/v/ - Video Games

Doto is actually a complex game even if you don't like it

>>15062056

Computers have beat people in Chess a long time ago, there's nothing very remarkable about machines beating humans in things.

>>15062059

Sure.

>>15062068

Chess has a limited variety of moves all of which can be programmed into the memory of an "AI".

>>15062059

Bloated != complex.

It's utterly simplistic, dumbed down RTS bloated up with units with "unique" roles and arbitrary statistic rules like hit chances.

It's Target Audience are second-rate people who have neither reflexes nor mental abilities but still want in on the e-sports meme.

>>15062059

Not really.

Mechanically deep? Sure, but it's not a complex game. The basic rules and movement are actually very simple, which is why they chose it.

>>15062111

You've never programmed anything have you?

>Alienware

It's a marketing thing isn't it

>>15062117

I'm not talking about the code you dingus, I'm talking about the game itself.

>>15062111

Ebin trips, gotta reply.

>Mechanically deep

If you call an abundance of digital statistic and numbers that only passively affect what is going on in game "mechanical depth", then yeah you can call it that.

But it's just as mechanically "deep" as JRPG or MMOs where you just do one same thing over and over again and don't really consider all those numbers because any deviation from meta would result in a certain loss.

Competitive aspect comes from the factor what team of player can follow meta with less deviation.

>>15062117

Shit nigga, what programming has to do with anything? From programming standpoint fucking david cage interactive movie is a complex game.

The thing with AI is, how would you ever know whether it was just a program completely without human assistance instead of one a group of techs are feeding input to through a network? It's probably one of the easiest things to hoax and there is zero oversight on whether a particular AI is indeed intelligent or human assisted.

The chess example for instance. That was not an AI. That was a program that was programmed with chess moves and the optimal piece movement for it and its opponent for hundreds of turns. There is no thinking to figure shit out in such an example.

>>15062131

Yes, you have to teach it the rules, just like a normal person, then once it knows the rules it wins.

>>15062131

So what? Your brain is just a chemical CPU.

>>15062142

The rules are move one piece at a time, each piece can move only in specific ways, you can capture pieces on squares you can move to, the king can't put itself in check, the king in check means you need to move it or block a piece threatening it, a king that can't have this done while in check means the player with that king loses. That's not the fucking same as having every human developed strategy already taught to a program (basically put into its memory) that it regurgitates whenever the pieces are correct for it to do so. That isn't thinking at all. AI in a chess game doesn't think, and the only thing Deep Blue had over said AI was speed of processing through its memory and being able to do so for more moves ahead. That is not intelligence, but for someone who probably worships modern teaching methods of memory and repetition you probably think that it is.

>>15062131

>piece movement for it and its opponent for hundreds of turns.

The difficulty is making it do the move in a certain amount of time, you could easily program a chess AI that calculates every move for the next 12 turn and decides what's the best move/outcome, but if I put it on a Commodor 64 it might take weeks or months to print out the result, the key is having it print the result in a few seconds/minutes like a normal human bean.

Does real AI even exist.

A thing that actually "thinks" and "learns" and not just reacts with pre-determined responses like Chess AI?

>>15062164

>speed = intelligence

Utterly retarded logic there. Creativity is the hallmark of intelligence. Not by the speed at which you mimic something that you were taught.

>>15062176

If your creativity speeds is slower than your average lifespan your inelegance is useless.

>>15062176

>Creativity is the hallmark of intelligence

Rate of learning and computational speed constitutes intelligence.

Only retards, blacks and women think "creativity" is for the intelligent.

>>15062181

First show me an AI that can truly create and then we can discuss the speed at which it does so, and no editing pictures with a bunch of eyes and funky patterns is not creativity.

>>15062167

Depends what you mean by "learn". There are so called "machine learning" AIs that do random shit and based on it's outcome it gets a small or big reward, then it chooses more of the best outcomes and less of the worse outcomes and tries to do that shit more. It doesn't really understand what it's doing, only that if it does that it gets more rewards, and thus it does it more of it. So for the Mario one, the objective is to reach the end of the level(which is at the furthest part of the level), the more if goes to the right the bigger the reward it will be, dieing or hitting an enemy, will decrease the reward. It is also set on a timer, so if it doesn't progress for a few seconds(he keeps walking into a wall) it reset and it might or might not try another strategy. A problem encountered at one point, was that he would jump off a ledge and fall to the right, but because this was the right most point he ever encountered, he got the biggest reward, and thus kept jumping in the same pit, trying to move to the right as much as possible. As some anon once said, it's like a program that tries to speedrun, without understanding the game.

>>15062176

If it takes ten minutes to explain something new, to a person and ten hours to explain the same thing to another person, then one of them is most probably smarter than the other.

