Why can't higher difficulty equal more intelligent play?

[ggsimmonds]

Re: Why can't higher difficulty equal more intelligent play?

Such as?

You're the one using Chess as your point of reference. What are the applicable advancements other games have made that sports games have not?

Whats the screenshot illustrating? Not sure what my takeaway is supposed to be there.

Do you play strategy games at all? Galactic Civilizations is a 4x space game and one of the features it has is if you type in a sentence or two it will use AI to dynamically create a civilization.

Similar games will have their AI adapt to strategies the human is using. I love space strategy games so more examples is that in those games you'll usually have different types of weapon technologies to development and each will have their own strength and weaknesses, e.g. mass drivers strong against hull/armor but weak against shield. The AI will adapt its strategy to counter you if they feel like conflict is likely.

And in war sims the AI is better able to analyze the human player and pick appropriate targets and exploit weaknesses.

And we are seeing some RPGs experiment with AI generated dialogue.

It should not be a matter of dispute. The AI of sports titles lags behind.

[MiracleMet718]

Re: Why can't higher difficulty equal more intelligent play?

Whats the screenshot illustrating? Not sure what my takeaway is supposed to be there.

Do you play strategy games at all? Galactic Civilizations is a 4x space game and one of the features it has is if you type in a sentence or two it will use AI to dynamically create a civilization.

Similar games will have their AI adapt to strategies the human is using. I love space strategy games so more examples is that in those games you'll usually have different types of weapon technologies to development and each will have their own strength and weaknesses, e.g. mass drivers strong against hull/armor but weak against shield. The AI will adapt its strategy to counter you if they feel like conflict is likely.

And in war sims the AI is better able to analyze the human player and pick appropriate targets and exploit weaknesses.

And we are seeing some RPGs experiment with AI generated dialogue.

It should not be a matter of dispute. The AI of sports titles lags behind.

I think the point is he is making is that those type of games have expected results based on actions. So RPGs have a more simplistic “if this, then that” type of result. In sports games, you have multiple outcomes based on a variety of other factors such as player attributes, # of players, location of the ball/puck, offensive players in the area, defensive players in the area, and a bunch more that I can’t even think to list.

So the point is it’s easier to train AI when the results are linear vs. sports games where the AI needs to be very dynamic. Nobody is denying is lagging behind, but there are reasons for it due to the complexity of what is needed.

[ggsimmonds]

Re: Why can't higher difficulty equal more intelligent play?

I think the point is he is making is that those type of games have expected results based on actions. So RPGs have a more simplistic “if this, then that” type of result. In sports games, you have multiple outcomes based on a variety of other factors such as player attributes, # of players, location of the ball/puck, offensive players in the area, defensive players in the area, and a bunch more that I can’t even think to list.

So the point is it’s easier to train AI when the results are linear vs. sports games where the AI needs to be very dynamic. Nobody is denying is lagging behind, but there are reasons for it due to the complexity of what is needed.

I think you are misunderstanding what exactly I mean when I mentioned AI driven dialogue in rpgs. This is a community mod but was quickest to search for and illustrates what I'm referring to (its a long video, the first few minutes are enough to give you an idea) https://www.youtube.com/watch?v=xNPF9VKmzxw

As far as this part: "In sports games, you have multiple outcomes based on a variety of other factors"
Thats true for strategy games as well. And virtually everything else that requires AI.

Sports aren't really that complex. They have a nice set of rigid rules to follow and the outcomes are fairly limited. We've had AI that controls the rudimentary aspects of sports for decades. What the OP and myself are talking about is an AI that can strategize. Like previously mentioned, an AI that can matchup best on best should not be beyond the ability of 2k's AI but it is.

"The amount of complexity that goes into getting 10 players to run up and down the court in unison is not remotely similar to an AI moving a Pawn from A2 to A3."

