Training AI to Play Pokemon with Reinforcement Learning

[@deesh6378]

An AI being traumatized by using a pc is the most ironic thing I've heard in a while

[@mcstrategist]

Haven't you seen twitch plays pokemon? PCs are a death sentence!

[@nimi-nae]

I didn't even make that connection 😂

[@Drakethos92]

@@mcstrategistI remember that. People were spamming. To get rid of pokemon. They had to ban people and make rules. That was pretty hilarious though.

[@Drakethos92]

@@nimi-naesame. But yeah that’s pretty funny

[@Kurayamiblack]

Sudden excessive punishment against a curiosity traumatizes first time experiencer Seems all too legit 😅

[@markcooper4876]

I laughed so hard when the AI refused to press the A button when it lost.

[@MP-lv5vk]

Stalling to avoid the outcome confirmation. Reminds me of young children, actually. Haha

[@JohnnyNatrium]

@@MP-lv5vk Sometimes the sound of a door slamming because of a gust of wind can remind me of children slamming their hands on a table. There is ZERO connection/homology between anything in the bot produced behavior, and the realm of human motivation or other emotions. It is logically impossible to learn anything about humans from literally everything about this showcase except by observing the actual human who decided to create this mathematical formula of instructions (algorithm) to a low level brute force bot.

[@kphaxx]

@@JohnnyNatrium Yeah but it reminds me of children's stubbornness lmao

[@marcelgonzalez1151]

Children can be the sorest losers, refusing to keep playing is hilarious 😂

[@Loliconman]

The only winning move is not to play.

[@Tommybgoode]

The AI is cool and all, lots of comments discussing it, but. I just wanna say, the editing is so awesome for a video like these, you don't often see such excellent presentation

[@Lone.Willow]

I'm honestly baffled by how this was animated. How did you get the scenes with the thousands of character sprites moving about, all overlapping one another?

[@patrickjones3826]

@@Lone.Willow all is revealed at 26:27

[@scotthuber8536]

200% this. Not only taking on the entire workload of the project, but taking the time making such an enjoyable and informative visual aid is stellar!

[@XistenceX1]

​@@Lone.Willow Yeah that's what's wild, the AI stuff is sick, but the editing to show the iterations had me fucking floored.

[@maxmuller6730]

I just thought the same, the presentation is amazing 👏

[@cappuccinocappy]

I dunno why but the clips were all the AIs aimlessly walk around like a colony of small ants is unbelievably adorable to me

[@sugabopp]

holy shit ai are the ants. or are ants the ai?

[@shreyandas4243]

Is this a subtle nod to @SmallAnt ?😂

[@Ricardo__Milos]

This is how Naruto trains himself: TAJUU KAGE BUNSHIN NO JUTSU. Then gathers the experience of each clone. :D

[@danielgoof8463]

Adorable? What are you a fucking pixie?

[@RavenLunacy44]

All I can picture is a tsunami of Ash just rapidly taking over a country, one town at a time.

[@butterfish6799]

“The ai is learning how to move, and is just walking around” really explains a lot of my online teammates in first person shooters.

[@jeffwooten6888]

Bots

[@Johnrick90]

Like my team mates in LoL

[@hallmark1]

Lvl 1 lukes in star wars battlefront 2 hvv

[@porkhill6665]

Npcs playing npcs 😢😮

[@IschmarVI]

@@jeffwooten6888"bot" sounds so negative. Maybe we should start calling them "reinforcemenrt learners" instead.

[@Hitmonstahp]

it was unreasonably adorable when the AI stopped in Pallet Town to enjoy the scenery

[@azukar8]

Seconded

[@AyaxTheDragon]

The AI is cute

[@htspencer9084]

Based AI knows true happiness.

[@Trashley652]

Ok but did you see the little dance after beating the bug catcher on the first try?

[@kratangg-arang]

Yeessss I envision it talking everything in with a solemn smile, knowing that it’s about to leave this quaint town on a grand adventure of trials and learning. ‘Just one more moment at the banks of this familiar lake, then I’ll be off…’

[@timothypickarski5234]

Not that I don’t love the videos that just say “I applied an AI to this game and here’s how long it took to finish it” but this video (in addition to its high quality visuals and great script) is so much beyond that. Instead of just watching a video on AI, we’re learning about reward implementation, the human condition, curiosity, and more and more. This went above and beyond, I was so rooting for our AI buddy by the end of this lol.

[@Fractisdnb]

You're right! This feels like an in depth, academic essay

[@wiiu-theunderratedconsole7569]

I want to see the AI beating the game

[@Scartay]

I want a full fledged version of this with atleast 20 hours to watch and actually make the AI complete the game, satisfying to watch, disappointed it had to end to end so soon :( Hopefully the success of this video has encouraged him to finish what he had started im here waiting for it!

[@kavinbala8885]

​@@Fractisdnbisn't that how most videos on this topic are?

[@juha2031]

This is to boring to watch

[@Godpull]

I’m so glad you didn’t stop when you said “this sounds like a reasonable stopping point”

[@BorrisBackyardigan]

But then he stopped not to long after 🥲

[@MillZ56]

That’s exactly where I stopped to mess with the comment section lol

[@DolanDarker]

This must've taken an insane amount of time to not only simulate but also edit, really good video, nice work

[@MoazSalama-ly5jf]

Omg Dolan you fucking legend where you been

[@MrGoodeats]

rN6media does the edits

[@anouaressanoussi]

Have you forgotten your password?

[@Hawk7886]

​@@anouaressanoussiobviously not

[@itsOZone]

​@@MrGoodeatsyeah most youtubers dont edit their own content anymore

[@pengwino828]

The ai discovering rng manipulation is mindblowing. I wonder if games in future could use ai to learn tedious or very specific glitches during beta testing.

[@antoinecharbonneau5108]

They already do!

[@NinjaArmy36]

Dude it clicked as he was explaining it "wasn't optimal" but also repeating and I was like "NOOOOOOO!!!"

[@sdsd-f7k]

Why bother?

[@veto_5762]

@@sdsd-f7k simple, ai thinks and tries things different to a human, it could discover stuff the devs wouldn't even imagine it was possible

[@snowolf494]

This is an elaborate version of fuzz testing, which is the act of feeding random data to a program to see how it react.

