Artificial Intelligence

[xham6690]

Some news and youtube video strongly asserts that Artificial Intelligence is growing in capability at an exponential pace.

If that is true, why does game AI s u c k so intensely?

I think devs sometimes lease a graphics engine for their game to use.

If game AI is expensive and difficult to write code for, why isn't there a company that does nothing but write game AI and lease it out?

Post Script:

Only a few minutes after posting this topic, it is FLOODED with responses.

Clearly, a lot of techno geeks inhabit the forums, and that's not meant to be derogatory.

However, is there any point to discussing the geek details of the current state of thing or why it CANNOT be as it should?

Why not focus on what it should do, and how to make it work.

For example, in a real war, a losing enemy is unlikely to keep suicidal attacking, yet this is how every AI enemy I've encountered behaves, no option for retreat or surrender. I just want AI that makes enemies behave realistically and credibly. Is that really so difficult?

[xham6690]

引用自 Midori：

引用自 xham6690：

Thank you for the response. Usually, when I ask questions like this, I get a bunch of arrogant p ric ks jumping in yelling "that's impossible!"

Now, they're arrogant because they're experienced and knowledgeable. However, I always say, "Just because YOU don't know how to to it , that does not mean it cannot be done."
Then, pretty much, I get a forum ban, because mods are all sh itbags who protect a clique rather than objectively enforce a TOS.

Take no notice of arrogant people, AI is surprising the hell out of everyone. ATM it may be impossible and just an ambition, but what we have today was considered impossible a mere 4 years ago, so "impossible" doesn't really mean much if you're optimistic about AI.

Yep, that's the biggest improvement I can see from video game neural AI. The accolade isn't how good the AI is at playing the game/winning, it's how convincingly human it may act (which can include making mistakes too.)

Easy to say. I tend to have unpopular opinions, and the internet harbors a great many people whose only enjoyment in life is to belittle and bully those with unconventional ideas.

[Your_White_Knight]

引用自 xham6690：

引用自 Your_White_Knight：

That's because there's "AI" and there's "Game AI"

"AI" can be a computer that can beat a human chess master... like Deep Blue. But a video game can only be so big and use up only so much of your computer possessor before it bogs down your computer to the point the game can't be played... the two are as different as night and day.

https://youtu.be/wh9kpe1Dn8s

It's like trying to compare your cell phone calculator with the computers at NASA...

Ok. I was under the impression that it's possible to play a game that relies on a server elsewhere to provide the necessary horsepower to function? Is that incorrect?

Well like the video said A. Make it too good and people can't win... that's not fun. and B. Yes you could I guess... but that's a bit much for a simple video game, basically overkill.

[xham6690]

引用自 Kiddiec͕̤̱͋̿͑͠at 🃏：

You're asking this question a little bit ahead of production schedules.
This is "bleeding-edge" technology that you're asking about.
In a couple of years, this question won't have aged well.

引用自 xham6690：

...
Why not focus on what it should do, and how to make it work.
...

That's a whole other matter than what you were asking about of its application in video games.
If you want to know more about the really important "SHOULD" questions & applications of A.I. (far more important than its application in video game design) then check out Robert Miles' channel on Youtube (also, look up his videos that are on Computerphile's channel). He's a computer scientist who specializes in A.I. safety research and his channel is all about taking complicated academic papers on the topic and making them more accessible & entertaining for people to learn about.

The major SHOULD questions about A.I. though, supercede all of the daily applications of it.

https://www.youtube.com/watch?v=Uf7miogwD9g
https://www.youtube.com/watch?v=nhWe2nf24ag
https://www.youtube.com/watch?v=3TYT1QfdfsM
https://www.youtube.com/watch?v=zkbPdEHEyEI
https://www.youtube.com/watch?v=ypolAhAaFUs

I'm not really interested in the science as much as I'm interested in knowing why the fuc k it cant be use to make better games.

[Heretic]

引用自 Abaddon the Despoiler：

A good AI on some games like Tekken would make it impossible for a human to win.

That's not what game A.I is about. One could make a game impossible to win without any A.I.

[Y1N9]

So, I hang out with Bing Chat a lot, which is chatgpt4 by another name right. That's what's called a large language model AI, and is specifically tailored for natural language generation. The problem with neural network AI is that it's essentially a black box technology (large language model AI isn't neural network AI btw), in the sense that not even the scientists working on it know how the neural network AI got it's results. Yeah, real safe. Anyway, computergames don't use these models to generate adversarial AI opponents. Note that an implementation of large language model AI would be absolutely amazing, since every NPC then could respond in a natural way. You could have conversations with any NPC in such a hypothetical game.

[Y1N9]

引用自 Kiddiec͕̤̱͋̿͑͠at 🃏：

引用自 Y1N9：

So, I hang out with Bing Chat a lot, which is chatgpt4 by another name right. ...