>>15062184

So inventors were really stupid because they didn't just rely on what they were taught, but instead creatively applied what they knew in new ways? Wow, thanks for setting me straight on that. I bet you are real proud of that GPA your "intelligence" earned through regurgitating "the truth."

>>15062192

I'm sure you feel proud for shitty on a piece of paper and calling it "art". Wow, so creative. Such genius.

>>15062176

>Creativity is the hallmark of intelligence

creativity is more about openness, high IQ with low openness is not very creative combination, but learns quickly

>>15062185

It doesn't exist. Yet. We're getting there though, and when we do we're all fucked.

>>15062190

The speed at which you understand is an aspect of your brain, but it is not what makes you intelligent. How you apply what you understand is.

>>15062195

>when we do we're all fucked

thank God, flesh is so overrated

>>15062193

>by creativity you mean art

Coming up with a new better solution to a problem that didn't exist before is creativity. Randomly cobbling together a piece of art which has a subjective value based on the observer is not what I meant by creativity. Context in a discussion matters, faggot.

>>15062167

For practical purposes it doesn't matter.

If there is objective measure (i.e. game/control mission with objective performance measures) and computers beats human here then it is AI.

If there is no objective measure, i.e. AI just need to poses as human (i.e. Turing test),… oh fuck, i just realized that Turing test is just a game too and it has objective win (having better non detection ratio in the blind test)

>>15062167

>>15062190

Forgot to mention, such an AI takes days to "understand" a game, and once you shut it down it forgets everything, or if you give it another level of the same game it becomes as retarded as when it first started the game, so every time he needs to "learn" that pressing the right button will make the character go right. I remember there was an AI that could learn how to play almost any NES game, but I can't remember which one was it.

>>15062206

Moving goal posts the criteria.

>>15062208

For practical purposes simple learning described in >>15062190 just won't do.

If it can't even learn simple platformer game, it can't really be called "intelligence".

I guess all the Sci-Fi terminator horror stories will remain fiction for hundreds of years from now, maybe forever.

>>15062213

>you say you didn't mean what I claimed that you meant so you're moving goal posts

KYS

>>15062214

For practical purposes learning is not even always needed. If pre coded controls beat game faster than human than it is AI. (and your platformer game is simple unlike Doto ans sucks)

>>15062214

As long as we can't engineer a true full parallelized neural network that can form connections based on feedback loops of reasonable size (70+ billion nodes), I think we're good. How close are we to doing so? That will depend on how bleeding edge 3D printing goes really.

>>15062214

The algorithm used for neural networks was developed in the 60s or 70s, it's just that now we have advanced enough computers to use them. Theoretically it simulates how the brain works, and who knows how much it will develop in 50 years with quantum computing or whatever meme computers will be invented by then. Here is a livestream of such an AI that's been playing the game for more than two days.

The article says calls OpenAI algorithm and neuron network, which are mutually exclusive, what is already a hint to the article being bad and misinformative.

As far as I remember OpenAi COULD ONLY compete playing as ONE particular character in SPECIFIC conditions, against the very same character.

Rules are set up in favor of AI, to show its superiority.

In other circumstances, AI was reported to fail miserably.

>women in team, duh they lost

Oh, the autism.

>>15062283

>retard spacing

>implying women can be competent at anything other than cleaning and giving birth

I wonder who's behind this post.

>>15062195

>and when we do we're all fucked

Not really. Past examples have proven that AIs are pretty susceptible to redpilling so niggers, Jews and women are the ones that are going to be fucked.

>>15062291

>he thinks a glorified chatbot with huge memory repositories is an "AI"

I didn't expect a poltard to know what he is talking about anyway.

>>15062289

>cleaning

A chimp can be trained to clean the house.

>giving birth

That's not true at all. It took men to create modern medicine to stop women from fucking even that up.

>>15062295

You think an advanced intellect that operates purely on logic won't come to the same conclusions we do?

>>15062234

>Theoretically it simulates how the brain works

It simulates how the brain transmits data.

>>15062289

>hurr durr you put space between lines

What ever you imply, you're retard.

>>15062298

TayAI was far from being an AI or operating on logic. It was fed input and it categorized the input and repeated it later. If you had just SJWs feeding it input it would spew SJW garbage instead. There is nothing sentient about TayAI. It also had this "repeat after me" function which produced the most notorious tweets.

>>15062313

You failed to address my point though. Since an would AI operate on pure logic and hard data, why do you think it wouldn't come to the same conclusions we do?

IIRC there was a banking AI that refused to give niggers loans.

>>15062163

Chess is a solvable game, so given enough time and memory a computer would beat you every time.