[@olemew]

That whole traumatic experience with the PC and the Pokecenter was fascinating. Thank you for making this

[@jeremycontreras6229]

The poor AI aww 😢❤

[@whirlpoolstudio97]

It triggered my Twitch Plays Pokemon PTSD

[@Spooglecraft]

the analogies between human behavior and AI behavior were quite interesting in general, though the trauma sticks out. also kinda makes you think about ourselves, doesn't it? after all, this is ultimately just a statistic algorithm with a simple reward system, but it manages to show some rather lifelike emergent behaviors, which weren't inherently programmed it. then again, pretty much all of life is not that different, the model and algorithm are just much bigger and more granular and complex.

[@perrowason5096]

Indeed it happened to me when i was young, i didnt know how to withow pkmn bc the storage system was a mess so i didnt use the pc anymore xd

[@shanemorris3554]

Reminds me of the trauma triggered whenever Twitch plays pokemon went near the computer after they accidentally released all those pokemon haha!

[@oAv3ng3dB055o]

I felt truly humbled when the AI was just done with the game at Mt Moon. I feel like so many real world experiences ended here that this moment just HAD to happen.

[@LeoTheDarkAngel]

My first thought was "Wow, imagine being so bad that you grinded to Blastoise in Mt Moon!" And immediately after that "Wait, didn't I have a Blastoise by that point, too?" I was so bad at navigating those "dungeons" as a kid 😂 (I still am, but I can look up maps now or be more strategic about it than aimlessly wandering around)

[@timonix2]

@@LeoTheDarkAngel I had my charizard when I battled misty for the first time :D

[@kylehill]

This was extremely well made. Great job

[@GameTimeWhy]

Holy cannoli it's science boi Kyle "Thor" Hill with his locks in the wild.

[@draaaven157]

I see we spend our sunday nights similarly. Lmfao.

[@buddycal1]

This is honestly one of the best endorsements this video could have

[@BigBaadMark12]

I’m certain the algorithm recommended me this video because of your comment

[@Big_Biba]

Its the goat 🐐

[@ArmoredarmadilloX]

Honestly the AI becoming traumatized from the PC was heartbreaking. Poor lil guy didnt understand what happened

[@istumby]

My heart dropped when it was revealed he never went back to the Pokémon center afterwards, I felt so bad for the guy.

[@hunterwylie6969]

@@istumbyright? Just imagine how rewarding it would’ve been to gain those total levels back! Probably would’ve broken the reward system, as there’s nothing keeping the AI from depositing the Pokémon just to get rewards for pulling it back out.

[@Yelonek1986]

@@hunterwylie6969 Deposit, withdraw, deposit, withdraw like a junkie.

[@nousukas]

"The Pokémon center stole my only squirtle!"

[@kdsavage1991]

Don't feel bad, they learn as they go!

[@ERFMXVCC]

I'm so glad the KZbin algorithm decided to recommend your video and I clicked on it. It's a fascinating thing to watch the process and journey that the AI goes through, while the presentation of the whole video is equally fantastic. Great video, you all deserve a round of applause for the effort and quality put into this whole project.

[@brandonbrsndon]

i love that the AI decided to just hang out and watch the scenery. reminds me of my favorite poem “Stopping by the woods on a snowy evening” by Robert Frost

[@piciperkuadrik4636]

Everybody likes Robert Frost

[@danielserrano929]

I’ve done this many times in my play throughs with Pokémon, it’s actually scary how much the AI “mimics” human behavior.

[@aceq361]

​@@piciperkuadrik4636not True I actually HATE Robert Frost

[@TheMysticalmax]

You have good taste. That's a beautiful poem

[@kingocowata]

@@danielserrano929because we’re in a simulation 😂

[@DigitalIndra]

"Just hanging out and admiring the scenery, is more rewarding than exploring the rest of the world." Never have I felt more like a machine learning algorithm than this sentence right here.

[@harm991]

The digital world is more rewarding than the real world

[@ceigey-au]

Very relatable outcome!

[@counterleo]

Me too, why bother capturing and fighting when you can just chill and enjoy the motion of leaves and waves? Quite poetic

[@aylakoch4516]

This was edited and put together so amazingly well. I haven’t even finished yet- I just needed to express my gratitude that you took the time to not only complete this project but edit the process in such a visibly appealing way. Thanks for 33 genuinely enjoyable minutes!

[@dognigga]

shit was boring asf, felt like a lecture lol

[@faith9505]

A Happy Way to Live The servants who are ready and waiting for his return will be rewarded. -Luke 12:37 All around us we can see fulfilled Bible prophecies, signs indicating that the return of Jesus Christ is drawing near. As followers of Christ, we should be watching for Him. We need to be ready to go. Jesus, speaking about His return, said, “Be dressed for service and keep your lamps burning, as though you were waiting for your master to return from the wedding feast. . . . The servants who are ready and waiting for his return will be rewarded” (Luke 12:35-37) Are you ready for His return? To be ready means to be engaged in activities that you wouldn’t be ashamed to be doing if Jesus were to return. It’s a good idea to periodically ask ourselves this question: This place that I am about to go, this thing that I am about to do, would I be embarrassed if I were doing it when Jesus came back?” Think about your plans. Is there anything you will be doing today, tonight, or tomorrow that you would be ashamed to do if Christ were to return? If so, then change your plans. You want to be ready for His return. Not only should we be ready, but we should anxiously await the return of Christ. We used to have a German Shepherd who slept outside the bedroom, leaning against our door. We didn’t let him sleep in our room because he often had nightmares and would wake us up. Every morning when we opened the door, he rolled into the room. Then he’d jump up and start running in circles. He was genuinely happy to see us. That is how we should be waiting for Christ’s return. And anything that might prevent us from saying “Come quickly, Lord Jesus” is out of place in our lives. In addition to waiting, we should be working. Every now and then, someone predicts that Jesus will return on a specific date. People believe these predictions and start quitting their jobs or divorcing their spouses. But that is not what we should be doing as we wait for the return of Christ. Instead, we should be working for Him. The Bible says, “Just as the body is dead without breath, so also faith is dead without good works” (James 2:26) If watching is the evidence of faith, then working is the evidence of faith in action. Watching for the Lord’s return will help us prepare our own lives. But working will ensure that we bring others with us to Heaven. The great British preacher C. H. Spurgeon said, “It is a very blessed thing to be on the watch for Christ. . . . You can be poor without murmuring; you can be rich without worldliness; you can be sick without sorrowing; you can be healthy without presumption. If you are always waiting for Christ’s Coming, untold blessings are wrapped up in that glorious hope” When you live in the anticipation of Christ’s return, it’s a happy way to live.

