EA's AI Game Development Tools Are Apparently So Bad That It's Costing More Money To Fix Their Mistakes

Draugoth

Draugoth

Gold Member

I Might Not Be Saving EA Money After All

According to a report from Business Insider (thanks, Tweak Town), EA's rollout of AI in game development through its own AI chatbot has not gone as intended. It's claimed that the generative AI tool is prone to "hallucinations", requiring actual human developers to go in and fix its mistakes, costing time and money.

A soldier holding up an assault rifle and using his radio in Battlefield 6.

These problems reportedly stem from EA's generative AI tool, ReefGPT. It's said that the AI keeps adding incorrect code to projects, which have to be fixed by the developers who don't want to work with the tool in the first place.


It's further claimed that EA's developers fear that they are effectively training their replacement as they help the AI tool do their jobs.

Article:
Via TheGamer
 
AI is nothing without people behind it to fix or check if they aren't doing anything stupid. Its just like having numbers in a company and not having a speciallist analyst to interpret those numbers.
 
Last edited:
I don't know. I have never written code before and was able to make an executable that could dissect a PSD file, with multiple layers, into individual PNG files. A unique need for my work, but I was able to write working code in an hour with the help of chatgpt. Something that would take me a whole day to do now takes minutes with this script. It's kind of incredible. You guys will never convince me that AI isn't amazing. And in time, I'm sure EA will get this to work with little problems.
 
Last edited:
Not a surprise. That's how any new program rolled out to a company works.

Get a company to implement a new SAP or Oracle program. Will involve lots of fixing in the first year as the data loads and reporting will have errors. But at some point after everyone cries about it, it works great. And then ongoing maintenance going forward like any program needs.
 
Article is kind of dumb - it would still be saving them money if the time spent fixing the code is much shorter than writing the whole thing manually. No-one, hopefully, is expecting AI to just flawlessly write a game
 
Said it in the other thread. AI is just the newest C-Suite Trend that, while revolutionary/game changing in many cases and industries, is something execs (in many businesses) want now to cash in on the fad for investors/their boards. They aren't thinking about the implementation thoroughly enough or how and when to use it. I am seeing this play out at my job and it's clear that many just want it without working out the hows. It feels like it is being rushed.

And there is a throng of grifters willing to consult with them at high cost on how to make ai work for them in development. Selling them on bullshit promises that it is incapable of doing.

AI is a tool (a damn good one) and power multiplyer for employees and should be seen as one, not a wholesale replacement to established systems. I have sat through some meetings where it was being sold as that.

Many people are already suggesting the AI bubble will burst sooner than expected and they will move on to the next corporate/investor trend. AI will still be around, and hopefully be seen with more rational expectations and as an additional tool for professionals.
 
Last edited:
poppabk said:
Article is kind of dumb - it would still be saving them money if the time spent fixing the code is much shorter than writing the whole thing manually. No-one, hopefully, is expecting AI to just flawlessly write a game
Click to expand...
Exactly.

It's a stupid argument because a lot of the naysayers are trying to make it sound like AI is an all or nothing tool. Well, if it was so all or nothing then the company wouldnt need any employees. It's obvious it's to reduce costs, redundancies, and time. It's up to the users to use it and keep their job making themselves better.

It'd be like working at a bank and saying money counting machines are a waste of time since occasionally it'll count wrong. So let's go back to everyone counting stacks of money by hand one bill at a time since the machines arent perfect.

They might not be perfect, but I bet the machines are still faster and more accurate than hand counting. You still need a human to load the machine and process the amount. And if something looks off, the human takes over.
 
Last edited:
Plague Doctor said:
Many people are already suggesting the AI bubble will burst sooner than expected and they will move on to the next corporate/investor trend. AI will still be around, and hopefully be seen with more rational expectations and as an additional tool for professionals.
Click to expand...

It's impossible to know when it will burst. But if the circular deals tell anything, Big Tech is doing everything they can to stop it from bursting.

But when token usage prices drastically increases (which is inevitable) it will be very damaging to the technology.
 
Last edited:
With AI, today technology is the worse it will ever be. I still remember the cursed 1st AI generated videos. Now, not long after, AI is already generating videos that can fool anyone not trying to superanalyze the video. In a few years i bet half the studios are using AI to help develop they games. AI is a tool, and should be treated as such, you still need humans with artistic vision and creativity to make something worth it.
 