This shows a lack of understanding of AI, chess, and frankly basketball too. Moving a pawn from A2 to A3 is not done in a vacuum. You move a chess piece based on your opponent's moves, the positions of other pieces on the board, and most importantly as part of a overarching strategy. Chess players make moves in advance and strategize around potential moves by the opponent in advance. To categorize it as simply "moving a pawn from A2 to A3" is disingenuous at best. For an AI chess player it shows a remarkable level of insight and strategic ability to beat a worldclass human player. There's a reason it was considered a huge milestone in AI development

(the reason I didn't initially respond to this portion, I don't want to get sidetracked into a conversation revolving around the similarities in strategic decision making between chess and basketball. But AI 20 years ago could beat a human player in chess, it is more than capable of looking at a basketball lineup and saying hmm I should match my best perimeter defender on Jordan regardless of position, and then adjust my lineup accordingly)

[ConQrDie]

Re: Why can't higher difficulty equal more intelligent play?

Other sports games are in much worse shape, and I can't help but wonder why. The reason seems clear: constant complaints about the difficulty. Companies have little incentive to invest in AI development when the casual playerbase keeps asking for a "fun" and "dumbed-down" version, even if they claim otherwise.

I’ve seen how the AI in 2K23, which was almost poetic at launch, was drastically downgraded just two months after release because of "dialing in," meaning simplifying it.

This year’s AI has faced similar setbacks. I’ve saved the AI profiles from the game’s release, December, and now. The AI doesn’t perform with the same efficiency it did at launch. While it hasn't been dumbed down to the extent of 2K23, the complaints have certainly had an impact, though I’m cautiously optimistic that it’ll survive.

In competitive games, everything comes down to skill—particularly reaction time and the ability to stay focused. This is why games feature different difficulty levels. Wanting to simplify harder difficulty levels to match your personal skill level, when many others enjoy the challenge, feels morally questionable.

[ggsimmonds]

Re: Why can't higher difficulty equal more intelligent play?

Other sports games are in much worse shape, and I can't help but wonder why. The reason seems clear: constant complaints about the difficulty. Companies have little incentive to invest in AI development when the casual playerbase keeps asking for a "fun" and "dumbed-down" version, even if they claim otherwise.

I’ve seen how the AI in 2K23, which was almost poetic at launch, was drastically downgraded just two months after release because of "dialing in," meaning simplifying it.

This year’s AI has faced similar setbacks. I’ve saved the AI profiles from the game’s release, December, and now. The AI doesn’t perform with the same efficiency it did at launch. While it hasn't been dumbed down to the extent of 2K23, the complaints have certainly had an impact, though I’m cautiously optimistic that it’ll survive.

In competitive games, everything comes down to skill—particularly reaction time and the ability to stay focused. This is why games feature different difficulty levels. Wanting to simplify harder difficulty levels to match your personal skill level, when many others enjoy the challenge, feels morally questionable.

Its testing bro.
I worked for one of the world's largest information technology firms as a developer and later project manager and few years back the company decided it wanted to be one of the leaders in AI advancement, and testing AI is a massive lift.

I remember 2k saying teams were hardcoded to defend Jordan a certain way in the 90s myEra. When you have that kind of hardcoded logic you only need to test n number of scenarios where n = the number of hardcoded logic you implemented. If you hardcode (static AI) 4 possible defensive strategies you have 4 things to test. Cool, thats not too bad right?

But what we are calling dynamic AI is open ended. It can do whatever it wants within the rules you specify. So how many scenarios do you have to test for that? The answer is a f****** lot.
Technically thats not hard, but budget wise? Testing would eat up a massive portion of the budget, and for relatively little payoff.

Thats an oversimplification but it gives you the idea.

Funny somewhat related story, anyone remember the football game "Backbreaker" from several years ago? Back in the days when everyone complained about Madden's canned animations Backbreaker was an attempt to make a full physics based football game. Early on in development they made a slight boo-boo. They forgot to add in the facemask penalty into their AI so during early development and testing of the game when they unleashed the AI all the defenders did was grab the ballcarriers facemask every single time. Every tackle was grabbing the facemask.

[Real2KInsider]

Re: Why can't higher difficulty equal more intelligent play?

"The amount of complexity that goes into getting 10 players to run up and down the court in unison is not remotely similar to an AI moving a Pawn from A2 to A3."

This shows a lack of understanding of AI, chess, and frankly basketball too.

When I say running up and down the court, I literally mean that a script is running at all times and TEN players are making decisions based on a multitude of factors in REAL-TIME.

This is not the same as a turn-based, stationary game of Chess. Basketball is an ACTION game.

