One of the themes that runs throughout this on-going investigation is the question of utility, or — if you like — value. For something to be “of value” one has to first come to the problematic with values. The way I usually do it is to ask a primary, structuring question: What is the condition in the world that I want to see? If I can answer that clearly, I have a goal, and on the basis of that goal I can begin to assess what is “good” in this world that can help me move successfully, or confidently, or at least better in the direction of that end-state.
A tool is of value if it helps you move better towards that goal.
So far so good.
In some cases, we almost all share the same preferred end-state and we have little trouble sharing our evaluations of tools. Take the stapler, for example. Unless you have some questionable habits, your preferred condition is: “the sheets of paper stick together (within the standard expectations of a staple rather than, say, glue).” Is this a good stapler? That’s pretty easy to evaluate now that we know what we want.
If, on the other hand, you’re a museum curator and your task is to collect design artifacts from the 20th century and demonstrate how certain eras drew off of shared influences, you might be in the market for an iconic stapler design (in great condition) as your display object. Whether the damn thing works (e.g. the internal spring is missing) is the least of your concerns because you have no interest in sticking sheets of paper together. This stapler is, in effect, retired from that career.
So … different values based on different criteria, based on different goals.
(Conceptual frameworks are very, very helpful when trying to get your head on straight).
Today the topic is screenplays. Why? Well … I’m writing one and I anchor everything I’m doing here in the practice of a creative writing professional (that’s me). I engage and report on my use of AI and what I’m learning. Someday this will be useful. (As the line goes in the movie, “Maybe not today, maybe not tomorrow, but soon and for the rest of your life …”).
I am turning The Curse of Pietro Houdini into a feature film. Whether the director and producer get the thing off the ground is a conversation for later.
So why did I approach AI (in this case, ChatGPT 4.0)? Was it to drop large parts of my novel into and say, “Make me an Oscar-winning screenplay from this, or, failing that, a beloved film that will make me gobs of money”?
No. Although, now I’m starting to wonder why.
Never mind.
The short answer is: Formatting!
Q: Do you know Final Draft, the industry standard for screenplays?
Yes. Yes, it did.
Q: If I provide you with a scene I wrote can you help me format it better to industry standards?
Yes. Yes it can (really).
So off I went. Scene by scene I asked it to clean things up. It helped nudge me into crisper if more boring descriptions of action; helped me differentiate a “transition” from a “shot” and also helped me understand better when to add less common notes into the “action"
At one point Dino (a young Italian soldier who has deserted) is explaining his ordeal to a group of monks and Pietro Houdini. The story is a bit long and involved but very dramatic and I decided (at least for this draft) that I wanted the scene to transition to a series of flashbacks with Dino providing the voice over. But I wanted the scenes to be less “documentary” and more told as memory: which means Dino’s innocent, naive, and kind personality would structure and present the visual storytelling.
How do you explain that in a screenplay?
The short answer was: I didn’t know, and I wasn't keen to watch two hours of YouTube videos hoping to stumble on the answer, nor did I want to read twenty screenplays where I vaguely remembered a similar technique being used.
So I write the scene as best I could, but then explained to AI what I was trying to accomplish. This was the result:
STYLE NOTE:
The following flashbacks are not perfect historical representations but are filtered through Dino’s personality: comedic, naive, and slightly exaggerated.
BEGIN VOICE-OVER: DINO
EXT. GREEK ISLAND - DAY (FLASHBACK)
A peaceful Greek village under occupation. Italian soldiers, including Dino, patrol the streets. Dino is trying to peel an orange and drops his rifle. An old Greek woman hands it to him upside down. He tips his hat, embarrassed.
DINO (V.O.)
Mussolini was deposed and captured and sent to the top of the mountain in exile. But then these German commandos captured the hotel he had been placed in and flew him away. Okay.
CUT TO:
EXT. ITALIAN MOUNTAIN - DAY (FLASHBACK)
A German commando unit swiftly overtakes the hotel. The commandos escort a disheveled Mussolini to a waiting plane. It takes off into the sky.
DINO (V.O.)
So then they put him back in power and we have the new Mussolini government called Salò. Okay.
So that helped.
Scene by scene, I fed the material into AI and it came back with tweaks and recommendations that assured a better alignment of my vision to the expectations of a director or professional screenplay reader.
