Can AI play board games?

Рет қаралды 1,000

Күн бұрын

Join my crew on Patreon:
/ nightsaroundatable
In the very first episode of Ryan vs. the Robots, Ryan squares off against the ChatGPT 4o AI large language model from OpenAI in a winner-takes-all battle for Japan in Magic: the Gathering creator Richard Garfield's King of Tokyo.
Ryan further tests the robot's capabilities with Will Shortz's Tribond, and Alexander Pfister's Port Royal: Big Box.
Use these links to skip ahead:
0:00 Introduction
10:05 OpenAI's empty promises
19:22 Warm-up Tribond questions
28:24 King of Tokyo
1:07:34 Asking ChatGPT for more game ideas
1:13:45 Tribond Round 2
1:23:23 Port Royal: Big Box
1:48:53 Fun's over: we reach the message cap
Buy your own copy of King of Tokyo: Monster Box with all the expansions using the link below, and we'll receive a small commission:
amzn.to/3wQmNYy
Check out King of Tokyo on Board Game Geek:
boardgamegeek.com/boardgame/7...
Note: this video is marked as containing paid promotion not because it's sponsored by OpenAI (it isn't) but because Ryan received a demo copy of Port Royal, one of the games featured in the video, from publisher Pegasus Spiele.
.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.
Email Ryan: nightsaroundatable [at] gmail (dot) com
Chat with Ryan and friends on Discord: / discord
Stream with Ryan on Twitch: / nightsaroundatable
Follow us on Twitter: / roundatable
Like us on Facebook: / nightsaroundatable
Visit our Website: nightsaroundatable.com
Dance like kaiju: • Godzilla Fiesta Kaiju ...

Пікірлер: 21

@alexxpfister Ай бұрын

Yes, lets do a german interview! But I'm sorry to tell you, I don't have a cat, so the cat won't eat eggs :-). It was a blast to watch your video, Ryan. Rules: When you take 0-3 ships, take 1 card. So usually I stop drawing after the first or second ship (if the first ship wouldn't give me a lot of gold). Of course wanting to buy crew members, changes things a little.

@NightsAroundaTable Ай бұрын

Ah! i remembered the rule incorrectly. And i’m not wrong in thinking that the robot had an insane first turn drawing 20 cards. i’m not quite ready for a German interview. But i am determined to be!

@AlphaFoxtrottMike Ай бұрын

Thunder road vendetta? Take pictures of the board

@NightsAroundaTable Ай бұрын

Fingers still crossed that they’ll release the camera-enabled version of 4o and i won’t have to!

@MetalGearChamploo 7 күн бұрын

Fascinating

@robcannon Ай бұрын

Amazing results for a model not specifically trained on those games. It wouldn't take much effort for someone to produce a model that is trained on each of those games, specifically. What is really needed to make it shine would be some service that keep track of the full game state so is is always available to the model. That would keep it from forgetting the special cards it had available.

@NightsAroundaTable Ай бұрын

Yeah - i was impressed with its grasp of the rules, but unimpressed with its ability to actually play. At the end of the King of Tokyo game, i was going to win if the bot didn’t do something to stop me. The robot still had cards she could have used, and i think she was keeping symbols that wouldn’t help her. Another crucial element is the camera access. A whole new world opens up if the human doesn’t have to explain absolutely everything about the board state to the robot.

@RozanovaHunter Ай бұрын

I think the biggest consideration here is just how much computing power has to be burned up to have chatGPT play the second player. There have been plenty of discussions in recent years about making board games more ecologically friendly to reduce the impact they'll have on the planet. I can't see this application garnering any good will from that angle.

@NightsAroundaTable Ай бұрын

Hmm. You might be confusing Large Language Model AI with crypto? Crypto requires large server farms to crunch numbers day and night in search of ever-higher primes. i think a lot of power goes into training a LLM, but i didn’t think a ton was used in actually running one. Am i wrong?

@RozanovaHunter Ай бұрын

@@NightsAroundaTable From reading some articles, to my understanding it takes a lot of energy and resources to build the model, and the larger the model and more complex the operation the more power it uses. Simple chatbot functions use less than generating images, and small images cost less computing power than large ones, but since they haven't yet implemented the "process and interact with your environment" feature there aren't any studies or figures to compare that to yet. All told, AI is using the power of a small country just like crypto did and growing at a rate that will apparently break Moore's Law, but you're right that they are two totally different things.

@NightsAroundaTable Ай бұрын

@@RozanovaHunter Sure, BUT... different approaches to producing computation will outpace silcon in the years and decades to come. We'll be able to get more computation from fewer resources. And of course, the hope is that AI will grow smart enough to help us get there faster. It's kind of like... it's dirty and destructive to burn coal. But burning coal supplied us with the energy we needed to study alternative sources, so that we can eventually ween ourselves off it. But we'd never get there without easy, "dumb" ways to literally keep the lights on while we searched for more advanced methods.

@Commander6444 Ай бұрын

So I'm of two minds vis-a-vis AI and board gaming. On one hand, I am very interested in how generative AI models might be able to attain general proficiency at board games as a whole. The computers have long stomped us at Chess and Go, and it would be quite something if that kind of capability could be extended to any board game. I don't really believe that's on the horizon anytime soon, but we'll have to wait and see. After all, if I'm with two friends and the game we want to play works best with four, it would be great to add a computer player on a whim. (Though perhaps not at Deep Blue level of play.) On the other hand, some people have tried to position AI as a cure for loneliness. I _wholly_ reject using AI in this manner. At best, I see it as a band-aid solution for healing only the most chronic of loneliness. At worst, I fear AI companions inspiring all sorts of antisocial behaviors in many people. Why learn complex social graces when your robot friends will abide even your worst impulses and actions? Anyway, I know this is all a bit heavy for a board gaming channel, so I'll stop here. But that's my general outlook on things as I currently see them.

@NightsAroundaTable Ай бұрын

Oh, no - please. Let’s absolutely go there. i also find it unsettling how gosh darned polite LLMs are. i have had numerous arguments with AI, mostly calling it out for its overconfidence in giving me answers that i know are factually wrong. When i began using 4o with a mind to doing this livestream and it misled me about setting up Anachrony, i was angry with it for just agreeing with me about the setup rule instead of… you know… checking the rulebook. But i got this unnerving Stepford Wives apology (made all the more unsettling since the voice was female). i really want to delve further into why those high-ranking staffers left OpenAI, but the cynic in me thinks it’s just to they could start their own firms and rake in a little of startup cash.

@sevind7 Ай бұрын

Still plays faster than the in-laws

@NightsAroundaTable Ай бұрын

Way better at drinking oil too.

@tehpickle1250 Ай бұрын

Maaan I just find this stuff so mind-bogglingly impressive, even with the AI brain farts. But holy heck, it’s a lot of money to pay considering the arbitrary “wish” allowance. I would love to see how it copes with Lisboa, and how you’d cope with trying to explain the game state every turn! 😇😅

@NightsAroundaTable Ай бұрын

Yeah! Forget the Turing test… let’s establish the On Mars test.

@BobLee-df4zh Ай бұрын

It's even more mind boggling when you remember it's still in infancy.

@NightsAroundaTable Ай бұрын

@@BobLee-df4zh Yeah... any of the shortcomings you see in this video will be eradicated, and (judging by its pace so far) right quick, too. Within the year, i'd guess.

@ThisVojnik Ай бұрын

i didnt watch this video. but just saw the title. you playing with skynet bro. havent you guys seen the terminator? need to stop teaching it before its too late!!!

@NightsAroundaTable Ай бұрын

COME WITH ME IF YOU WANT TO PKAY ANACHRONY