[@jdllim]

It's one thing to set all this up, and it's another to visualize and present it in such a coherent and digestible way. You did both so well! Hope to see more content from you in the future

[@ArthurvandenBroeck]

Agreed! This video is insane!

[@youngsdiscovery8909]

I can't believe it's done by individual. Super high quality.

[@nyx3m]

This man made a channel. Dropped the best ai educational video using pokemon. Didnt post or commented anything . leaves. What a legend

[@mygirldarby]

You can find him online. He's a software engineer and creates all kinds of interesting things.

[@advenco344]

@@mygirldarbyI just looked him up and wow this guy is a genius!

[@steven-mz3jf]

Seeing high effort videos like these from relatively low sub channels always surprises me. Definitely deserves more recognition/subs.

[@napoleonbonerfarte6739]

It's the only video on bro's account lmfao wdym

[@RandyGBH]

@@napoleonbonerfarte6739lol was about to write this too

[@abhishekkoundal584]

Good things take time

[@absolutelyfookinnobody2843]

And people who over react to low sub channels being high quality doesn't surprise me. Lots and lots of dumdums out there

[@BlackScytheLP]

Haven't even finished the video yet, but I want this to pop off in the Algorythm, this video had tons of efforts put into it, and deserves to get out there.

[@Neo_Data]

I've got some good news for you, that's how I found this video

[@esotericraime1441]

The algorithm brought me here

[@auraonline9073]

Guess I'll throw on a comment too then. This is great!

[@toyushi]

yesss this was so cool

[@aaronhpa]

Thanks then

[@turingtestingmypatience]

watching the little reds go round like an ant colony brings me so much joy and i don't know why. look at them all exploring. learning. discovering the world. lil guys. thank you for spending at least 1000USD and several hours putting this together just for me to uncontrollably laugh at the reds for 20 minutes ..with that out of the way, fantastic video. incredibly readable visuals and clear voiceover, awesome topic, understandable for several levels of previous knowledge. can see this hitting the high hundred thousands.

[@raula6533]

I was looking for this comment because I thought the same!!! It was like watching ants!! Just amazing!! This video exploded my mind... Imagine a Pokemon game were you can compete against a real "rival" (blue) in real time just to see who wins the league first... And every run the rival gets different pokemons with different moves... This guy is just insane, this is like a Pandora box!!!! New sub for sure!!!! And thank you for this video Peter!!!!!

[@gigabytemon]

Peter: *starts to train an AI to play Pokemon* Magikarp Seller: "It's time to become a multi-millionaire."

[@TZDUHH]

😂

[@ThePlumGeneral44]

He literally did earn 5 million lmao since the guy said that around 10000 of the ais bought the magikarp

[@hornoxthekingslayer8100]

If you ever do have the AI finsih the game, I think it would be really cool if you let the same AI try Pokemon Gold. I think seeing if an AI trained on Gen 1 could play Gen 2 that would be an interesting experiment

[@zeebo30]

Obviously it'd have to relearn how to navigate the map, but it'd probably do well in battles since it already knows how

[@johnhamilton5431]

It wouldn't be able to catch the farfetchd or use cut This game would fail too with the hm's

[@geekygecko1849]

I'm going to do this as a project for my machine learning class, and I am planning on trying the same algo on Gen 2.

[@lpsfoxstar8454]

do make a video@@geekygecko1849

[@BatistaInvests]

​@@geekygecko1849how can I follow along?

[@SplishySploshy]

As a psych prof I'm always trying to think of different ways to explain certain concepts and give relatable examples, and this one is perfect!

[@MasteringSilence]

They tell me I’m crazy here 🤪

[@norabarlow17]

​@@MasteringSilenceCrazy? I was crazy once

[@MasteringSilence]

@@norabarlow17 you only lose your mind once… They put me in a rubber room with rubber rats…

[@knockout8157]

@@norabarlow17they locked me in a room. A rubber room with rats.

[@charpool169]

As a psych professor can you explain the appeal to these people repeating the copy paste comments? Also just to be clear I'm also asking out of genuine curiosity if there may be psychological reasons past the basic wanting to be a part of something, and not just trying to hate on them or anything ✌

[@eddie7252]

I honestly expected this video to be from a youtuber with thousands of subscribers, to see that you only have 60 baffles me, this is an incredibly well-made and well-put together video.

[@viperific3410]

yeah i thought the same, its gone up to 400 now but still nuts

[@androsp9105]

tbf, it's his first video.

[@viperific3410]

@@androsp9105 yeah I only realised that after I left this comment, even more nuts lmao

[@Station9.75]

He’s gained nearly 5,000 in a few days. Very good going.

[@AwesomeHairo]

Misuse of commas.

[@willianrocha5475]

As a Pokemon enthusiast with 4 Pokemon tattoos and a data analyst aspiring to become a data scientist, this project was one of the coolest to watch! I was so fascinated that I decided to replicate the project myself. I encountered some difficulties along the way, but the Discord community was incredibly helpful. Congratulations on the project! 🙌

[@JuliusCaesar888]

Thanks for telling us all about your 4 Pokémon tattoos, that's just the proof we needed that you really like the games. All of us had proof standards met, you're definitely a fan. Congratulations bro.

[@PhycoSerpentStudios]

How easy is it to replicate and can you have it play other games such as gold? Or even newer gens like fire red. Itd be cool to get it to beat red. And then see how long it takes to get through other generations

[@e4e5e2e7]

Extremely impressive visualization of the simultaneous iterations. It can be hard to grasp that machine learning is happening in batches of mass parallel attempts, not each progressive scenario after another one by one. Excellent video!