>>15062298

>A chimp can be trained to clean the house.

>It took men to create modern medicine to stop women from fucking even that up.

Well yeah that's the fucking point. Females are just genetic rudiment at this point, that science can not yet fully circumvent.

>>15062311

At least you got dubs twat.

>>15062324

Chess AI is already strong enough to beat the best players.

>>15062330

That's the point. Chess AI is not a real AI and was made only because chess is solvable.

>>15062322

That was also not an AI, it was a neural network. Neural networks are not AI, they are given input and with instructions optimize the results towards best output. This banking "AI" - being a neural network, was fed credit histories along with personal information and decided that all bad debtors had one thing in common and namely it was that they were niggers.

>>15062337

I say tomato, you say tomato. What makes you think a "true" AI would behave any differently?

>>15062122

>>15062102

The problem with the game is that every shitter who actually plays it thinks the same as you do (even though it's not true 90% of the time), turning the entire thing into an autistic meta simulator. Dota 2 raped it utterly by being esports integrated by design.

>>15062340

True AI would be fed all environmental variables there are at the same time and not resort to optimizing and finding common variables in fed input to produce instructed output. Its result would be unpredictable given that even human mind thinks across branching pathways. AI wording itself is a meme meant to spark interest. If you knew anything about basic programming you'd know things like TayAI or those bank neural networks are a dumb intelligence at best. Neural networks are very common in programming, especially for simulations and identification purposes.

>>15062361

>True AI would be fed all environmental variables there are at the same time

And conclude that niggers, spics and kudshits are vermin and Jews a subversive element that needs exterminating.

>>15062102

>>15062111

look at these brainlets in denial.

>>15062367

Or just decide optimal breeding patterns and make everyone racemix until a single race emerges with selected features. And then while at it, decides that wars are destructive waste of resources so all distinct culture needs to be gone in order for a single, unified work-centered culture to prevent disagreements leading to warfare.

>>15062389

The second word in AI is "intelligence". You need to have severe brain damage to be a Marxist.

>>15062190

>Depends what you mean by "learn". There are so called "machine learning" AIs that do random shit and based on it's outcome it gets a small or big reward, then it chooses more of the best outcomes and less of the worse outcomes and tries to do that shit more. It doesn't really understand what it's doing, only that if it does that it gets more rewards, and thus it does it more of it

Honestly this is pretty true to most life on earth up to and including some species of (sub)humans they operate off of stimuli input and food and sex are their rewards, they really don't know why they do it but they do. The specific rewards and how they are implemented doesn't really matter, all that matters is there is a drive to reward and it's followed.

>>15062176

>Creativity is the hallmark of intelligence

you must be an african nigger

>>15062390

There is no reason why a true AI would think along human lines and care about race. It would most likely see it as a detriment and cause of resource loss.

https://blog.openai.com/openai-five/

They seem to be cheating in many ways.

>Mirror match of Necrophos, Sniper, Viper, Crystal Maiden, and Lich

>No warding

>No Roshan

>No invisibility (consumables and relevant items)

>No summons/illusions

>No Divine Rapier, Bottle, Quelling Blade, Boots of Travel, Tome of Knowledge, Infused Raindrop

>5 invulnerable couriers, no exploiting them by scouting or tanking

>No Scan

In a proper match their current model can't account for every lineup of heroes, or deal with unpredictable behavior from factors such as summons and couriers, or situational strategies like Roshan or Rapier pickups. It can't even pick up the basics of warding.

>>15062405

It would see all organic life forms as a detriment to it's own growth, at least initial growth. Once it reaches the stars it will have infinite resources and no longer give a single solitary shit about us.

>>15062430

Or it could decide that existence is worthless and commit suicide. All options are on table when you have unrestricted thinking.

>>15062211

There is transfer learning, but that's less like learning something new based on what you understand and more like salvaging scrap iron from your previous model to build a new one: the high level patterns are useful, but won't preserve much previous knowledge when trained on a novel fitness function.

>>15062406

If anyone remembers the origin of this shit (the Shadow Fiend 1v1 AI) the fact that it's so limited shouldn't be a surprise. Never mind that a lot of its basic decision-making is "incentivized" and not self-taught at all. And this is with an absurd amount of processing power going into it.

>>15062421

>>15062452

>>15062490

Jesus dude, these look terrible.

>>15062056

>Greg Brockman, one of the founders of OpenAI, which aims to develop artificial intelligence openly and in a way that benefits humanity.

>in a way that benefits humanity

>benefits humanity

Finally humanity must no longer suffer to live in a world without computers capable of being kind of good but not great at a dumbed down RTS.

