Cold Readings — you either love them or you hate them.
You have very little insight into the script. You don’t know how the scene is going to resolve, let alone the entire story. In a way, it can be scarier than improv because unlike improv there can be a wrong choice that you’ll need to go back and fix because, in the end, your choice might not really fit what you later understand the point of the script to be.
In the realm of AI, GPT stands for Generative Pre-trained Transformer. In the case of chatGPT, it is making its best prediction for what the next word should be based upon its dataset and what it has learned so far. In the case of the audiobook narrating AIs, they are making their best prediction for what the next emotional response should be based upon their dataset and what they have learned so far.
Making predictions based upon a text that I have never read? Guessing at what will happen on the next page, let alone the next chapter or the one after that?
That sounds like trying to cold read an audiobook!
Essentially, they are not trying to build a generalized acting AI, they are trying to build an AI that is good at cold readings. And that is a much smaller subset of acting than almost anything else could be.
The AI scans through billions of sample sentences, finds similar content, aligns that content in a way where similar content also appears from both the previous and the following sentences, and then makes the most likely choice for how the voice is modified to mimic the most appropriate human-like response to what is being read.
It is a cold reading, fullstop.
The AI is not acting. It is mimicking the most usual human reaction that it can find in the myriad of samples that it can pull from its database.
But Steven, you might say, the computer is reacting to the words on the page. And wasn’t it Stella Adler who said, “Acting is reacting?”
But, the computer is not reacting. The computer is analysing its database of usual reactions to similar pieces of text and then it is manipulating its vocal emulation to indicate a response that it believes will have a high probability of being the expected reaction by the listener.
There is no surprise. There is only predictability.
Cold Readings are fun, but they are rarely where one makes deep discoveries within the text. Those usually happen deep inside the rehearsal process during a moment that someone like Stella Adler or Sanford Meisner would be cheering on from beyond the grave.
The AI are not, as Meisner would have put it, having a true emotional response to an imaginary circumstance. They are choosing the most likely response from a list — a huge list, but a list nonetheless.
There are some people predicting the end of audiobook narration with the coming of AI. I am not.
I think traditional human storytelling will live alongside AI-Enhanced Text-to-Speech for a long time. And, since many publishing decisions are made based upon cost and bottomline concerns, I do believe that there will be many low earning titles that will be produced using AI in the future, both fiction and nonfiction, from major publishers and indie concerns all.
My advice: to borrow from a title of a book by Constantin Stanislavski, An Actor Prepares. So should we all.
Dive deep into your texts to find the treasure buried within. Find the core of the content and express that gem with surprise, sorrow, wonder, and glee. In both fiction and nonfiction, bring your inquisitiveness and sense of awe to the fore.
Be prepared, not predictable.