I thought this was going to be a SrslySirius spoof video.

Instead it's some nerd whining about poker.

What a disappointment.

To be fair: I reckon I could probably beat the starcraft AI if I got to slow the game down to a level where my human slowness wouldn't affect things, let's say 10-100x. I would still not make better decisions than liberatus if I got to take 10-100x as much time as the bot to compensate for my slower computing power. And I think I'm better at poker than starcraft.

Some points for this discussion:

- The headsup match was a narrowly defined thing. Every hand reset to 200bb stacks, this indicates humans were still better at NLHE headsup overall because the AI wasn't advanced enough to adjust to different stacksizes yet. I'm not sure the brute force method that is used works for all stacksizes, at least the decision trees will be much more massive. Adjusting to 3-10handed and antes requires exponentially more. You need some kind of intelligence in this case if computing power is limited as it is. I wouldn't be shocked if things advanced to this point in a year though, given it was a bad reg 2 years ago and the best reg one year ago at 200bb headsup.

- I don't claim to know how GTO works perfectly, but generally the further you deviate the more you lose to it. I guess there will be some decisions where you accidentally break even with a suboptimal play, but in general I'd expect a GTO bot to win more at nl50 than the best NL50 reg, the best NLHE player at any stake. A non-perfect bot had a pretty chunky winrate over 4 very competent humans, after all, it would probably beat me by 2-5 times that winrate. But perfect exploitable play will of course still win more than even GTO, but it's always at risk of losing if people find the adjustment.

https://youtu.be/2dX0lwaQRX0?t=1618

it seems as though it would be pretty easily capable of dealing with antes and variable stack sizes, it only adds 1 extra variable preflop (stack size) and changes the size of the initial preflop pot.

They did it this way to make it faster at making decisions preflop/flop by essentially precomputing all possible preflop and flop scenarios. Considering when they played the games, liberatus was a little slower than you would've hoped on the streets where it wasn't precomputed (turn/river), and I'm guessing they did so for the sake of watchability without sacrificing the core of the experiment which is to show off the effectiveness of their cool machine learning technique.

That's kind of my point, it requires a lot more time/computing power to do all the things than one narrowed thing. It can beat humans at a certain precision given an amount of time we find acceptable, but it may need an unacceptable amount of time to get the same winrate if you keep changing the variables because it's not pre-calculated.

Also, changing from 2 to 3-10handed is obviously a much bigger change than adding antes or playing 50 or 500bb stacks. But maybe I underestimate how much it was already capable of a year ago.

pretty easily? zzzzzz

How much less utility to a society do I produce compared to a casino? You know, casinos that PREY upon PROBLEM GAMBLERS to DESTROY LIVES yet somehow aren't held to the same moral standard that I am.

lol, wow. You're pretty dumb, guy.

the professional poker players that played against libratus are Uncle Toms

They should be deeply ashamed, but they're all a bunch of math nerds and math nerds never understand or appreciate the bigger picture. Luckily, I am the type of nerd that has a sufficient lack of Aspergers to get what's happening.

or you have too much

There's a pretty solid argument that casinos are negative utility to society cuz they actively exploit and enable addictions like gambling, alcohol, tobacco, or that they are marginally positive because they provide entertainment and are a useful social space.

But there's no argument for them having 0 utility. Which is what being a poker player as a career is. It brings you, the individual gambler, joy and fulfillment and the opportunity to make money in a way that doesn't make you want to blow your brains out, but it's not doing **** to provide any type of good or service or anything tangible for anyone else.

there's nothing wrong w/ being a 0, utility isn't the only thing that matters and plenty of people make a living being actively destructive like you said, but its still true.

That goes for all sports in general.

Tiger winning at golf really doesn't help anybody but him and his sponsors

Phil ivey winning at poker doesn't help anybody but him and his sponsors.

Life is a game of poker if you dig deep enough.

dude what are you talking about? tiger receives millions of dollars to PROVIDE ENTERTAINMENT, which obviously has a massive amount of utility given the number of people that find it entertaining. the nerd in this thread is just trying to win some money from other poker players, providing 0 utility as the other guy said

giving an AI more move per second than what is physically possible to accomplsih with a mouse and a keyboard is legit cheating though so is giving it greater sight than what a screen + minimap can offer

I actually agree with some of OP's points.

Libratus was successful in proving what it set out to do. I think a perfect GTO bot could out-winrate basically any human in any realistic game environment. Also lol at OTB not being able to out-win the best 50z reg. I mean maybe it'd take him a few hands but yeah.

But to OP's point... It would have been better imo if there was a 15 sec time constraint, a time bank, and only 6-8 betsizes. Even a lesser computer/strat will crank out some EV vs the human by using a large number of very precise (predefined) ranges/betsizes. It's like asking the human to beat a calculator. Yeah it's no-limit, OK, but that aspect is a little outside the spirit of what I'm more interested in.

Anyways, people with PIO-live-advisors are already crushing the bosses. So it's not surprising brilliant people with a pile of $ behind them can beat the best humans.

Nobody is "giving" the AI more moves. It's simply a feature of computers that they are faster at performing simple actions. There is no limit to the number of moves per second in the game rules, so by definition it isn't cheating.

That being said, I agree that when we're discussing AI in real time games like star craft its probably helpful to try to isolate out the AI's mechanical advantages (e.g. actions per second) to figure out if its actually smarter at the strategic/tactical part of the game.