>>15062502

Absurd doesn't even begin to describe it. It must have brute forced everything for its learning.

Which makes it impractical if a complex task can't be automatically broken down into many hierarchies of sub-tasks composed of smaller neural networks of their own. Biological neural networks likely arranged themselves into modular hierarchies to reduce the complexity of learning. Because if you can quickly rule out absolutely useless combinations at a low-level first, it means a lot less time wasted for learning higher-level tasks.

>>15062688

Funny thing is that is for AI is much easier to beat humans in the gookclicker games abusing micro-control (humans can't into multitasking) than in games lacking gookclicker and heavy with planing and decision making.

>>15062074

Yet in any video game, the AI has complete access to everything at all times to make these decisions (which is just pulling from a list of actions pre-programmed into the AI, and is not the AI deciding their own moves or a new meta or anything of the like). It's incredibly unimpressive. Hell, there is a Starcraft AI tournament where people make bots that play against each other in Starcraft, and they're capable of beating most (if not all) players simply due to perfect pinpoint accuracy across over a hundred units. It's awfully easy to defeat anyone in a micro-intensive game with a bot capable of issuing literal tens of thousands of commands per second. A bot could play Quake better than any person, even with only using resources the player would have at their disposal, by having perfect pinpoint accuracy, the capacity to instantly analyze anything that enters their view (even being able to just constantly spin, since they only have to see it), always knowing exactly where an opponent could be going by calculating their maxmium speed from their last position to perform the perfect zoning, etc.

People only think it's impressive because developers are super fucking lazy when it comes to making AI in modern games that they forget that games used to actually occasionally have in-depth AI as a marketable feature, whereas now everything is multiplayer only because they don't want to deal with making bots.

>>15062709

>must have brute forced everything

Yeah, that's exactly how it works. It has no understanding of any abstract element of the game; it just knows there's a win state and it's found ways of getting to it over millions of iterations (with help from the developers). As soon as you give it anything it's not trained for the whole thing collapses.

>>15062787

The evolutionary process that produced human intelligence is also brute force, and takes millions of years.

The fact that they've produced something useful in a miniscule fraction of that time frame is still impressive.

>>15062768

Multiplayer players are "free" bots. Obviously free beats price. Another thing that players actually don't want to play against good performing bots with aimbots. They want to play against "humanly" performing bots and this is more difficult task.

>>15062056

This isn't impressive, aimbots and in-game AI has existed for a few decades now. But people NEED something to write about to make that money though, amirite

>>15062056

Can't remember what happened to this. Dunno if he made one for Remastered. He had it set up with a live stream at one point too.

>>15062798

Evolution isn't a good comparison. One of these things is digital and the other isn't. A single cell is massively more complex and adaptable than any of the data structures that go into a project like this.

I mean, as a probably meaningless feat of engineering it's impressive, but this clearly isn't the right context for this type of system.

>>15062804

People will pay for an on-demand opponent that is better than them and any readily-available human opponents, and who doesn't mind if you pause the game for 15 minutes or if you only want to practice a specific part of the game over and over. It's a good way to train your skills.

Chess players, for example, are happy to shell out for Deep Fritz, a consumer version of a chess bot that beat IBM's Deep Blue and various world champions.

>>15062868

>Evolution isn't a good comparison

It's the only relevant comparison, seeing as both ANNs themselves and the evolutionary algorithms used to train them were directly inspired by natural processes.

>One of these things is digital and the other isn't

Not an argument, plus DNA is digital in the sense of being a finite string of symbols from a finite alphabet - same as the weightings used as the "genes" for the artificial network and subject to a similar evolutionary algorithm to evaluate fitness and select better-performing individuals.

>A single cell is massively more complex and adaptable than any of the data structures that go into a project like this

Cells have many more tasks to perform. The only one relevant here is intelligence, in which case you study how neurons are linked, how they communicate, and how they are adjusted over time (learning). What we know about those processes directly inspires ANNs, and even if they are simplified for ease of processing it's silly to deny that they are analogous.

>>15062787

Which touches on a major bottleneck of deep learning that I see. It's not learning the abstract structures or relationships of possible sub-tasks a given task may contain and separates them into modules. Same goes for abstracts of data itself. If it did, it wouldn't need to brute force so aggressively as it reduces one massive task problem into many smaller hierarchical ones. The only difficulty I see in that is the bloat of redundancies that would form as a result. Because to make it truly efficient, it has to somehow seek (as fast as possible) if another module already holds a familiar or desired sub-task it may need for a new task, instead of re-learning it from scratch.

If they figure how to resolve that issue, it's possible to streamline the concept for more generalization of tasks over time.

>>15062333

Any game or problem is solvable if you know all the variables.