Nevermind that Chess pieces have set values and every NBA player is individualized. I'd imagine most Chess AI's would go right in the gutter if both sides started with two Queens. Or one side with One Queen and the other with 4 Rooks.

And to be crystal clear, 2K's AI has improved a ton in the last 25 years, but again, asking it to understand basketball better than 99.9% of the people on planet earth is a tall task.

it is more than capable of looking at a basketball lineup and saying hmm I should match my best perimeter defender on Jordan regardless of position, and then adjust my lineup accordingly)

But that's literally what the game is doing now.

Is it capable of picking out a starting lineup or making substitutions without positional distinctions? No.
(For reference see the old argument of "2K should be positionless")

What about teams with multiple defensive stoppers? What about teams with none?

What about lineups where you're replacing Jordan with a Zach LaVine level equivalent?

What about Dream Team lineups with Jordan + Bird + Magic etc

None of this even takes into account the User skill level where they could be Greening everything, or Bricking everything. How is AI supposed to factor slider success rates into it's decision making? If the AI "knows" that everything is jacked to 100 and is adapting to that, find and protect John Connor at all costs.

So how many scenarios do you have to test for that? The answer is a f****** lot.
Technically thats not hard, but budget wise? Testing would eat up a massive portion of the budget, and for relatively little payoff.

Okay, well, we all know what the development cycle is for sports games. It certainly doesn't make it more conducive to large advancements.

[ggsimmonds]

Re: Why can't higher difficulty equal more intelligent play?

When I say running up and down the court, I literally mean that a script is running at all times and TEN players are making decisions based on a multitude of factors in REAL-TIME.

This is not the same as a turn-based, stationary game of Chess. Basketball is an ACTION game.

So yes, I was correct. Its a misunderstanding of all 3.
Processing moves for ten players is not a lot. You drastically underestimate modern technology's processing power and what AI is capable of if you think 10 players is a significant number. Sure, Chess is a turn-based stationary game. You're just throwing words out that you think help your argument. Google the Shannon number and do some research on the significant of AI in chess. I guarantee you with each move in that turn based stationary game it does far more complex calculations than 10 virtual basketball players.

Nevermind that Chess pieces have set values and every NBA player is individualized. I'd imagine most Chess AI's would go right in the gutter if both sides started with two Queens. Or one side with One Queen and the other with 4 Rooks.

Rashidi bro, we are talking about a video game, not real life. In video games, every virtual player will have set values. Your point would be valid if I was arguing that we should replace real life coaches with AI coaches, I'm not. I'm not asking or hoping for a videogame AI that could outcoach Eric Spoelstra. I'm asking for a videogame AI that could engage me while I sit in my room wearing a 3 day old shirt and stuffing my face with cheetos during play stoppages.

And to be crystal clear, 2K's AI has improved a ton in the last 25 years, but again, asking it to understand basketball better than 99.9% of the people on planet earth is a tall task.

1. Its not a tall task
2. Sure 2k's AI has improved a ton, but nevertheless it lags significantly behind advancements in AI both in general and other genres of videogame titles

But that's literally what the game is doing now.

Is it capable of picking out a starting lineup or making substitutions without positional distinctions? No.
(For reference see the old argument of "2K should be positionless")

Its literally not what the game is doing now, you said so yourself in the very next line. If I run a game with today's Thunder vs Kawhi's Spurs and I start killing the AI with Shai, the AI will not put Kawhi on Shai. He will continue guarding whoever is the 3 for OKC. This isn't positionless basketball. Thats a separate matter entirely

What about teams with multiple defensive stoppers? What about teams with none?

What about lineups where you're replacing Jordan with a Zach LaVine level equivalent?

What about Dream Team lineups with Jordan + Bird + Magic etc

None of this even takes into account the User skill level where they could be Greening everything, or Bricking everything. How is AI supposed to factor slider success rates into it's decision making? If the AI "knows" that everything is jacked to 100 and is adapting to that, find and protect John Connor at all costs.

All I'm going to say in response to this portion, read up on the abilities of AI. You say all that like its a tall task for an AI. I assure you its not.

The challenge for an AI would be when the user goes rogue and starts behaving unpredictably. AI pulls from historical data so if a user starts operating outside of that historical data that is likely to trip up modern AI.

Okay, well, we all know what the development cycle is for sports games. It certainly doesn't make it more conducive to large advancements.

