An episode of Stargate may be written by AI. The task was taken up by Google AI

Writing a television programme often includes a writers’ room and a lot of time, as people try to come up with the story and the dialogue that work together to create a show.

Character and narrative creation for the Stargate science fiction franchise’s three shows (SG-1, Stargate Atlantis, and Stargate Universe) was overseen by Stargate co-creator and executive producer, Brad Wright. Towards the end of 2021, Wright asked on Twitter whether an episode of Stargate might be written by artificial intelligence for publication on the SciFi insider site The Companion.

Laurence Moroney, Google’s AI leader, stepped up to the plate to show what artificial intelligence (AI) was capable of.

– However, at first he was unconcerned that artificial intelligence (AI) might replace him or other authors.

It was something I tossed out to The Companion as a notion since I had seen a number of AI models for scripts,” Wright told VentureBeat. “It’s amazing in some ways. “In other respects, it’s a pretty safe place.”

By November 2021, the AI-generated screenplay had gone through its first round of revisions. Wright found the writing to be intriguing, but he also found a lot of nonsense in it. Second time around, Moroney and Wright are working on a screenplay for a more complex and intriguing Stargate story.

If we’re going to do this again, I told Laurence (Moroney) that we need really pick up our game, and he welcomed that challenge,” Wright added. I was blown away by how much better it was than the previous version.

How did the AI script for Stargate come to be?

Stargate’s AI script was produced in the same manner as any other AI model: by training the model, of course.

Moroney fed the AI model every single word of dialogue and narrative from every single Stargate episode screenplay ever created. He used a wide range of tools, including Google’s TensorFlow machine learning framework.

A technology called transformers was employed by Moroney, as well as pre-trained natural language models.

To avoid confusion with Hasbro toys, Moroney said that “transformers” were first employed to translate languages. An ideal situation for script creation would be to take an input sentence and match it to an output sentence, as is the case with language translation.

Capt. Samantha Carter says something to General Jack O’Neil, for example, as an example of a suitable input phrase It is based on the transformer training on how Jack has reacted to input phrases like the one Samantha just stated.

This would allow me to programme a transformer to anticipate what Jack would say in any given situation, Moroney reasoned.

The universal sentence encoder, another key technological method, provided a numerical value for the context of a sentence. The semantics of phrases may be mathematically encoded, according to Moroney, such that the links between sentences can be understood both forward and backward. This time, Wright felt that the encoder in the second version of the Stargate API was an improvement over the first since it prevented the script from degenerating into meaningless gobbledygook.

Although Moroney could have utilised several Google machine learning operations tools to increase automation, he noted that most of the work to construct the Stargate script was manually. He asked the trained TensorFlow model to respond to a question he had given it. Moroney painstakingly constructed the script based on their replies.

“The model wasn’t putting out a screenplay with correctly structured text,” Moroney remarked. Models generated the relevant tags for insertion at the right times using a distributed approach.

Introducing scriptwriting AI to Google’s business operations

Enterprise might benefit from Moroney’s Stargate scripting approaches and concepts.. As for the event speech script for Google and Alphabet CEO Sundar Pichai, he joked that it’s doubtful that his AI approaches would be employed anytime soon.

“I could train a model on Sundar (Pichai’s) vernacular if I had all of his words,” he remarked. What if I could write something that would make me sound more like him, rather than myself? That would be a great benefit.”

As far as Moroney is concerned, an AI model may be able to speed up the process of creating keynote speech scripts for corporate events, but it won’t be able to handle all of the job.

When making an announcement at an event like Google I/O, “you’re going to speak about something fresh that nobody’s ever seen before,” he said. What this means is “by definition, you cannot train models like that.””

Artificial intelligence (AI) might also be used to create deep fakes, posing a significant cybersecurity concern. In theory, artificial intelligence (AI) that has been taught to speak a particular dialect may be used to assist in the creation of a deep fake, but Moroney said that he likes to see technology utilised for good rather than evil.

In an interview, he claimed that he was concerned about “deep fakes.” I’m also hopeful that there are tools available that you can employ to detect deep fakes, and they are really pretty simple to identify with the correct use of these technologies,” he writes.

The AI script for Stargate is excellent, but it will not replace people.

A common fear concerning AI is that it will eventually replace us, but Moroney isn’t concerned — at least not yet. AI-assisted technologies, in his opinion, have the ability to help lay the groundwork for future scripts.