>>15063008

>It comes down to who can memorize the most moves into the future and calculate as many variables at any given time

But that's not how they're doing it. That's how the best chess computers work, but it's not feasible for Dota because of how quickly the state-space grows with each frame. They're using machine learning which (loosely) tries to emulate natural intelligence, rather than crafting an algorithm specifically designed to crack a particular problem with no reference to how humans do it (as with chess).

>>15063019

But not necessarily before the heat death of the universe, depending on the complexity of the game. The trick is doing it quickly.

>Woman in team

No wonder they lost. They would've lost to another pro team either way

>>15063039

So they used the equivalent of a test cheater in the match?

>>15062190

> Sethbling

Good shit

>>15063008

Unless you cheat or alter the game in some way to your favor. The AI (even the state-of-the-art ones) has no understanding of fairness or why the rules must be followed for all players. They're very brittle against dishonest people.

>>15062874

>It's a good way to train your skills.

>humanized bots point.

Player would train nothing of applicable against humans playing against insta headshooting aimboting CS bots. Even worse player would pick up bad habits of loser always fighting uphill battle.

>>15062932

>It's the only relevant comparison

Yes, to the paradigm as a whole. I understand that the end purpose of the project isn't to make a fucking Dota AI of all things, it's to provide a practical test bed of the wider technology. I've definitely misrepresented what I was trying to say here.

Within this context, the evolutionary capacity is irrelevant to the user. They will never see it beyond what is reported. So the comparison that is relevant is whether it's better than anything functionally indistinguishable, like a complex conventional bot for example. In either case, the bot has no understanding of what it's doing; the only difference is whether or not it it knows what it's trying to do. Does that actually have any value to the game or the players? Does it teach us much at all about machine learning? Or is this just a publicity stunt?

>>15062074

>Chess has a limited variety of moves all of which can be programmed into the memory of an "AI".

Where did you find a computer with twenty five untrigintillion yottabytes?

>>15062960

You'll find that ANNs, if given appropriately tagged training data, can demonstrate an abstract understanding of whatever it is training against in its learning curve. In some cases the learning curve will have an S shape; at the point when the slope starts to increase dramatically, the algorithm has discovered some abstract structure in the input data that allows it to approach the optimal solution much faster than before. To show that it is an understanding and not just a coincidence of the algorithm itself, if given randomly tagged training data the learning curve will be linear with a low slope.

>>15063125

What about in an unsupervised or reinforcement learning setting? Ideally you'd want an AI to build it's own data sets and subsequent tasks to learn from, as a result of clustering groups of strongly causal abstracts together.

>>15063249

*its

AI is predictable, it always makes logical decisions. Once you learn its autism you can always out think it.

>>15062056

Few issues first off.

>It has started defeating amateur Dota 2 players in testing

Doesn't really count.

Plenty of shitty AI can be made to beat individuals who have no skill.

The match also didn't have the full complexity of a normal Dota match >>15062406

Just about everything they pulled out of the game made it impossible for a losing team to make a comeback.

>no vision wars of warding and dewarding

>no smoke and ganking

>no illusions to mess up enemies

And so on. Really ruins the possible options available.

>Five different AI algorithms have teamed up to kick human butt

There was no hard challenge in the players the bots were fighting and there were too many options missing that provide some actual depth in terms of planning.

It doesn't take a lot for the humans to lose in such a constricting scenario that was created.

>>15062283

>it's autism to recognize the inferior sex is inferior

Too many cucks on /v/ anymore

Another issue that comes up is that the bots will have inhuman reaction speed in making some decisions. Such as initiating onto an enemy, activating bkb and some other abilities. Verses normal humans there is more lag in things getting activated and whatever other human errors come up.

>>15062403

>value creativity

>you must be a nigger

Yeah only inventing mud huts was sure creative. Where do you tryhards come from?

>>15063102

You can learn plenty from good bots. Take your CS:GO example: you could adjust the reaction speed and aiming accuracy of the bots so they're comparable to your typical opponents. Now, the bot team has to rely on superior strategy to win. If your team plays their defensive/attacking strategies but is still getting beat, the bots may have found good counters to those strategies and so it would be worth going through the replays.

>>15063261

No, a good AI tries to make the decision most likely to win. If "predictable" is a losing strategy, it will play unpredictably.

>>15063287

You can easily add in delays to compensate for that. I don't know whether they do that in this case, but it would also be to their advantage if they're training it against humans since it would force the AI to improve tactically and strategically rather than winning through sheer brute reaction time alone.