This part is true. Like all business videogame studios are obsessed with year over year sales. They want that X% increase year over year above all else. They are reluctant to make investments that would jeopardize those sales goals. That's why we are seeing 2k throw in half baked new eras each iteration.

Now the truth of the matter is anyone who cites "development cycle" is a) being disingenuous or b) they are got by the industry. Because the development cycle is a creature of their own creation. Citing the development cycle is nothing more than a (usually successful) attempt to justify the studio's yearly sales goals approach

[Real2KInsider]

Re: Why can't higher difficulty equal more intelligent play?

So yes, I was correct. Its a misunderstanding of all 3.
Processing moves for ten players is not a lot. You drastically underestimate modern technology's processing power and what AI is capable of if you think 10 players is a significant number.

I am aware that 10 players is nothing in terms of PROCESSING power, given the Kasparsky Chess computer was evaluating 14,000 moves per second or whatever it was in the 80s.

Sure, Chess is a turn-based stationary game. You're just throwing words out that you think help your argument. Google the Shannon number and do some research on the significant of AI in chess. I guarantee you with each move in that turn based stationary game it does far more complex calculations than 10 virtual basketball players.

What I'm saying is all 10 players are moving up and down the court in unison. They are puppets on a string. One player moves, the rest follow suit. That is how the game is programmed.

The rules & goals in Chess are very clearly defined. An NBA emulation? Not so much. A single basketball possession is exponentially more complex.

Your point would be valid if I was arguing that we should replace real life coaches with AI coaches, I'm not. I'm not asking or hoping for a videogame AI that could outcoach Eric Spoelstra. I'm asking for a videogame AI that could engage me while I sit in my room wearing a 3 day old shirt and stuffing my face with cheetos during play stoppages.

But we do get that. The game does make adjustments based on the player's strategy.

advancements in AI both in general and other genres of videogame titles

At this point you're going to need to make relevant examples. What action-based games have these evolutions?

Its literally not what the game is doing now, you said so yourself in the very next line. If I run a game with today's Thunder vs Kawhi's Spurs and I start killing the AI with Shai, the AI will not put Kawhi on Shai. He will continue guarding whoever is the 3 for OKC. This isn't positionless basketball.

There are definitely instances where a wing defender like a Pippen will take the POA, that was something they worked in a few years ago. It can be a value proposition - does moving Tony Parker create a larger mismatch elsewhere on the court? Does Kawhi actually have the requisite Speed etc rating to actually keep up with Shai? Does hiding Tony Parker somewhere else create a larger mismatch based on his defensive ratings, size etc?

There are also plenty of instances IRL where the coach follows the gameplan and doesn't make an adjustment. Doc Rivers, Mike Budenholzer, Tom Thibodeau, etc are not exactly known for making in-game adjustments. Teams have different values for the frequency of which they make adjustments. Six straight points might be "killing it" to some, but a lot of that is contextual to the situation & game clock, it isn't cut and dry.

It also bears noting that this is precisely what happens when the AI calls Timeout. That is when they are changing up the strategy. NBA players IRL aren't making scheme/matchup adjustments on a possession-by-possession basis either.

The challenge for an AI would be when the user goes rogue and starts behaving unpredictably. AI pulls from historical data so if a user starts operating outside of that historical data that is likely to trip up modern AI.

The User is going rogue like 95% of the time. The game plays much differently when it's AI vs AI. We don't move the same.

Because the development cycle is a creature of their own creation.

The sports leagues are the ones that want yearly releases, not the Studios. This dates back to the 90s, well before the cash cows these games became today.

DaCzar (who was the offensive AI guy who moved the game well beyond where it was 20 years ago), has been pretty open about devs not preferring a yearly cycle, but it's not the business reality.

[ggsimmonds]

Re: Why can't higher difficulty equal more intelligent play?

I am aware that 10 players is nothing in terms of PROCESSING power, given the Kasparsky Chess computer was evaluating 14,000 moves per second or whatever it was in the 80s.



What I'm saying is all 10 players are moving up and down the court in unison. They are puppets on a string. One player moves, the rest follow suit. That is how the game is programmed.

I started playing basketball games with NBA Live 98 and that had 10 players moving up and down the court in unison. You are saying that it can't be done because its too complicated, when its been done for 30 years.

