. Settlers is not my favourite game either but it naturally directs all discussion towards the game currently being played, which I really like.
9:08 still watching the video and this feels woefully unrelated to the important part but I think this segment has convinced me to try this game with some friends despite not knowing what it was 10 minutes ago, my problem with Chess is how everything feels futile since I know I could mathematically be making better plays and it feels like more of a numbers game than strategy to me in that, but the way you describe it as "playing it for the game" rather than playing it for the best move sounds like it fixes all my issues with that set up and sounds really awesome to try out, so, thanks for that lol.
"AI" is certainly the right term in the context of computer games. In that context AI means _whatever_ logic controls the computer's units, nothing more. And as a wise game designer once said: it's not about making the opponents smart, the player is supposed to beat them eventually, it's about avoiding Artificial Stupidity, avoiding them behaving in a jarringly bad way.
The ghosts in Pac Man were a great early example of good AI: they seem to work together cleverly to surround the player, but are each just a few simple lines of code.
Yeah, the field as a whole has been called AI since the 60s or so. It is and always has been a very broad term. Anyone who insists on confusing it with AGI cannot be saved by terminology, and creating yet another disconnect between academic and public language wouldn't be worth it anyway.
I highly recommend Tabletop Simulator on Steam. Not only can you play with your out of town friends: you can also play expensive or out of print games (like LOTR Stratego) without having to scrounge up a copy. Sometimes my local friends are so lazy that we play on Tabletop Simulator anyway... but it is a lot easier sometimes to not have to all meet up.
If you're looking for good two-player games, I recommend Hannibal (full name Hannibal & Hamilcar: Rome vs Carthage). It's an asymmetrical two-player wargame which is really fun if you like history. I have stayed up until 4 AM desperately trying to get Scipio Africanus off the board before
Hi, to answer the question: "where do you people even find the time to play boardgames with your friends?" we don't, we go to boardgame clubs, where all the people there LOVE playing 3-hour 4-player games, and we learn new games together, and we make new friends together.
I would totally recomend that.
I will listen to you say things like "it's a black box" or "AI doesn't exist" for hours and do it happily.
The "It's Fine" shirt is either future merch or the perfect purchase. 🙂
Only 90 seconds in, but please keep up the good fight. Everything being AI drives me up the walls.
Back in the day my crew had board game parties. We (wife, friends, kid) were bug into a bunch of different ones like Catan, Munchkin (Adventure Time version is the best!) and Stratego!
My daughter was young and wanted to play, so Catan became a matching and adding game for her, and it was fun for us having this force of randomness added.
For Stratego I made a list of maybe 8 rules for her to set up the board: 1 place flag, 2 protect flag with bombs or strong pieces, 3 place bombs, 4 place miners, 5 protect miners, 6 place scouts, 7 place spy commander and general,8 place all other pieces.
We had some really great games! I usually won but she displayed some great strategies based off those rules like flag in corner, bombs on all the front lines and she learned to defend against scouts flying thru the gaps to steal the flag :)
the bombs must be seen as a minefield maybe?
In the original french version of the game, they actually were called "mines" to represent a mine field, and the defusers were actually mine sweepers. That also explains why they don't go away when they explode. Even if a single mine explodes, the mine field is still there.
One variation in Stratego is that in every battle, the spot of the losing piece is occupied by the winning piece, unless the winning piece is a minefield, in which case the minefield stays put. In case of a tie in level number, level 3 (colonel) versus level 3 (colonel) for example, both pieces are destroyed in a battle.
In that variation a good initial set up is to have one lakeside path blocked by two minefields, a second lakeside path partially blocked by using one minefield, and the third lakeside path unblocked by mines. This approach leaves 3 mines for protecting your flag, which is good enough if your flag is on an edge, or leaves a spare mine if your flag is in a corner. Wherever your frontline minefields are, you want to have a rank 7 (sergeant) or rank 8 (miner) piece adjacent in a rear area ready to destroy an enemy miner who destroys your minefield, since if you can succeed in destroying all enemy miners, your flag is safe if surrounded by mines, and then the only way for the enemy to win is to defeat all of your pieces with the ability to move.
A spy should ordinary be very close to the general, so that if the general is defeated by the enemy field marshal, then the spy can defeat the enemy general on the next turn.
In the gaps usually near the frontal minefields, a level 1 or 2 or 3 piece should be nearby in order to be able to stop most strong assaults. An series of level 9 (scout) pieces might be initially laid out in a left to right manner in the second rank, to be intentionally quickly gotten rid of in order to create a strait line lateral path for higher ranking pieces to quickly shift leftward or rightward.
Being the first to take a level 3 (colonel) or higher piece usually results in a winning game if you can also strongly deplete the enemy's initial supply of rank 8 (miner) pieces, since you can usually trade down until you have the highest ranking moving piece on the board once you have reasonably good flag safety.
At 28:54 that second piece of information given to the algorithm that a Stratego spy will always attack a known enemy field marshal is not always true.
Aside from a blunder, there can be tactical situations such as a perceived "flag rush", during which the delay of an attack toward the opponents perceived flag position is not worth taking the time out from the advance toward that perceived flag position in order to capture a field marshal that is not essential to either stopping one's own flag rush or in speeding up the enemy's flag rush. A better approach therefore is to give the mystery non-attacking piece a hard-coded multiplier of around 0.03 on the possibility of that piece being a spy (unless all other opposing pieces have been absolutely ruled out), and then making mathematical adjustments accordingly on the possibility of other opposing pieces being the spy piece, so that all the individual piece probabilities add up to approximately 1.000 of a spy, in accordance with the game start setting of 1 spy per side.
i taught my wife stratego during the pandemic. we haven't played a lot but I think it was in july that we played last.
Is she still your wife or what's going on with that?
@@anita.b no, unfortunately i taught her Stratego to use to decide how the divorce settlement would play out but she was too good at the game. She beat me so thoroughly she got all of my stuff and even my google account. this will probably be my last comment and then she'll own this account too.
I used to play Star War Stratego with my dad, so there has always been this, faint, almost whisper of a voice in the back of my head, whenever anyone brings up chess and checkers and it just goes, "I wonder if Stratego's fun?" As I've last played it when I was six. So, thank you so much for this video.
Does it have to be a 50% win rate to be considered a Nash equilibrium? My understanding of it was a Nash equilibrium is reached when both players in the zero-sum game have no incentive to unilaterally change their strategy.
It does not, because games can be asymmetric. But for games like stratego that are symmetric (assuming you flip a coin to decide who goes first), every Nash equilibrium is guaranteed to be 50/50.
If you start talking about asymmetric games -- including stratego where the same player always goes first -- then the Nash equilibrium can be biased, with one player only able to win 40% of the time or whatever.
@@just_some_commenter I see. Thank you for the clarification.
You are a much needed voice in the community, keep going. You slap.