Poker has no mechanics, only strategy.

In all of human advancement and technology you're still surprised that a computer can beat a human being at XYZ?

It was a totally fair match, it's just our technological capabilities have exceeded the calculating ability of our own brains. Just look at today's chess & go engines & more pertinent the AIs like AlphaGo, AlphaZero (& Leela more recently in chess if anyone follows that).

Tldr human beings can't beat computers anymore, so what?

yes its cheating because the game is meant to be played with a mouse, a keyboard and a certain screen size. those are the tools and the AI need to simulate them or he's playing another game

The AI thinks faster in chess and its a feature but allowing it to have 50000apm in starcraft is not a feature, its a cheat. Its physically impossible to do while outsmarting stockfish in chess is possible in theory.

To make this even you would have to somehow 'plug' the mind of the starcraft pro within the game so he can play without the physical limitations.

live in game pio isnt cheating come on guys just like a hud get with the times.

I agree with this and always thought they'd have to create a robot that's not connected to the computer the sc2 game is run on and that actually physically controls a mouse and keyboard and has to watch the computer screen to make decisions/actions. Until then it's not a fair comparison, because moving your hands and mouse rapidly and accurately requires a lot of calculation as well and it's the type of calculation where a human isn't beaten by an AI yet. The vast majority of difficulty in sc2 is execution, not understanding. Until a robot can do that, they're not better at playing sc2 than humans, but of course that day will come as well.

Though I guess limiting APM to about 100 is the next best thing (yes pro gamers get like 250+ but most of that is redundant, I doubt they make as many as 100 true actions per minute).

fair point

Progamers have very high APM during intense parts of the game, like when they are fighting and macro-ing at the same time, albeit lower in SC2 than in SC:BW because you have more hotkeys and can bundle buildings together much smoother in SC2 than you ever could in SC:BW. Like in SC:BW all the koreanpro´s were racking up 300+ apm during highintense situations aswell as in the first 5 minutes of the game, but obviously it was just them keeping warm/warming up by spamclicking stuff while waiting .

Id say 70-110 for the most of the game and 160-180-ish during highintense situations in SC2 (which can be anything from a few seconds to like 1,5minutes) would be a ballpark of what most pro´s are doing now.

Shouldnt it be trivial to code that the AI can only use what it can see on minimap+actual vision and not have automaphack and also limit its apm? Like give it a certain amount of minutes a game it can go up to the upper threshhold of what the pro´s achieve.

Yes, it wont be the same as omnishakira pointed out still but the next best thing until we got robots so "aim-execution" gets to be a thing too.

Yes, but you still have a massive advantage if you don't actually have to move your body parts to execute. I'd probably beat anyone in the world if I had a few weeks to prepare and my thoughts were just instantly executed flawlessly on screen. If I'm playing starcraft I'm using a very small % of my brainpower to think strategically, the vast majority for pro-gamers is just executing a fairly shallow strategy tree as well as they can.

Isn't POKER on ESPN also? must have entertainment value. A handful of Poker players make big $$

Would an aspiring ameteur golfer bring the utility? Only A handful of golfers make big $$ I don't think too many people go watch the club championship at local course.

In business "the nerd" is always trying to steal the competitions $$.

SPOILER: There is no meaning to life--If a person can find a way to live while playing poker & is content I find no less utility in that than selling latte's at starbucks.

I will agree online poker as a career is pretty dismal way to earn $$

Hi again, I fixed your SPOILER above.

you seem to think utility is about whether you happen to think something is ok or acceptable but it is a more defined concept than that

yes of course televised poker also provides entertainment, so those guys getting paid for that if other people want to watch is all good. an amateur golfer isn't a career. if he makes it to the pros then fine he'll provide entertainment and earn a living, if not, he'll do something else. the point of the post you originally replied to was about the butthurt grinder who felt it was unfair that it was more difficult for him to win money from other poker players while providing zero utility to society.

and your business example isn't correct. businesses exist to provide goods and services (which must have utility) to consumers who choose to pay for them. no utility, no customers, no business

Businesses exist to make $$, successful ones have found which goods & service(utility) is profitable then exploiting it. When Orginizations are worried about getting goods & services to people without $$ they call it charity.

my utility to other poker players is I bring an unknown dynamic that they can test their poker knowledge agains't. With poker we have a way to measure SKILLS with the fellow human, which I think is more important than most of the "utility" in the world. The score just happens to be kept in $$.

yeah we definitely veered of GTO/AI topic, so heres some AI

"Google is beating hospitals at their own game — at least when it comes to death risk assessment. Google’s Medical Brain team has begun training its artificial intelligence system to evaluate the risk of death among hospital patients, and results for the time being are more accurate than those provided by existing medical tools." (copied)

"Google first detailed its new system in a paper published in the Journal Nature in May. At the time, company researchers noted, “These models outperformed traditional, clinically used predictive models in all cases. We believe that this approach can be used to create accurate and scaleable predictions for a variety of clinical scenarios.”" (copied)

Google's algorithms are matching doctors abilities and AI will be threatening their jobs and countless other jobs in the near future.

So to think poker will be profitable against AI is a mere pipe dream.

Somehow you managed to completely miss his point, maybe try reading before continuing to post, unless you want to just keep writing for no reason