Last edited:
Aenima said:
With AI, today technology is the worse it will ever be. I still remember the cursed 1st AI generated videos. Now, not long after, AI is already generating videos that can fool anyone not trying to superanalyze the video. In a few years i bet half the studios are using AI to help develop they games. AI is a tool, and should be treated as such, you still need humans with artistic vision and creativity to make something worth it.
Click to expand...
They are probably already using AI, but dont want to admit it for sake of optics.

Many gaffers have said their companies have already been using AI in their tech related jobs for many years. It's just that all the normies think AI has only been a thing the last year or two goofing around on websites making fake images and reading business articles. As if AI programs were just invented in 2023 or 2024.
 
Last edited:
Aenima said:
With AI, today technology is the worse it will ever be. I still remember the cursed 1st AI generated videos. Now, not long after, AI is already generating videos that can fool anyone not trying to superanalyze the video. In a few years i bet half the studios are using AI to help develop they games. AI is a tool, and should be treated as such, you still need humans with artistic vision and creativity to make something worth it.
Click to expand...

Generative AI use architectures just like we do for CPUs and GPUs. When you try to go past what the architectural allows, you start to see lots of problems and no gains in performance like we did with Raptor Lake. I wouldn't be surprised if we are already at the diminishing returns territory with this specific tech. The biggest problem as well is that Generative AI is heavily reliant on data, which they basically ran out of already.

StreetsofBeige said:
They are probably already using AI, but dont want to admit it for sake of optics.

Many gaffers have said their companies have already been using AI in their tech related jobs for many years. It's just that all the normies think AI has only been a thing the last year or two goofing around on websites making fake images and reading business articles.
Click to expand...

I've seen people train AI with the Pascal GPUs back in 2016.
 
Last edited:
Here's the thing with AI. It's good with:

Things that are small prototypes
Things that don't have to be accurate 100%, like close enough will work
Things that don't have to be repeatable

It's bad with:

making maintainable code that's error free
working with existing code
trying to making things that require 100% accuracy

So it might be good for art, sound, etc that doesn't have to be 100% accurate, but is going to be more trouble than it's worth on a LARGE codebase that has to be maintained for many years. You use AI to generate sub standard code to start, then you'll be fighting it forever, it'll never work out. It's good for rapid prototypes or small projects where you never have to maintain it.

The problem is that the current AI models never tell you "I don't know" or "I have a 75% confidence factor in this answer". It just says the answer as if it was always right and a large part of the time it isn't. But people don't pick up on it because a lot of the times the answers seem reasonable or close. Researchers have had issues where they have AI analyze their dataset and spits out a reasonable answer, then reload the same dataset and get a different answer. It's not consistent which makes it very difficult to trust or use once you figure that out.

To many execs have just heard the buzzwords and have jumped on the bandwagon so they can get raises and bonuses for implementing AI. Very few people at big corporations are making big advances in productivity in programming using AI, they've done studies that have shown that....for what parts it frees up time it adds it in other areas, leading to a net zero gain. That's why all these corporations aren't suddenly pumping out games 30% to 40% faster like they claimed AI was making them, if it was you'd see release dates getting moved up all the time.
 
My friend is an experienced developer at a large mobile game publisher. He says the what the ai does well is the job of inexperience coders. As in, he has to review it all either way. His job didn't fundamentally change. The ai pumps it out a lot faster and cheaper though. So he's still coding but he's having the ai go whip up some kind of function for him and he has to review that, and decide to add it or not.
 
Crayon said:
My friend is an experienced developer at a large mobile game publisher. He says the what the ai does well is the job of inexperience coders. As in, he has to review it all either way. His job didn't fundamentally change. The ai pumps it out a lot faster and cheaper though. So he's still coding but he's having the ai go whip up some kind of function for him and he has to review that, and decide to add it or not.
Click to expand...

Yup, as soon I realized you still had to review what GPT wrote down the whole 'We will replace everyone with AI by 2030" fell apart.

It's obvious cheap now but it wont't be the case in the future, that's more than enough reason to ditch it for faster and less powerful local models.
 
You must log in or register to reply here.

Similar threads

Report: EA's internal AI is causing issues with games development
Opinion Business Game Dev  2
50
125
mckmas8808 replied
Hideo Kojima Wants AI To Handle “Tedious Tasks” To Cut Down Development Time & Cost
Opinion Business Game Dev 
26
171
yogaflame replied
Project Shadowglass - AI meme inspired game in development.
News 
9
236
Kikorin replied
[GamesIndustry.biz] 87% of game developers are using AI agents in their workflows, says new survey
Business Game Dev 
47
984
Loomy replied
EA and Stability AI partner to empower artists, designers, and developers
6
69
pqueue replied
Top Bottom