The rules & goals in Chess are very clearly defined. An NBA emulation? Not so much. A single basketball possession is exponentially more complex.

This is just wrong. An NBA emulation will have rules and goals that are clearly defined, every thing in programming has to be clearly defined

But we do get that. The game does make adjustments based on the player's strategy.

And I want to see more lifelike adjustments. The game doesn't make holistic adjustments. The AI may decide this is a mismatch in the post and send a double team. But it is incapable of a strategy like front the post and pressure the ballhandler

At this point you're going to need to make relevant examples. What action-based games have these evolutions?

No, I don't have to oblige your arbitrary moving of the goalpost and irrelevant distinctions. I wasn't born yesterday, I know how this works. Anything I bring up you will cook up an argument on how its not comparable and you will then move the goalpost again. Knock it off, "action-based" games don't matter, you acknowledged as much further down.

There are definitely instances where a wing defender like a Pippen will take the POA, that was something they worked in a few years ago. It can be a value proposition - does moving Tony Parker create a larger mismatch elsewhere on the court? Does Kawhi actually have the requisite Speed etc rating to actually keep up with Shai? Does hiding Tony Parker somewhere else create a larger mismatch based on his defensive ratings, size etc?

There are also plenty of instances IRL where the coach follows the gameplan and doesn't make an adjustment. Doc Rivers, Mike Budenholzer, Tom Thibodeau, etc are not exactly known for making in-game adjustments. Teams have different values for the frequency of which they make adjustments. Six straight points might be "killing it" to some, but a lot of that is contextual to the situation & game clock, it isn't cut and dry.

Those instances are hardcoded.
As for instances when real life coaches don't make an adjustment, okay so? We have a coach "call plays" slider, we can also have a "gameplan fluidity" slider.

It also bears noting that this is precisely what happens when the AI calls Timeout. That is when they are changing up the strategy. NBA players IRL aren't making scheme/matchup adjustments on a possession-by-possession basis either.

Okay so your "action based game" distinction doesn't matter then. Its basically a turn based scenario.

The User is going rogue like 95% of the time. The game plays much differently when it's AI vs AI. We don't move the same.

No they don't, and stop obsessing over movement. I don't care about movement. Going rogue would be something like playing Steph Curry at the 5. You ever see those stupid "I made a team full of kickers in Madden" videos? Thats going rogue.

The studio executives are the ones that want yearly releases, not the Studios. This dates back to the 90s, well before the cash cows these games became today.

DaCzar (who was the offensive AI guy who moved the game well beyond where it was 20 years ago), has been pretty open about devs not preferring a yearly cycle, but it's not the business reality.

FTFY

[ggsimmonds]

Re: Why can't higher difficulty equal more intelligent play?

I'm going to bow out of this conversation, but I'll leave it with an example of something I want to see out of a future 2k game and this is something that is within the capability of modern technology. Its also an opportunity to plug one of my favorite youtube channels and I'll never pass that up :P

https://www.youtube.com/watch?v=rsf84_ZEFwk

The Lakers crafted a holistic gameplan to deal with Jokic. The strategy had different parts to it, it wasn't just a "double team Jokic in the post on the catch."

~75% of this gameplan can be implemented in 2k today using the current defensive settings, the issue is only that the 2k AI is not capable of using the different settings to form a holistic strategy. Its limited to single strike adjustments, i.e. single setting options in a vacuum.

All the talk about movement and 10 guys in unison is background noise, it doesn't matter. All we need is an AI that can use the tools that already exist* in an intelligent manner.

There are 2 limitations or obstacles to this:
1. Modern AI is able to do this via machine learning technologies. What this requires is access to historical data. Long story short -- if we want this level of improved AI in our offline single player franchises, we need a constant internet connection so that the AI has access to the data it needs.
2. The level of machine learning that is reasonable to hope for in a videogame is incapable of being creative. In everyday terms, it can't create revolutionary never before seen strategies; it would only search historical data like a menu

* We would need the addition of more defensive options, but they wouldn't differ in kind from the settings we already have, i.e. they wouldn't be significantly more complex than what we already have. For example we have a deny ball option under off ball defense, well we would also need something like a "deny space" option. Also something similar to the pre-rotate setting but for off ball.