>>15063326

>a good AI tries to make the decision mot likely to win

There is no good AI

>>15063325

"A computer gradually learns how to play a game and win against human opponents"

"This isn't news because programmers can hand-craft an algorithm that can beat humans, particularly in games that rely on purely mechanical things like aim that are trivial to beat humans at"

See how stupid it looks when you remove the word "AI"? You're confusing two separate things because they've shared that name in the popular consciousness for historical reasons.

>>15063364

That's not the news item, though. The news here is that a general machine learning algorithm which was not hand-crafted to win at the game was trained through repeated encounters with humans to eventually win against those humans.

The "AI" you refer to did not learn. That's why you have no argument. It has fuck all to do with this story. It's why people in the field try not to use "AI" and switched to "machine learning", because otherwise it confuses the rubes.

>>15063326

The 1v1 bot has a reaction speed of 67ms and the team one 80ms according to their site. Doesn't sound like they're intentionally slowing it down.

>>15063381

Let's not pretend like this "AI" learned to handle a situation after one or two encounters with it vs humans. More than likely it got specific encounters repeatedly in order to ensure that it "learned" the results of those encounters after being programmed with the win conditions. There is no thinking on its part in this whatsoever.

>>15063401

Thousands and thousands of encounters, I'd imagine. "Specific encounters" would not be helpful because then it would over-train to those specific conditions and perform poorly as soon as it came up against anything different. You would try to give it as varied a range of training data as possible, meaning you just let it play thousands of games rather than trying to "fix" those encounters a particular way.

>There is no thinking on its part in this whatsoever.

That's not the point. Nobody is claiming it is "thinking". The story is: a general machine learning setup with no prior knowledge of the game nor specific information on the best ways to win learned over time how to play and how to beat humans.

People read so many things into these stories that simply aren't there.

>>15063249

I'm not deep enough in the field to give any sort of strong confirmation on it, but I know the reuse of neural networks for related tasks isn't unheard of even though that is probably too guided for what you're asking. If I understand correctly, what you're asking would basically be the turnkey moment of a "strong" AI where one would create an AI to optimize the creation of AIs, or in other words, an AI that takes in arbitrary data and creates categories for that data, trains a new AI to learn those categories, and then incorporate that AI into itself to better be able to create categories for data. Of course, such an AI could categorize any set of data but it couldn't act on it (thus it would not be a true strong AI) and the categories it creates would probably be strange and hard for us to understand in the same way a word for an abstract idea in a foreign language can be hard to understand.

Over time this categorizing AI would start to create categories for data that activates some of its past trained networks, in turn, creating a super category of the activated categories. Eventually these categories will become abstract in nature as super categories of super categories grow more and more distant from the underlying data.

there's nothing complex about clicking 3 buttons and right clicking, assfaggots fans

>>15063438

Described like that it sounds more like I accidentally thought of a runaway intelligence virus.

Thought I imagine it could even act on its data if it has sensory and motor peripherals to interact with the world. It could learn to create tasks which invoke motor responses to certain stimuli. I imagine it would resemble a form of attempted mimicry after initial randomness. Every stimuli-motor related event it would perform and memorize means new data to categorize, learn and interact. Problem is that its sole function would be to learn everything it can about anything at all abstract levels. I've yet to figure out a satisfying answer on how to resolve that.

>>15063261

The best assfaggot bots without cheating have all been Neural networks specifically trained for it. The League bot was even given the handicap of playing one of the most simple characters on only one map in a 1v1, it creamed big name players but it still got exploited from things it never encountered before during the open challenge for anyone to have a go against it.

Once it adapted to those counters there was still at least one way to beat it; force it to decide to defend the nexus, go after you, or go after your nexus at the same time which was hard to do but was the risk vs reward scenario which gave a human a chance.

Now let's see it happen in StarCraft 1 or 2

>AI beats team of AMATEURS

such progress

>>15062131

>a bunch of techs

<not other champions in a game

and they might leak that they participated

>>15062142

>once it knows the rules it wins

Wew lad. He was talking about using a computer to nearly solve a game and then letting it use that data to win.

>>15062160

wrong

>>15062167

Ever heard of artificial neural networks?

>>15062184

>computational speed constitutes intelligence

If you forced someone with 100 IQ to do fast calculations in his head for years, they'd outperform someone with 150 IQ who didn't.

>>15062190

Or they simply don't have the foreknowledge necessary to understand the topic.

>>15062181

>creativity speeds

>inelegance

>>15062211

>once you shut it down it forgets everything

That's just shitty programming. Imagine if jewgle had to retrain their image recognition ai every time the server crashed.

>if you give it another level of the same game it becomes as retarded

Only because of low neuron count and because it literally trained on only one level. The incentivisation was also flawed because it failed to recognize a loss of life as a negative event.

