Lately, I’ve been getting this question a lot: As a screenwriter, am I worried about the increasing use of artificial intelligence in the entertainment business? Will AI put writers out of work?
I have two answers to this question. The first one is a little flippant. “No,” I say. “I’m not worried. Intelligence in Hollywood has always been artificial, and we’ve managed to make nice livings anyway.”
SENATE AIMS TO NAVIGATE CONFLICT BETWEEN COPYRIGHT AND TRAINING AI
The second answer is a little more nuanced. The key to any successful artificial intelligence contraption is the dataset. A computer only knows what it’s allowed to know.
When ChatGPT was launched, for instance, its computer brain had a limited awareness of the world. It was only taught about events that occurred before 2021, which meant that a lot of the weirdness of the past three years — COVID-19 vaccines, Elon Musk, Harry and Meghan, the midterm elections, Squid Game, that sort of thing — were totally absent from its memory banks.
And while that may be a desirable state of mind when it comes to people — personally, I’d love to erase the past three years from my memory banks; hell, the past five years would be even better — when you’re talking about an automatic robot brain that’s supposed to know everything, that’s not so great.
Of course, some AI setups don’t have this specific limitation, but all of them are reliant on a dataset to function. And while the ChatGPT dataset may be too small for now, eventually, every AI application will have limitless access to … everything.
And that’s why I’m fairly sanguine about AI’s effect on my writing career. AI simply knows too much. And it has read too much. It will soon have access to everything ever written or documented, and I think we can all agree that everything ever written or documented is a fatally flawed dataset to use because most of those things are really terrible.
It’s a simple case of garbage in, garbage out. Throughout history, most novels have been bad. Likewise, most movies and television shows. And as long as that’s the dataset that AI uses to help greedy movie studios replace (expensive) human writers, I think my paycheck is safe.
Put it this way: If I ask AI to help me create and develop a new half-hour television comedy, it will draw on some of the very best examples of the category — Cheers, for sure, but also The Honeymooners, Friends, All in the Family — as well as some of the very worst. How will it know, for instance, that Night Court is hilarious but that Small Wonder, a first-run syndication comedy about a little girl who is a robot, was really awful? AI might, in fact, prefer Small Wonder, on a robot-to-robot kind of level. The script that comes out of that AI bot is going to be just as bad, and maybe worse, than everything else.
Eventually, of course, these drawbacks in AI dataset design will be addressed by the marketplace. Some enterprising character will start to train AI applications on only the very best examples of television and movie scripts. Instead of drawing on a wide base of everything Hollywood has ever produced, both the very good and the very bad, what will emerge are some very expensive, luxury high-end datasets — just the very best inputs, curated by experts and influencers. Available for a premium price, of course.
In the future, before a screenwriter begins a project, he or she will order up a bespoke dataset. Writing a romantic comedy? The high-end dataset will include You’ve Got Mail, How to Marry a Millionaire, Breakfast at Tiffany’s, and How Stella Got Her Groove Back. But not, say, Just Go With It or I Now Pronounce You Chuck and Larry. Writing a thriller? Load the set up with The Bourne Identity and The Wages of Fear, but make sure to expunge any traces of Basic Instinct II and I Know Who Killed Me.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
There will be datasets that are trained exclusively on scripts for movies that have made more than $1 billion, and some that are only aware of television shows that have inspired multiple spinoff series. The Hollywood writers of the future won’t be hired to write. They’ll be hired to compile a dataset of previously released movies and television shows to copy and steal from, which is sort of what they do today. It just takes longer.
Rob Long is a television writer and producer, including as a screenwriter and executive producer on Cheers, and he is the co-founder of Ricochet.com.