In almost no circumstances did AI provide new or better dialogue. I say “almost” because sometimes it would omit a word or two and I decided that — yes — less was more and the line was a little better that way. But this was not “writing". At best it was suggestive copyediting.
What did happen, though, was AI would sometimes suggest or change the Actions or descriptions and it provided me with a baseline from which I could them imagine something better. Said differently, I would read it and think, “If that’s what the computer can do, I can do better and will.” So the automatically generated text set the bar that I then set out to exceed in terms of writing quality, directorial insight, and viewing experience.
It was not difficult. Maybe, someday, it will be. We’ll see.
What we’re back to, then, is AI being extremely helpful in providing a natural-language interface to provide standardization solutions to my creative acts.
While some screenwriters may try and make the case that every shot, every transition, every word is a creative or artistic act, my response is to roll my eyes. I’m a novelist. I already consider this a less art form. Yes, I absolutely do, and unapologetically. It is a skill and writing a good screenplay is very hard especially if you’re writing it from scratch, which I don’t do because I’m adapting my own novels. But the complexity, richness, and open structure of the novel is a nearly infinite canvas to compared to the expectations, limitation, and much shorter duration of the screenplay. I’ll defend this more later if asked to.
In conclusion: AI is an excellent editorial assistant, and occasional professional mentor, in formatting screenplays once you already understand the basics. It is not a writing assistant, and it is certainly not a writing substitute but — as with the valuation of the stapler — it all depends on your definitions and standards of value. Because, on request, it can create something (derivative and bad) that is perfectly formatted.
More later.
— DBM, 23 August, 2024
You say, Derek ...
"In conclusion: AI is an excellent editorial assistant, and
occasional professional mentor, in formatting screenplays
once you already understand the basics."
No! I don't think ChatGPT is "an excellent editorial
assistant, and occasional professional mentor," but, from your
brief illustration and explanations, it may be, for you, a
useful aid in "... formatting screenplays once you already
understand the basics."
ChatGPT did not bring to the work you describe the knowledge,
understanding, reasoning, experience, and skills, of a
professional editorial assistant. It didn't do this because
ChatGPT doesn't have any of these things, none! It just looks
and feels like it does to you. But this is an hallucination;
an hallucination you have, not ChatGPT. ChatGPT doesn't have
hallucinations. It uses a massive statistical model to
generate stuff that is close, but not necessarily correct or
truthful, to forms and patterns in the [ginormous amount of]
data it is programmed with in its construction.
What I think really happened here is that you used ChatGPT,
which can generate sufficiently good screenplay formatting in
your judgement, as a kind of mirror which reflects back to you
in certain [strangely] changed and distorted ways what you
present to it. And it proved to be useful to you in what you
were doing because it can make changes and distortions in how
it reflects things back to you of a kind you can't [easily]
do, but which directly aid you in doing what you're trying to
do, and, in particular, which aid your thinking and reasoning
as you do it.
We use other common [computational] tools like this; a
Spreadsheet App, for example. We use a Spreadsheet App to
tabulate some data in a neat and tidy and visually accessible
way, to see things in our data, which results in us changing
our tabulation and presentation, allowing us see other things.
And we can get this App to do things for us we can't do, or
which would be difficult for us to do, like calculate some
useful statistics on the tabulated data, and to make graphical
presentations of these statistics, again, helping us to see
and understand more about the data we have tabulated. It's a
conversation with ourself, aided by -- indeed, made possible
by -- the Spreadsheet App. We've been doing this kind of
cognitive working for thousands of years, in writing, in
drawings and sketching, in making, in improvising, in playing.
They all aid us in sustaining useful and effective
conversations with ourselves as we work out how to do
something. ChatGPT is another aid to doing this common kind
of talking-to-ourself working.
So, I would say, in working on your screenplay script you
usefully, and successfully, in your judgement, used ChatGPT to
have a useful thought and reasoning provoking conversation
with yourself on what you were doing, as you worked on and
formatted your script. It's good stuff. ChatGPT was
important in making this useful talking-to-yourself work
happen, but the only intelligence present in all this, of any
kind, was yours. There's none in ChatGPT. To think there is
is to fool ourselves, and to then misrepresent to others what
ChatGPT is and does.
-- Tim