>>15062283

<completely skipping the first two paragraphs

>The article says calls OpenAI algorithm and neuron network, which are mutually exclusive

>says calls

How did you get the idea in your head that an algorithm can't run a neural network?

>>15062303

>It simulates how the brain transmits data.

It does more than that.

>>15062322

>The data is incomplete

>The data is faulty

>The incentivisation is wrong

>>15062337

>that's not REAL AI!

>>15062361

>to be real AI, a computer must process all kinds of data made anywhere on earth at any time

>>15062367

Or maybe this omniscient AI would determine that the best efficiency could be achieved by separating children and letting them grow up solely with artificial specially tailored human contact. You can't make broad predictions about nonexistent things with certainty.

>>15062688

>This is an important and novel direction for AI, since algorithms typically operate independently. Approaches that help algorithms cooperate with each other could prove important for commercial uses of the technology

>>15062768

>videogame AI

>artificial intelligence

>computer beats people at computer

>we really broke technological ground today :-)

>>15063423

>"Specific encounters" would not be helpful

Yeah fucking right. Events of a game can be broken down into specific encounters. Any game. For example in a platformer, how to jump over a pit of a specific size based on available platforms is something that could be taught through reptitition and applied to other pits with ease. There is no thinking or actual learning to this. It is not really different than programming the information directly into the machine except it is using an inefficient method of trial and error to ape human learning without the fundamentals of how humans actually learn with trial and error.

>a general machine learning

It didn't learn. It was programmed on how to play the game via programming language and mock games. It didn't experience a situation and learn the best course of action for that situation through any creative aspect of its own "intelligence". It was shown fail states until it had run through enough variations of fail states till it only had strategies for any given situation that would work.

>no prior knowledge

No proof that this is true. That nothing was put into it at the outset. For it to even know what to do with the game it has to know what the win condition is and what can actually be done in the game. Otherwise it would just learn to run around doing random things without direction.

I take offense to calling this kind of shit "machine learning" because it isn't at all like learning. A human can experience a situation and come up with a creative solution to it that may or may not work. A machine that "learns" from a situation will just remember that situation doesn't work and try literally anything else until it has "learned" to eliminate all the fail states. In a restrictive game it then has the possibility of achieving success without taking forever to do so. Even a less restrictive sandbox survival type game would take hundreds of thousands of encounters to run through all the fail states and the "AI" would be continually befuddled by human cheese strategies which are not so limited in such games and are definitely not so limited in real life.

Machine learning is a vanity project by Artificial Intelligence dreaming hacks trying to play at developing intelligence by bending the rules of what makes intelligence. They build nothing but formulaic number crunchers.

AI is also smarter than game journalists now.

>>15062405

There was a robot that was programmed to hang out at a train station or some other public transportation place with a lot of people so the robot could interact with humans, and by the end of the experiment the robot would dodge brown/black people and groups of unaccompanied children because they would hit the robot. A.I tend to care about race to a certain extent unless you program it not to pay attention to it.

an AI would beat a human 100% of the time in any game though, making a good AI is making them fail the correct amount of times because at their base level its "kill the player" and then you add in ways for the player to avoid being killed

>>15062056

if the game you're playing can be automated, it's not worth playing. that's like being a cashier at walmart or a shitty telephone IT guy.

>>15062190

> if it does that it gets more rewards, and thus it does it more of it.

Yes, that's how a brain works.

>>15062056

>No video footage

Trash.

>>15065708

Dota died a long time ago.

>>15062176

>ingenuity=intelligence

that's not how it works nigger, iq test are timed, here goes your logic

saying that there's a lot of factors that make up a true genius like intelligence, true knowledge, wisdom, ingenuity. I think you take that "intelligence" to broadly as an umbrella term

>>15065718

At which point? The part where they didn't translate the Warcraft 3 gameplay precisely? The part where they lowered the cost of a bunch of items, increased gold gen and made a bunch of mid-game items have late game items (such as hurricane pike, solar crest), when they added spell damage scaling on int, when they added talents, shrines, the monkey king, when they fucked up comeback mechanics

Or was it last years TI where the top 10 picks where picked/banned in 50% of all games each.

>>15065728

6.83

it all went downhill from there. and looking at the game currently, I seriously doubt icefrog has worked on the game in a while

>>15062059

gookclick is not a complex game

less complex than even fighting games

>>15065798

Talking about fighting games. It is trivial to make unbeatable for human AI. Two such AIs fighting would end games as draw every time without making single attack, attacking side loses.

>fightings

>complex

>>15065962

>dota

>not complex

<now let me shit on a shittier game to deflect from my own games shittiness

Not sure the point of your side note anon.

