How’s that for a come-on line?
For many years I have pointed out that there’s a huge stock of soap opera scripts that could be mined for interactive storytelling. With LLM technology, we now have exactly what we need to accomplish that. Doing so will require a combination of efforts, so we’ll need a startup to put those pieces together. Here’s a sketch of the basic scheme:
Personnel
First, somebody forms the startup. I will NOT be part of it; I just turned 74 and the manic pace of startups is for younger people. I’ll be happy to advise, but I think it’s best for all if I’m on the sidelines.
After much contemplation I have come to a conclusion that I deem crucial to the success of the operation, yet many readers will be displeased with. I strongly believe that the CEO and driving force behind this startup must be first and foremost a storytelling artist. Not a techie: a techie CEO will strive to build the best technical storytelling LLM possible, which would be very impressive technically but a failure commercially. Nor should the CEO be an academic; academics just don’t have the commercial mindset to run a company. My ideal CEO would be a person with a background in storytelling, be it cinema, television, or popular novels.
Obviously the startup will need a CTO with strong credentials in AI. While such a person will probably be young, the position will demand some leadership talent, especially in matters of dealing with underlings. There are plenty of great techies out there who couldn’t manage a herd of three sheep.
And of course the startup will need a bold marketing person. This company will be trying to sell a completely new kind of product to an audience that does not yet exist. The marketing person’s primary task is to create a market, not sell into an existing market.
Procedure
The first task, it goes without saying, will be the creation of the business plan. This permits the next step: assembling capital. I know nothing of these two steps, so I leave them unaddressed.
With capital, the CTO can begin hiring their team and work on the custom LLM AI for the company. Meanwhile, the company must retain an attorney skilled in the new art of licensing source material for the LLM AI. That attorney and the CEO will then tour Hollywood studios to obtain rights to all their soap opera scripts for use in their LLM AI. There are 10,755 episodes of “All My Children”; 12,000 for “The Young and the Restless; 11,136 for “One Life to Live”; and 14,000 for “General Hospital”. 13,858 for “As the World Turns”; 8,891 for “Another World”; 9,000 for “The Bold and the Beautiful”, and on and on. These here add up to almost 80,000 scripts; surely there are in excess of 100,000 scripts for daytime soap operas. Add in the prime time soap operas and you’ve got quite a large data set.
My guess is that most of these scripts will be on paper and will need to be digitized; with this done, the material can be used to train the LLM AI. We would want to include a number of rules to hone the performance of the system. I don’t know enough about the strengths and weaknesses of this technology to confidently propose such rules, but here are a few examples to consider: identify certain standard roles, such as innocent sexy female, wanton sexy female, sexy male, lechorous sexy male, lecherous wealthy senior male, advisory older woman, and so forth. When each actor is assigned to one of the roles; this makes it easier for the system to unify the material in different soap operas. I expect that much more complex algorithms will be required to produce good storyworlds.
Once the system is operational, the system sells “interactive soap operas" on the Internet. I expect that the potential market for this product would be stupendous, but developing that market would be a slow and expensive process.
Expansion
Once the soap opera system is operating, it could be expanded in a number of dimensions. First, it could be internationalized, with soap operas in different languages for different countries. However, given cultural differences, it would not be possible to simply translate American soap operas into different languages; there are plenty of native soap operas that would have to be licensed in order to correctly address cultural differences.
Another dimension of expansion would be different classes of stories. There are thousands of scripts of TV Westerns that could be used, although Westerns don’t have much appeal these days, and the scripts of TV Westerns date from the 1950s and 1960s and reflect those values. Fistfights aren’t that big a draw these days.
Then there are mysteries. Again, we have thousands of these to work with, but I worry that a great many mysteries hinge on microscopic details that are not easily handled by LLM AI systems. I recall an episode of “Poirot” in which a woman who is an artist mentions to Poirot that she hates horses. Much later, while searching through her art studio, Poirot notices a clay head of a horse, and immediately begins tearing it apart, much to the consternation of the other investigators. Sure enough, Poirot finds the murder weapon inside the horse head. That’s not the kind of thing that generalizes well; I really can’t see people tearing apart clay eagle heads, clay fish heads, clay crabs, and so forth. Moreover, most mysteries are more puzzle-like; this may prove frustrating to players.
Another field to plunder is science fiction. Indeed, we now have two gigantic franchises, Star Trek and Star Wars, each of which has enough material to be handled separately. I think we could finally realize Brenda Laurel’s dream of sitting in the captain’s chair of the Enterprise, making life-or-death decisions.
My role in this
As I earlier stated, I will not play an active role in this. I can provide one useful service, though: I shall act as a communications hub in the earliest stages of gathering people. Once a group has formed and established an identity, I can hand off that function to them. You can contact me by using the Contact Form at the top of this page.
So get to work!