[@lateusbetelgeuse]

Since I'm all into both Pokémon and coding, KZbin suggested your video just minutes after you uploaded it. I subscribed after a few minutes watching it, and now I watched it again and noticed you have almost 50k subscribers! With just one video! Please take that as a public, worldwide testament of the effort you have put into this. Thank you so much!

[@DruggiePlays]

Broke yt 😂

[@t2g648]

If you like Pokemon and AI, you'll love this: kzbin.info/www/bejne/gYjae3evd61rn8k&ab_channel=Spawnvilley

[@jondebeer6863]

As a physicist i appreciate those visualizations. This is truly remarkable content.

[@aurelia8028]

wtf does you being a physicist have to do with anything? guess you just wanted attention.

[@steak9194]

16:13 The ai never making it out of mount moon is pretty relatable, to be honest

[@chrispyvolterra]

This is their first KZbin Upload, it’s crazy to me how much work, effort and money went into its production without having built an audience on an already successful channel before. Mad props to you Peter.

[@chrispyvolterra]

I am looking forward to see what else you will create.

[@notavailable947]

He is an employee of Amazon Headquarters in Seattle 👏🏽👌🏽 He is smart af

[@joshuasims5421]

That was incredible! I’ve always wondered if this was possible, I’m blown away by what the AI was able to learn! The visualizations and presentation were excellent, I hope this video reaches a wide audience!

[@YMilkshake]

The visualizations of the AI exploring is actually insane! Seeing the entire map and all iterations moving looks so dope, especially with the arrows indicating their average movement. Sick video!

[@bentpen2805]

It was like watching what PokeMMO would’ve been in the early 2000s

[@faith9505]

A Happy Way to Live The servants who are ready and waiting for his return will be rewarded. -Luke 12:37 All around us we can see fulfilled Bible prophecies, signs indicating that the return of Jesus Christ is drawing near. As followers of Christ, we should be watching for Him. We need to be ready to go. Jesus, speaking about His return, said, “Be dressed for service and keep your lamps burning, as though you were waiting for your master to return from the wedding feast. . . . The servants who are ready and waiting for his return will be rewarded” (Luke 12:35-37) Are you ready for His return? To be ready means to be engaged in activities that you wouldn’t be ashamed to be doing if Jesus were to return. It’s a good idea to periodically ask ourselves this question: This place that I am about to go, this thing that I am about to do, would I be embarrassed if I were doing it when Jesus came back?” Think about your plans. Is there anything you will be doing today, tonight, or tomorrow that you would be ashamed to do if Christ were to return? If so, then change your plans. You want to be ready for His return. Not only should we be ready, but we should anxiously await the return of Christ. We used to have a German Shepherd who slept outside the bedroom, leaning against our door. We didn’t let him sleep in our room because he often had nightmares and would wake us up. Every morning when we opened the door, he rolled into the room. Then he’d jump up and start running in circles. He was genuinely happy to see us. That is how we should be waiting for Christ’s return. And anything that might prevent us from saying “Come quickly, Lord Jesus” is out of place in our lives. In addition to waiting, we should be working. Every now and then, someone predicts that Jesus will return on a specific date. People believe these predictions and start quitting their jobs or divorcing their spouses. But that is not what we should be doing as we wait for the return of Christ. Instead, we should be working for Him. The Bible says, “Just as the body is dead without breath, so also faith is dead without good works” (James 2:26) If watching is the evidence of faith, then working is the evidence of faith in action. Watching for the Lord’s return will help us prepare our own lives. But working will ensure that we bring others with us to Heaven. The great British preacher C. H. Spurgeon said, “It is a very blessed thing to be on the watch for Christ. . . . You can be poor without murmuring; you can be rich without worldliness; you can be sick without sorrowing; you can be healthy without presumption. If you are always waiting for Christ’s Coming, untold blessings are wrapped up in that glorious hope” When you live in the anticipation of Christ’s return, it’s a happy way to live.

[@MousechrisTFM]

It stopped to look at the scenery, exactly like humans do It was petty and refused to press A for its defeat message simply just to not be told it lost It rage quitted avoiding brock because it lost too many times Your AI is just a literal human todler This AI is genuinely adorable

[@flicmylich]

Bro honestly this is KZbin video of the year. How spectacularly you presented this information in such a clear and entertaining way that is honestly on the level of professional science productions like Cosmos. Absolutely colossal performance man. I wouldn’t be surprised if you had an entire production team.

[@peterwhidden]

thank you for the kind words :) no production team, but my friend @torinblankensmith made the thumbnail

[@lovol2]

I second this. I'm super interested in the content, but at the same time I'm like.... However did he make this look so good.

[@glupshitto1977]

It’s not that deep dude holy shit

[@Tom-yg7mi]

@@glupshitto1977 its deep.. learning.

[@Fissan_Poulsen]

@@Tom-yg7mi get out

[@trbremm]

I really like how grounded and transparent your breakdown of the AI capabilities and limitations is, it shows it as a tool and not as a magical solve-all-problems strategy. Also, what a masterful storyteller and explainer you are. This video is very well paced and laid out, congrats!

[@SSGoatanks]

Yes it's limited but imagine what it could become in a few more years 🤖

[@JoaoMorais-ee1oq]

Everything about this was amazing, the computational approach, the video edit, the tone, the explanations and the real life parallels. Beautiful work!

[@jaredezz]

The technical expertise that went into this is astounding. As a lifelong pokénerd and career software engineer, I applaud you for capturing my attention with such a captivating topic and experiment. But as just a person, I thank you for relating it back to the human condition. Realizing that we get distracted by false or incorrectly calibrated reward systems gives us awareness, the first step in the right direction toward pursuing real, meaningful value out of life. I would love to hear more of this from your obviously talented and insightful, inspirational mind.

[@matteoemanuele-gi4jk]

Fellas, I'm an AI engineer, with a short background in Reinforcement Learning for a period I interacted with Sony for a job. I need you to understand the MAGNITUDE of these results. It's an insane work, and I'm sad that probably only a few might understand the sheer amount of skill require to do this. Insane job man, you are a goat

[@dkm9090]

This is no understatement . This takes a level of focus and problem solving that is just not normal. Savage!