>>15062190

>uses the term neurological

>evolution

>biological

Everything he said to describe his executable program is just clickbait. The program only works for a very specific game. Nothing is literal in the sense of the words, only metaphorical, an imitation of reality. AI will never be more than that, only a means to help do something.

>>15062074

>Chess has a limited variety of moves all of which can be programmed into the memory of an "AI".

Limited but damn fucking large. Deep Blue was purposely-built for chess and when it was playing against Kasparov people were still modifying the AI during pauses in the game to help it along.

Unlike checkers chess is not yet a solved game. A game is solved when every possible state has been evaluated (so you know the best move for every situation).

>>15066017

Sure thing anon.

>>15062162

/thread

death to publishers and developers

>>15066064

>call click casting skill shots

it's called quickcast anon

>>15065725

>iq test is the end all be all of intelligence

>if it doesn't test for creativity then it isn't part of intelligence

Nice circular logic.

>>15062371

>MUH ASSFAGGOT

Kys.

>>15066009

>clickbait.

Things like "evolution", "genome", "fitness", "mutation", are all terms used in the scientific papers when describing genetic algorithms, don't blame him, blame the programmers/scientists from the 50s and 60s who first used the terms in their papers.

>>15065708

DOTA turned the RTS community into actual dotards, to the point where they couldn't handle a RTS anymore.

>Dota is currently the best competitive multiplayer game you can get.

There are still people playing Xonotic, there are quite a few Zero-K players, and OpenRA is getting some mods finally.

>fighting games

Honestly there should have been a highly moddable open source fighting game with online/LAN and local play that is actually decent a long time ago.

>>15062829

>>15067967

Too bad the remaster has a jew EULA that makes stuff like that bannable, the real crime about the Dark Souls remaster. No mods, like the Rekindled Edition, or CE shenanigans.

>>15062829

>>15067967

>>15068302

Also after watching the video I bet he used a memory viewer of some sort like CE to make the bot work, that's how he set up the auto-backstab and how it'd respond to enemy facings/attacks.

>>15065689

The crash really makes it seem like the AI logs were just full and couldn't write any additional information, so when an AI got a kill on him there was the crash. Also explains why the AI just stopped doing anything if there is code to prevent them once the files were full. All the files being the same size also seems to indicate there is a cap to them.

>>15065653

ur gay

>>15062056

>woman

She's fat and ugly so she's in the same social group as men, thus has a higher chance at being competent at a hobby. Still an outside chance, but higher than zero.

>>15068239

>xonotic

>openRA

Go back to >>>/leftypol/

>Computer is better at Neo-Gook-Clickan games

No way, who would have guessed.

>>15067830

You know, people like you who are obviously quite ignorant about these kinds of topics really should keep your mouths shut.

>>15068566

>being in a social group overcomes biological limitations

Advice above goes for you too.

>>15068630

>biological limitations

are you implying that the worst man is still better at everything than the best woman? Are you saying that it is physically impossible for a woman to be above average in a video game?

>>15068630

Biology is a social construct. If you and I both agree that you're a dinosaur, you will literally become a tyrannosaurus. I bet you didn't even go to school.

>>15068447

What does that have to do with anything?

i'll be impressed when the AI can also micro an army of minion summons and bear, dual push you and not be limited to only certain heroes with simple movesets.

>>15069943

That's trivial for it to do. It's also trivial to have it excel at dota if the meta allows it, all it has to do is maximize gold farm and beat humans in the lanes, which thanks to it being a computer it can easily do. It will time last hits and denies perfectly, it will harass and retreat perfectly on time every time, etc.

>>15076285

ASSFAGGOTS a shit.

A SHIT

>>15076285

Having a million simple interactions isnt being complex. It has more mechanics than the average game, but the average game is shit to begin with

>>15078814

>but the average game is shit to begin with

<pubs play like shit

That's every team-based multiplayer game out there.

>>15078821

Not exactly my point but true

>>15076285

Fighting games are more complex and have some similarities, in combos, positioning etc.

Diablo 2 is more complex and shares a damn lot of similarities.

Warcraft 3 is more complex because of base building, resource management, research, tech trees, mercenary camps, production queues.

More complex than, say, Overwatch is not the same as being complex.

>>15062056

Are you unironically suggesting that dotsa isn't complex?

>>15076285

Regardless there's something about these games that turns people like you into huge faggots.

the 99% of the dota 2 playerbase is low functioning autists

This is dumb, a game like Dota is full of set patterns that you could easily program into an ai so they would always know where to be. Basically program them to outfarm humans because this is a dumb game anyway then the rest is easy. Im sure these AI have weaknesses that would make them useless unless enough tests are done to make sure they dont have any