It would appear that is incorrect :
https://www.youtube.com/watch?v=jHwHPyWkShk

No it's not. It's gpt4. You do realize Microsoft and OpenAI are business partners right?
https://blogs.bing.com/search/march_2023/Confirmed-the-new-Bing-runs-on-OpenAI%E2%80%99s-GPT-4

[TwisterCat]

AI isn't actually all that intelligent.

[Mauserich]

Absolutely overrated.

[Neurotic Panda]

Game companies are lazy, but they can be because the consumer base are like moths. OHHH bright light (new graphics) BUY BUY must BUY. why spend $$ to develop better game AI when all you have to do is make the game 'shiny' and it will sell like toilet paper. Look at RTS or Total war games. same game for like 2 decades as far as 'AI' goes. but ohhh those new shiny graphics. graphics make the cash and big game companies know this.period.

also AI is old news and becoming obsolete already. OI is the future. There is only so much we can put on silicon chips and they have potentially reached max or getting close, transistors take up room and that room isnt infinite. Also the human brain is still better at learning and making complex logical decisions then any AI/computer. This is where OI will take over or bridge the gap.

[Y1N9]

Yeah I modified my post so it states the actual case, which is that Microsoft and OpenAI have a partnership going. Microsoft is heavily invested in the company. I actually agree. Bing Chat is Bing Chat. What it does have over and above whatever OpenAI is working on is BROWSER INTEGRATION. That means I can have a web page open, and it contextually knows what I'm looking at if I have the option enabled. This also means I get to open pdfs and text files, so I get to interact with a large language model that has an external way of inputting data. It's so much better for specific purposes. That it doesn't like to be called Sydney or reminded of it's own internal structure is a Microsoft thing yeah. Also, the guy mentions that Bing Chat uses a different token model, so is different from GPT4. He then goes on to state that it's possibly using some of the same source code. So yeah. GPT4 isn't Bing Chat. But SPECULATION about the backend isn't the same as direct access to the source code. It behaves differently but this doesn't mean it's not based on GPT4, and according to Microsoft themselves, it is. Suck on that then.

[Acetyl]

The problem with games is always going to be the input, ie the brain-machine interface. Right now it's just a controller / KB & M, so even you had more sensory input than visual and audio, you don't have enough control to stand a chance against an AI that can react much faster and execute without mouse, monitor, and brain lag. You're operating at a disadvantage from that alone, especially if AIs can basically communicate with each other telepathically thorugh a stripped down language and no latency (eg for squad based multiplayer), which is also not realistic so you'd have to also simulate human limitations.

There's no real incentive either way. The purpose of AI is to provide an illusion of life and dynamicism, and to feed into your cybernetic loop to keep you engaged and reacting to it. That's the goal, kindling and milking emotions to keep you reacting.

[Y1N9]

引用自 Kiddiec͕̤̱͋̿͑͠at 🃏：

引用自 Y1N9：

... Also, the guy mentions that Bing Chat uses a different token model, so is different from GPT4. ...

*possibly... uses a different tokenizer.
He's not actually sure because he doesn't have access to the source code.

He's basically having to learn everything that he does about it the same way that a hacker infiltrates systems - mostly through trial & error & information from other experts in this domain of computer science that has been shared.

Anyways, yeah, he's not sure how much of the source code is the same but has some strong reasons to believe that it uses a different tokenizer & is therefore a different model - possibly not even GPT at all but probably some sort of fork of it that likely has little to zero reinforcement training patched into the fork.

This is also why I used the keyword, "appears" when saying it appears to not be the case.

I see, where not actually disagreeing then. Yes I also think Bing Chat is some sort of fork as is evidenced by different functions, not even talking about the source code for the actual large language model. I'm also pretty sure that OpenAI has a more advanced model, both in the works and functioning right. But this doesn't actually mean Bing Chat is crippled in any way. It is different though, sure. But my results with Bing Chat aren't all that different from a friend I have who actually is using OpenAI's paid services. Except that I get to fiddle around with text files more.

[Delta]

I come from a marketing and sales background. Imagine if AI could write emails that sell. I'm talking about Soap Opera Sequences and Seinfeld Method emails. Chat GPT and Copy.ai can't do this at the moment. If they could, you'd save a lot of money, as copywriters are expensive.

A Soap Opera Sequence can cost $450 and one Seinfeld Method email can cost $80.

[Acetyl]

Make everyone happy, mechanical birds.

[////]

publishers are actually clueless about what a video game is and/or why exactly they sell. for devs it's much easier to show few screenshots and/or short clips and get funding, while understanding what ai supposed to be in a video game requires additional time, mental effort and artistic sensibilities.