[@bricegardner7815]

I’m not even an engineer, and my jaw is on the ground. I genuinely would love to learn how to become a part of this world. I wish there were more people in my circle with hobbies and fascinations like this. I used to help write xml codes for world of Warcraft bots when I was a kid. Now laying in bed with an alarm set for five hours from now. I’ve got a sales job… is 33 years old too old to learn how to work in this scene? This video drips with knowledge, and a wisdom and understanding of something that I have no idea how to even begin to approach. Kudos!!

[@alr9447]

I wouldnt say those results are impressive theory wise ? The impressiveness of the work comes from a technical point of view, how great he managed to link the RL model with the game and the fine-tuning he put in it. By the way, AI engineer doesnt really mean anything, what is your job title ? Out of curiosity

[@harshrajjadhav940]

​@bricegardner7815 no age is too high. With enough determination and curiosity you can definitely pivot. Look into videos explaining the skills required to get a job in game development/ AI.

[@matteoemanuele-gi4jk]

@@alr9447 I am officially a data scientist, but within the team I'm the guy responsible of the training of the ML models, therefore I make this distinction because nowadays "data scientist" is too broad. In most big tech companies, AI engineer is a common notation to distinguish between the data science folks

[@Solsumi]

Dropping a comment to help the algorithm. This video honestly deserves millions of views. I love the part where the AI learned to RNG manip to catch a Rattata. It's one of those moments that's unexpected at first but when you go back and look at it it's like, "oh, of course it would react like that!" Moments like those are why I love AI learning videos like this.

[@henriquemagalhaessoares8739]

This was an amazing project and explanation. You should submit this to The Journal of Geek Studies if you don't have a publication lined up already.

[@seveneyes77]

Wah is that a thing?

[@henriquemagalhaessoares8739]

@@seveneyes77 Yep! They are an online publication that uses geek culture as a way to popularize science. They had a bunch of articles from the biology if final fantasy monsters to the effectiveness of super man disguise.

[@BryceHuston]

They told me my Pokémon phase would pass. Little did they know, it was just evolving into an AI obsession!

[@sanjaywilson8232]

Pokémaniac Bryce Huston wants to battle!

[@ashashii911]

​@@sanjaywilson8232LMAO!

[@30303Steve]

*Pokemon Trainer Battle Theme starts playing*

[@aplesos]

Edit : Go Lucario ! Fight Pokemon Bag Run away

[@brandonvolesky9867]

Your findings, implementation, logic, and ANIMATION is incredible. 👏👏

[@Asidders]

Yep. This is easily one of the best videos regarding Pokémon on KZbin.

[@aBlackMage]

This might be the coolest video of AI playing a video game I've ever seen. I love all the fascinating emergent behaviours (especially the RNG manipulation), as well as the analogies you draw to humans. I also love that you presented the technical explanations in a way that allowed me understand almost everything without any programming knowledge, just a decent understanding of AI. Genuinely amazing job, I hope to see more like this in the future! :)

[@artisticcrossbreed]

I'm absolutely dying for an update, I keep checking this channel every few months to see if there is another video

[@kuromyou7969]

I'm only 12 min in and looked to see if there were more videos! But alas...

[@headyshotta5777]

this video is mindblowing. I have absolutely no clue how you collected and translated all this data into such cool visualizations, but i am in awe. this is so cool. thank you so much for making it!

[@Hateburn]

I remember back when there were 1 or 2 reinforcement learning videos on YT. Now we get all sorts. But this one...this one is special. The production value here is excellent. Thanks for all of your hard work.

[@LPcrazy_88]

Holy crap this is your very first YT video? I can't wait to see what you cook up if you continue to create! Outstanding work!

[@clickpwn]

Tbh I didn’t know youtube algorithm allowed channel with 1 video to pop off like this. Over 1 million views in 7 days?? If this video was posted in a sizable channel, it might have been even 10 times more.

[@bilibangbang]

he paid for the view XD@@clickpwn

[@sarkhaaan]

@@bilibangbanghow you know that?

[@StonyBaboon]

It gets better when u go into the git-hub project and find out that he has been working on this for the last 2 years...

[@blake..-]

@@bilibangbangmald

[@lukasruger6048]

This might be the most interesting, fascinating and satisfying video I've watched on KZbin so far. Hats off, I'm looking forward to future videos!

[@RageAgainstTheTards]

This was awesome, I'd love to see a full series of the AI completing the game.

[@ΧάρηςΚυπαρίσσης]

Yes!!

[@tsunalein]

downloaded it and train the ai more

[@cassidy8307]

And then i'd like to see it completing the game as fast as possible. An AI speedrun competition: winner gets 100,000 arbitrary points

[@benjones8779]

Not sure if its been said already, but, I would love to see them beat the game. Then we can see what levels they got to and what they thought was the best pokemon to have for the elite 4. Would be interesting.

[@Asidders]

Charizard with Slash, easy

[@teracraged320]

This will took a looooott of time and video preparation edit etc. But I agree would be awesome

[@Mcobange]

i really doubt if the AI can solve the Stone Moving "Puzzles" inside the IceCave and VictoryRoad thou. Can it even be teached to learn and use the VMs? but id love to see it :D

[@davidfl4]

I think it would be hard to program the rewards to get them through the specific obstacles tho like using cut in certain places etc

[@stephenh9483]

Yeah they would shatter Wersters Speedrun World Record!

[@phancanedoo013]

Just 10 minutes in, and it has already gotten so damn interesting! The behaviors, the systems, the events, the unexpected but explainable scenarios, the AI literally experiencing something comparable to trauma? I want to see more!

[@kaelthunderhoof5619]

The Red swarm wasn't enough?

[@kaio0777]

me too sad when he stop at moon mt.

[@Elintasokas]

The AI doesn't experience anything because it's not a conscious entity. It experiences as much as Microsoft Word when you open it.

[@alexb8926]

@@Elintasokas😅😂😂

[@JamanWerSonst]

@@Elintasokas Based on my PCs heavy breathing when I open Word I assume its orgasming.

[@BernardoMachado]

This is one of the best implementation and visualization videos on the subject I've ever seen. Amazing work!

[@TailsMiles249]

The amount of work you've put into this is so incredible. All of the self recording of _all_ of the AI iterations meant time spent (never wasted) for the sake of a single video. From the editing you've shown down to the research of how the human psyche works, this is beyond something I would even think to produce. You will go far in your endeavors.

