GDC 25: Rez Graham on AI, Game Design, and the Death of Art

Courting again to 1988, the Recreation Builders Convention is an annual occasion held on the Moscone Heart in downtown San Francisco. In the course of the conference, many talks are held on all types of various matters, together with postmortems on accomplished tasks, retrospectives, and evaluation of the business at giant. Yesterday, designer David “Rez” Graham, who has labored on many video games through the years similar to The Sims 4, and who teaches recreation design on the Academy of Artwork College, delivered a chat dubbed Recreation AI Summit: The Human Value of Generative AI. If the response, a swell of enthusiastic applause and cheers, was something to go off of, Graham’s phrases struck a chord with the viewers, lots of whom (together with this author) headed outdoors of the presentation chamber to pepper him with questions afterwards.

Earlier than we launch into what Graham mentioned, let’s take a second to (broadly) talk about what generative AI (synthetic intelligence) is, for many who may not know. Generative AI is a brand new type of synthetic intelligence expertise that produces new artwork, writing, music, and extra with minimal enter from the consumer. Merely enter in a few prompts to specify the specified content material or end result, and the given program will make it. To perform this, generative AI software program is skilled by way of so-called scrapes of as vast a pool of knowledge for reference as doable. What is that this knowledge? That’s the place issues begin to change into murky and sticky. The info is… folks. Particularly, the inventive spark of human beings within the type of their drawings, their writings, their songs, their movies—no matter it’s that the generative AI may be tasked with reproducing.

Rez Graham at GDC 2025. Picture by way of Recreation Developer.

Within the case of many generative AI packages, these scrapes are performed with out the permission (or most often even the information) of the creators. Scrapes will cowl swaths of content material from internationally vast internet, in some circumstances 1000’s and even thousands and thousands of samples of a given topic, after which take that knowledge to derive new content material from. However therein lies the rub: generative AI is, at its core, extremely by-product. It’s not making something really new, however as a substitute learns from the works of others and makes use of it to supply a Frankenstein’s monster of hodgepodge property from throughout the Web. Generative AI isn’t succesful at this level of going past what it’s taught by way of knowledge harvesting. Which isn’t to say generative AI at all times produces junk (the AI pictures on this article are all very serviceable and, sadly, you may not even have recognized they’re AI if I didn’t let you know), however the outcomes aren’t what’s in query—it’s the ethics.

Every thing I’ve described to date would possibly sound comparatively benign—”so generative AI is only a software for creativity!” a few of you’re doubtlessly considering—however the actuality is extra sinister than a primary look lets on. Generative AI results in someplace darkish and oppressive. It’s this bleak future that Rez Graham sought to focus on along with his discuss at GDC. In it, he described how these generative AI makers are circumventing creators, and in some circumstances willfully so, to pad their libraries of content material by way of knowledge scrapes with none intention of crediting or paying these whose work is being utilized. Somebody can spend years perfecting an artwork fashion, for example, and a generative AI maker can swoop in, scrape all of that work, after which use it to supply content material extremely related, if not similar, to what the unique creator makes.

Samus Aran generated by way of Google Gemini AI.

That actually doesn’t sound very moral, but it surely’s change into an on a regular basis incidence in just about each inventive area on the market, together with video video games. What actually rankles, nonetheless, isn’t simply the usage of creators’ work with out their permission, but in addition the, because the discuss so astutely describes it, human value of this expertise. One value may be very easy: generative AI is a method for grasping corporations to supply content material (word, not artwork, which we’ll get to) with fewer staff. Within the dream world of those corporations, AI may finally get to the purpose the place it handles each single facet of the inventive course of, from manufacturing, to advertising and marketing, to promoting, and every thing in between. In an business hemorrhaging staff each month owing to the fixed layoffs happening, that’s a shady proposition for people who find themselves already struggling to seek out constant, steady work.

The opposite human value is maybe a bit bombastic in the way it’s worded, however actually, is there every other option to put it than to say that unrestricted generative AI will completely result in the loss of life of artwork? Graham used the quote from Jeff Goldblum’s character Dr. Ian Malcom within the movie Jurassic Park the place he states “Your scientists had been so preoccupied with whether or not they may, they didn’t cease to suppose if they need to.” As Graham so eloquently put it yesterday, what design downside, precisely, is being solved by generative AI? Human beings ideate, they riff off of the works of others, they innovate, and so forth, all as a part of the inventive course of. Generative AI capabilities seemingly as a rebuke on these inventive endeavors by suggesting that there’s no level in all of that foolish human creativity when an AI can do it in seconds.

 

My distaste for the usage of generative AI would possibly really outstrip Graham’s, as I don’t even actually agree with it as a method of lowering so-called monotony for duties, like, say, creating a number of cells of animation for a cartoon. I give it some thought like this: if somebody had been to animate a leaf falling from a tree, taking hours upon hours to meticulously recreate the movement of a leaf slipping free from a department and slowly swirling all the way down to the earth, that imagery solely has which means and price in understanding {that a} human being needed to put within the labor to breed that sight on display. What’s particular or fascinating about an AI getting a immediate from somebody to animate a leaf falling and churning out the specified end in two seconds? By this logic we should always cease consuming scrumptious meals as a result of a diet shake can clear up all of our sustenance wants with a few fast gulps from a cardboard carton.

Graham said in the course of the discuss that corporations producing generative AIs could be much less harmful in the event that they at the least took into consideration the wants and wishes of the creators whose works they so gleefully and greedily gobble up for scraping. That certainly, if creators within the micro had been in a position to seize extra management over how these AIs are skilled and utilized them in a fashion that fits their workflow and wishes, that the expertise could be extra useful to the business. It’s actually some sound logic and I don’t completely disagree with it, however for me, the answer needs to be (however sadly gained’t) to purge generative AI from inventive endeavors as a complete. That customers demand content material made by different human beings and reject these mechanically cobbled collectively abominations. Sadly, with the business so hellbent on pinching pennies and churning out content material, generative AI is extra prone to change into more and more current, not much less so.

 

Nonetheless, as somebody who thinks of himself as an artist, I discover generative AI to be distasteful, particularly within the predatory method that it’s being skilled and utilized. As I discussed earlier, corporations are involved with making content material, not artwork, nowadays. With the streaming mannequin, companies have created an insatiable beast whose huge enchantment to their viewers is at all times, “look how a lot stuff there may be to observe/take heed to/play!” However with ease of entry comes a brand new, persistent, fixed demand for brand spanking new, new, new from customers. Binge dozens upon dozens of episodes of a TV present in a matter of days versus years, and abruptly the streaming supplier realizes that with out that new, new, new to repeatedly provide, customers will merely abandon them and transfer on to whomever is providing them one thing recent.

The tip results of the so-called demand technology is an limitless want for extra content material that’s driving the uptick in output of digital slop from everybody making video games, films, TV exhibits, and all types of leisure in between the previous few years. So whereas some would possibly suppose “the loss of life of artwork” is an over-the-top assumption about the place generative AI leads humanity, I discover it to be completely apt, if not a bit on the conservative aspect. It’s a blight and it has no enterprise in gaming or anyplace else the place an individual may and needs to be paid for his or her distinctive imaginative and prescient, talent, and exhausting work. Graham’s discuss was refreshing, however I worry that it’s a warning that gained’t be heeded.