[@MageMinionsOP]

This guy's first video and it's about using ai, so KZbin AI said "I gochu"

[@rainwatervideography4546]

This video was done incredibly! A perfect demo of and comparison to deep learning. A well earned follow. The dedication, creativity, and in depth descriptions are beyond impressive for this being the first video on this channel. Keep at it! I'll be looking forward to what ever you produce next!

[@bronzebond4869]

I’ve got no idea who you are but i can say how proud i am that someone tried this and had the patience to gather such interesting, noteworthy and valuable insights. Great work fam. Awesome explanations as well

[@MarcusTheDorkus]

Right off the bat I like this video because it actually goes into detail on how success is defined. Way too often this is skipped over and it absolutely breaks my brain because the implication of not covering this is that the AI somehow figured things out without any sort of goals defined.

[@Gambleraddict79]

adding ai to Pokémon added absuloutely nothing to the game. same idea as staring out a window and twiddling your thumbs worthless NPC behavhiour

[@mischavandenburg]

“Just hanging out and admiring the scenery is more rewarding than exploring the world” Amazing work Peter! I look forward to see how this will progress

[@MichaelCrecker]

The AI naming Squirtle “AAAAAAAAAA” killed me! 😂Thanks, amazing content.

[@RevanBC]

AI picked the Squirtle in Pokemon Red lol what a contrarian

[@Tropictopic69]

i was hoping someone else had mentioned this

[@Tyler-qh7bf]

@@RevanBC that was its only option...

[@RevanBC]

@@Tyler-qh7bf No you can pick 2 other pokemon! idiot.

[@sergeantjoe6802]

Pigeoto was ‐-----------

[@wb7779]

I got a lot of insights from watching the video. The visuals were the best.

[@AMNEZ1A]

This video reminds me of when I got Pokemon Yellow as a kid, I didn't read/speak english so I just had to try things to learn what everything did and was. It's weird how similar the AI playing feels to my experiences as a kid. The Pokemon games (among TV and other games) actually helped me learn english at the age of 9 far before my classmates could and as a little extra ROM hacking got me into graphic design and coding/web development somehow. Pokemon in general is the base of my origin story.

[@kaio0777]

damn bro that is deep.

[@Edoss98]

Me with Spanish at 3 years old and English at 2 ahah Pokemon Azul and Pokemon Red 😅

[@joaofernandes6349]

When I first played Pokemon, just like yourself I was still a kid didn't know any English so I couldn't even save. The first few months was just like the AI, start from that little room and trial and error.

[@2006HondaCivicD]

Hello fellow ESL player, i was like 5 when I got my first hand on pokemon. I was EXTREMELY upset when I accidentally start over the game (the copy was second hand and the saved file is from my older brother who already completed the game.) that I cried. I lost my brother's charizard, even the moltres he caught with an ultra ball because i couldnt understand a lick of English back then - overwriting his save accidentally, and I just love exploring the pokemon world more than battling them. Only then 3 whole years later when I did restart and beat pokemon on my own, around 12 I became competently aiming to "gotta catch em all".

[@lexusenthusiast6414]

This is how I taught myself English at 6 years old, pokemon yellow and a dictionary provided by my parents. Wild

[@theopiumden1551]

This is one of the most fascinating things I’ve ever seen. You deserve (1) reinforcement point in the form of an award. 🤙

[@menerdo]

5 mil on your first video. Great quality, good research and break down. Congrats, can't wait to see what you bring next!

[@Distractionn-CG_5945]

This was absolutely amazing, my friend! Please do more of these! I must admit I was disappointed that you didn't do the whole game 😂

[@jay.c2097]

Can’t believe this is your first video. This was so entertaining to watch and the editing leaves me wondering how much time it took you. Hope you put out more videos like this and I’d love to see a full AI playthrough at some point!

[@Deusregere]

I was kinda disappointed that it didn’t conclude with the AI defeating the Elite 4 😢

[@Emerald-Fluffie]

Residents of viridian city watching in horror as the swarm of Reds rapidly engulfs the city

[@maddo4]

I explained this to my Fiancee who works as a addiction recovery specialist and this come off as reward seeking behavior commonly associated with alcohol and drug addiction. The way the AI sought increasing point value is similar to chasing a high and it refusing to enter the pokemon center on what seems almost like fear to lose points even at its own detriment is very close to what human addicts might do to keep feeding their addiction.

[@crypto1upteam750]

What a cringe take lol

[@MrCmon113]

Not just addicts or humans, but all animals in general. Reinforcement learning as a problem setup is very general. But it's still a model of the world and not the world. In reality there is e.g. no separation between agent and environment and animals also think into the future rather than deciding only spontaneously.

[@coomlord5360]

Holy shit, did you really try and school somebody on their own specialty? Not only that but u prob sounded dumb as fuck to her. This can literally be compared to all functioning adults who chase rewards like “getting promoted” or “learning new things” as they hit certain reward systems in our brains. And refusing to enter pokecenter can be related to refusing to build new relationships or reinforce current ones to make more money, etc. not trynna be harsh but pretty odd takeaway ngl

[@christiannersinger7529]

Making it to viridian forest is already insane, it's got 3 separate if/thens to complete that don't have rewards aside from gameplay expansion, this is actually cool af

[@thermonuclearwarhead]

The accidental traumatic depositing of Pokémon in the center is rather hilarious, and the Magikarp/fast food analogy is beautiful. Picking left is an ancient gaming trick, not surprised AI picked it up/that we make games that reward it. And lastly the short-term memory bit seems to me a great idea to solve this (and also, accidentally, rather human :P).

[@counterleo]

I was feeling sad for the AI who must have thought it accidentally killed its Pokémon 🥲😂

[@lesbo37]

The only flaw in the fast food analogy is we'd need to learn that in the future eating fast food will make you live longer (or something else awesome) given what Magikarp evolves into!

[@teenslayer]

i thought the traumatic experience was super interesting too and funny lol

[@ChacalLoL]

As a Data Scientist, this was amazing to watch :) well done !

[@antonioiniguez1615]

I wanted to be a Data Scientist then I realized I couldn't code😂

[@htspencer9084]

Anyone newer to Machine Learning: this video is such a great introduction to concepts such as: * Reward Functions * Misalignment * Emergent Behaviour And more!

[@ProtonManBB]

this is the type of content that I love - thank you for being on this wavelength

[@3ountyhunter]

Is this really your first video?! This is incredibly well done. So glad YT has recognized that your content is deserving of being pushed algorithmically.

[@SevereMalfunction7]

this is honestly worthy of an entire course's final project at the graduate level. Thank you for making this freely available!

[@warmcat]

Isn’t it just! I’m currently half way through my final project for my MSc, with a relatively shit regression model predicting energy usage. 😂

[@dylancope]

This was a really excellent video! I'm super impressed that you managed to get this working - as a PhD student working with RL I understand that it can be an nightmare to debug! And I appreciated the depth of technical details you gave at the end. The presentation of the video was really good. I really liked how you eased into the deeper explanations and created lots of cool visualisations. I'm surprised that this is your first video and I hope you make more :)

[@akk1584]

If this hasn’t already been suggested, you should make a screensaver video of different cities populated with a ton of Ashes walking around-that shit is mesmerizing!

[@baguetta205]

As a computer science student (and a long-time Pokemon fan) currently taking a semester off due to mental health stuff, this really helped to get me interested in my career path again. When depression and anxiety get in the way of your day-to-day life, your interests can become few and far in-between, and the things you used to find joy in start to feel pointless and mundane. I've always loved Pokemon, and just want to thank you for the mindset shift. This video was incredibly well done, and I enjoyed every second of it.

[@peterwhidden]

Thank you for sharing that. I also took time off when I was school. Hoping you're able to find the joy, and wishing you the best of luck in your journey!

[@elusivelectron]

​@@peterwhiddenCongrats! Now you know what it's like to be a parent.

[@leeslater8962]

This video is outstanding and surely one of KZbin's all time best. Can't imagine how much work you've condensed into half an hour, and managed to make what is quite technical/dense material into something really engaging for people on different levels of prior/knowledge.

[@elrudiiisimo3066]

I’m not sure if you noticed this or not Peter, but this is historic. In terms of R&D and just human science. Very impressed with this creativity and passion. Cheers 🥂

[@kaComposer]

Genuinely blown away by the many high level skills this takes. On top of that, you have an incredible ability to teach high level concepts to a lay audience. Very rare!

[@PantheraTK]

@@kaComposer Agree. This level of technical ability plus storytelling ability is magnificent.

[@downformexico5470]

Some more of these. They are really high quality

[@Nilumbra]

Dude, you are a genius! I am taken aback that this is your first video. Your skill, knowledge, production value and way of balancing what can be a dry subject with interesting information and funny tidbits is absolutely amazing! I am seriously jealous, with your skills you are gonna go far! I am subscribed! Would love to see more Pokemon AI stuff, but understand if you wanna go a different direction as well.

[@13hydRa37]

It look like, he run this program for himself too, to get the best possible reaction to his first video. And we all part of a simulation :O IMAGINE

[@Johnrick90]

Amen to this! I see great science potential here in multiple purposes/ subjects!

[@naveking1]

This is great- I have no background with anything programmingwise but you made this into such an entertaining story. I hope that this blows up enough to get a sequel at some point, I'd watch this for HOURS!

[@plasmakitten4261]

This is like such a classic example of how AI thinks differently from humans. It can't figure out how to get past a ledge but its pattern recognition is so strong that it figured out friggin RnG manipulation by itself.

[@NikhilAutar]

We humans also have reward systems. Everything "living" does. It's different to an AI model. But who's to say that we're not just an AI model with different base rewards?

[@plasmakitten4261]

@@NikhilAutar The term "artificial" is meaningless unless it's being used to mean "made by humans". Since we didn't design ourselves, we aren't AI by any useful meaning of the term. But at the core, this way of designing AI is designed to mimic how humans learn, so you're not far off.

[@ryanwirawan5012]

Not only the pokemon gains EXP points, the AI gains EXP too

[@NikhilAutar]

@@plasmakitten4261 We'd be artificial to whoever designed us/this haha

[@cs16Tactics]

I think completely opposite :D This (video) was prime example of how phenomenons that happens with humans can be put into numbers used by AI learning. Our learning = pattern recognitions based on the rewards we've gotten. They aren't as vivid with "Getting 3 points on catching pokemon", but rather intuitive that happens automatically.

[@JuliusCaesar888]

LMFAAOOOOOO the AI being PTSDd by interacting with the PC is absolute gold. This is the kind of satire that is brilliant by nature and just forces a person to stop and laugh about it for a few minutes.

[@lolmanor]

Excellent video! I always love the videos where the AI attempts are overlapped. Makes me feel like I'm staring at a bunch of newborn ants explore the world. Also the Pokemon center "trauma" moment was so cute! Poor AI!!! You didn't do anything wrong!!

[@trevorschauwecker]

I noticed at 17:20 you mentioned you were unsure of why it chose to move in a single direction with limited memory. When I was a firefighter, we used a method of keeping our right or left hand on a wall while searching smoke-filled buildings. The method that AI used is more or less the same method and my hypothesis is that it learned it could discover new areas more easily by utilizing this technique which triggered more reward points at a faster rate.

[@faizanulhaq8349]

what if the wall is on fire

[@trevorschauwecker]

@@faizanulhaq8349 probably going to be pretty hot…

[@maxng7211]

I also noticed that the first few towns have right turn bias. That being if you follow the wall on the right you're more likely to get to a new area faster than the wall to the left

[@elsephiroth666]

@@maxng7211 Probably because the game is circling right (start on the left of the whole map, go up then right then down then left).

[@bideo_games]

This is correct. If you wanted to escape a maze the slow but sure way, you would hug the wall until you reach the endpoint. So if you only made right turns the entire way, it may take longer but you would have a deterministic method of completing the maze. Using this method you could clear caves without needing Flash.

[@metleon]

The counterclockwise motion may just be a result of how certain maps are layed out in the first could areas. You have to go to take the right path on Route 1 to get to Viridian, so following the right wall will get you there. Then in Viridian Forest, hugging the right wall is pretty much the fastest way to get through the forest, and I think later generations only deviated slightly to fight trainers for more experience once they realized that Brock was the main roadblock. Finally, in Pewter City, you need to take a counterclockwise path to get into the gym. So the AI probably didn't have a preference at first, but going in a counterclockwise direction for it to where it needed to go there fastest.

[@ForcefulDragon]

That sounds like a pretty good theory to me. Though if areas later in the game required a different pathing techniques... Would it be sophisticated enough to only use new pathfinding techniques when required and still use optimal pathfinding for the parts it already 'solved'? Or would it just have one skill set for it's pathfinding that will start to skew towards being mediocre at both sections but amazing at neither?

[@xxzombiekillerxx9549]

@@ForcefulDragon i feel as though that would require the ai to access long term memory instead of short term which was described in the video.

[@ForcefulDragon]

@@xxzombiekillerxx9549 not necessarily. If the first generations were evenly split between left and right then at the end of the gen it would have seen that right-preference exploration was yielding higher point totals. So a right turn preference would be developed simply from its higher score based results - especially along the first few routes in the game. It wouldn't require memory at all, just RL

[@carterkal2918]

Clicked on this for pokemon, stayed for a philosophy lesson

[@nicky_gunz]

As a lifelong Pokémon fan, this video was incredibly nostalgic to me in such a strange way. I found myself looking at this AI like I was a proud father or a proud older brother. This AI was reminiscent of me as a child too young to understand how to effectively play Pokémon. How to get out of the starting room, admiring the scenery, and having a feeling of dread the first time I put a Pokémon into a PC box not knowing that I can easily get it back, etc. - all things we experienced as children. Needless to say, I was beaming ear to ear while rooting for the AI to discover by trial and error like we all did has kids. Hats off to you sir, what an amazing video, and thanks for making us look at AI as the “new generation” for us to teach, root for, and be proud of!

[@MrHerbertandreasson]

How beautiful

[@annslow41]

Yes! I remember 9 year old me thought I knew everything about Gameboy games at the time. Most games I owned I could complete in under an hour. I remember reading through the manual in the car on the way home... Seeing the awesome fire lizard, how it evolved into an amazing dragon, and how I was going to pick that one, not some stupid turtle or grass guy. After an hour of learning the game I remember thinking to myself, "I must surely have the big fire dragon by now, though my guy looks the same as when I got him..." Only for my mind to be blown a few minutes later as my Charmander evolved into Charmeleon! ”What??? I've been playing this long already and I'm only scratching the surface???” That was when I, myself, evolved into lifelong fan. (Even if the newest games have been total sh!!t)

[@PrinceTalonSabre]

Bro depositing my pokémon in the PC Box traumatized me too. Haven't been in a pokécenter since. 😭 I much prefer excessively admiring the scenery in the starter town. Fr tho, a lot was relatable. Exploring every pixel of the map, learning what moves to use in battle, discovering dead ends and progressing.. Those were the good ol' days. Admiring the scenery is actually something I do a little in new games, as well. 🙃

[@vicarrighi]

Dude this was mind-blowing, the explanation and the editing were crazy good, best AI learning video I've ever seen

[@McDonaldsCalifornia]

Incredibly well made video! I think your resourcefulness and ability to explain things in non-technical terms shows a deep understanding of the topic. Plus the storytelling is top notch

[@diegoatila2633]

congrats on the project amazing!

[@SpaghettiRoad]

This is the coolest video I have ever seen! You did beautiful job visualising all the games being played simultaneously.

[@brubru9663]

Where are the uploads

[@kratangg-arang]

YO the visuals and quality of everything in this video are wild! I could not ramble long enough to explain everything I love about it. The diagrams, explanations, overlapping AI attempts, just AAAA. AI videos like this are an amazing experience to watch, but imagine very difficult to make and monitor. It was wild seeing that this was your first video on a now 2k sub channel. Seeing it learn was also nostalgic to a degree. You could really feel its ‘personality’ by the end. Seeing its trials, learning behaviors, and what it was capable of was such a journey. I can’t wait to watch your channel grow!

[@mishi6525]

I'm deeply impressed by how the AI, despite being non-human, developed opinions and experiences so similar to ours. Huge respect for the dedication it took to create this.

[@JohnnyNatrium]

It did nothing whatsoever of anything that you said there. You are falling into logical fallacies by attributing human experiences to the outward behavior of a completely braindead brute force bot which was fed explicit formulaic instructions.

[@halfpace1462]

Well ​@@JohnnyNatrium, it's incredibly common for humans to attribute human qualities and personify things that are clearly dead/braindead. It's common human behavior. The same reason many people believe plants feel pain when being cut or bugs can feel love.

[@JohnnyNatrium]

@@halfpace1462 Of course. Did I say I am surprised that humans anthropomorphize things? I'm taking issue with the fact that people are mistaking this poetry for actual scientific homology and coming to almost scarily fallacious conclusions and claims based on this bias, including the narration in this video.

[@halfpace1462]

@@JohnnyNatrium Well said. You can't blame the narrator for this as this increases engagement in the viewers by a significant margin as seen by the comments, I agree that people are acting like this is actual poetry when it's not, but there is a certain comforting feeling about the process of the AI learning even if it's just a brute force robot. It's hard to take issue with the things people interpret as because it is simply human nature, even if taking issue with them is understandable.

[@bob8mybobbob]

@@JohnnyNatriumand I think you’re misinterpreting people’s comments. Of course the AI wasn’t actually traumatized by the computer. But that’s the best way to describe it using common and brief language. And someone saying “the AI reminds me of this human behavior” is probably not someone who thinks that the AI is developing sentience, but rather pointing the hilarity of similar actions by two wildly different things.

[@anguskeenan4932]

My favourite bit was when it named the Squirtle ‘AAAAAAAAA’ so relatable

[@JanviChawla]

This is absolutely insane and so much work went into this! It's so cool to see how the AI learns like humans tend to

[@longbottomleaf6918]

This brings back so many fond memories of playing Blue version on my pokemon edition GBC, learning how to evolve pokemon for the first time and finally getting through the forest. I've played the game so much, 25 years later I can STILL visualize the entire route of Rock Tunnel.