As a member of The Companion, you’re supporting original writing and podcasting, for sci-fi fans, by sci-fi fans, and totally free of advertising and clickbait.
The cost of your membership has allowed us to mentor new writers and allowed us to reflect the diversity of voices within fandom. None of this is possible without you. Thank you. 🙂
Laurence Moroney is a man for whom the stars – if you’ll forgive the pun – seem to perpetually align. Luckily for us, he uses that power in the service of Stargate fandom and advancing the appreciation of artificial intelligence rather than scratch cards.
As A.I. Advocacy Lead at Google, his technical expertise birthed 2021’s Stargate A.I. – an innovative world-first table-read by the Stargate cast of a new script written by artificial intelligence – but the web of happenstance that set things in motion has been decades in the making.
“When Stargate Universe first launched,” he explains taking us back to 2009. “I was working at Microsoft and we had this technology called Photosynth, where you could take photos of a scene and then stitch them together into a 3D model. MGM and Microsoft were partners, and they’d brought in some folks from Microsoft to say, ‘Hey, we got this new show. They’ve built these expensive sets up in Vancouver. We’re looking for some innovative way to digitally market the show’.
“A few folks were scratching their heads not really sure. They wheeled me in, and I said, ‘Wait a minute, Stargate!’”Laurence Moroney
Rather than a corporate shotgun wedding, the end result was a 3D virtual tour of Destiny, the epic and enigmatic spaceship at the heart of SGU. Rooted in his appreciation of the fans and their values, the project successfully whetted appetites for the show and earned Laurence a page on the Stargate wiki. Although Stargate Universe came to a premature conclusion after only two seasons, leaving audiences without Stargate for the first time in 14 years, it’s in the times of franchise famine that fandoms thrive – just look to Doctor Who in the 1990s.
“It kind of got me a little bit involved in the whole Stargate ecosystem,” he recalls. “I was a writer on the now-canceled comic book that never actually saw the light of day. For Stargate Universe, I wrote the prequel comic books, a series called Icarus. But for various reasons, the publisher canceled it.” Redditors still trade his totally unofficial scripts for SGU Season 3, IMDb holds credit for his consultation on a 20-minute fan film, Stargate Universe: Distant Hope, and has a couple of original SF novels sitting on his Amazon Author Page alongside AI and Machine Learning for On-Device Development. It’s a strange sort of legacy for Microsoft Photosynth.
Stargate | Everything You Need to Know About Stargate A.I. Version 2.0
A new script reuniting Richard Dean Anderson with Amanda Tapping and Michael Shanks in celebration of Stargate SG-1′s 25th AnniversaryRead
Head in the Cloud
In the meantime, Laurence had joined Google’s A.I. team. It was another odd coincidence, as his history with the field began as far back as 1991 when he graduated with a physics degree “right into the worst economy the UK had had since World War II. In a terrible economy, being a physicist with an Irish accent was not a good thing. So I was unemployed for a while. Then [Prime Minister] John Major’s government started this scheme – and it was amazing foresight – where they created a cohort of 20 people who know A.I. and maybe that’s something that could be useful to industry.”
This was the death rattle of the ‘Alvey program’, launched in 1983 as a response to Japan’s rapid advances in computing and electronics. Beset with limitations – amongst other things, the working definition of artificial intelligence was in constant flux – by the time Laurence found his way into one corner of it, Britain’s tech sector was running up a deficit with the rest of the economy, and funding dried up.
“It failed within three months, but I was able to parlay that through a lot of hustle into a master’s degree. And that’s when my career began. But the vision had been there and that I was bitten by the bug.” His deeper perspective on artificial intelligence found its moment when towards the end of 2017, Google declared its ambitions to be an “A.I. first” company.
“A.I. was seen as very heavily mathematical. And I’m like, ‘I don’t want to see calculus ever again in my life’. I left school for a reason. But I realized there was a lot of power there, but we’re teaching math, so I brought it to the A.I. Leads at Google. I was like, ‘If you want to train millions of people, you’re doing it the wrong way.’ They were like, ‘Okay, you do it.’ And here I am.”Laurence Moroney
This mandate to educate Google has since become a mandate to educate the globe on the potential of artificial intelligence. At the annual Google I/O conference in 2018, the company’s CEO Sundar Pichai gave the world a glimpse of how the company’s new focus looks, from the jaw-dropping – a recording of a very human-sounding Google Assistant booking an appointment at a hair salon – to functions we already take for granted and probably don’t associate with AI. These included Gmail’s Smart Reply function, bespoke curation of Google News, and suggested edits in Google Photos, but it’s the fear of Google Assistant going full-bore “I’m afraid I can’t do that, Dave” that generated the most headlines.
“The thing that I really hope to drive,” says Laurence, “particularly in the sci-fi community, is that A.I. is not really what you think it is. It’s very, very different than what you think it is. For generations, we’ve been reading sci-fi books and A.I. is either like an evil all-conquering machine or it’s a Pinocchio. There’s so much more that can be done in between those extremes, that when you understand what AI really is, there’s a whole rich vein of potential storytelling there and storytelling that’s very much in line with the real world. That’s one of the things I would love to be able to influence in the sci-fi creator community, given that I’m a sci-fi geek myself.”
Star Trek | Universal Translator Tech Is Pretty Close – So Are Its Challenges
In The Science of Star Trek, Becca Caddy discovers how A.I.-aided translation is quickly gaining ground, but Shaka when the walls fell all the same
Stargate A.I. my Destination
So over there was Laurence Moroney, a Stargate fan with access to the most sophisticated artificial intelligence on the planet and a mission to spread the word about A.I.’s potential. Over here, Stargate SG-1, Atlantis, and Universe co-creator Brad Wright and Companion co-founder Lawrence Kao were dreaming up a way to get the cast back together for a table read.
More than anything else, fans longed to see their favorite characters again, but with a new Stargate series in contention at MGM, Brad was understandably leery about giving too much away. His stars would need to read something, so how about an A.I. write a script? It was a solution to potential studio sensitivities, and best-case scenario, based on the track records of chatbots and open-access A.I. like GPT-2 and 3, it would be funny bad, not bad bad. Brad put the call out on Twitter for an A.I. expert to help him out. Stars aligning, yet again.
“I’m sure they’d forgotten all about me. They didn’t know that I’m doing what I’m doing now. I was like, ‘Wouldn’t it be great if you knew somebody who was like an A.I. lead, like a major AI company?’, kind of mocking Brad’s tone a little bit, and then Brad responded, ‘Oh, yeah, it would.’”Laurence Moroney
The result of that is clear to see (literally so, here’s the full video), but perhaps what flew under the radar a little was just how innovative this was, and that this innovation didn’t come from brute force, but from smart use of smaller A.I.s.
“That’s been the difference with the Stargate A.I. project over every other project that’s been out there for generating scripts. I didn’t feed all of the Stargate scripts into one model and have that model write stuff. I’ve created… I can’t remember how many models now, there are at least four models for each character. There’s a model for how the character responds to action descriptions in the scene, there’s a model for how the character responds to somebody speaking to them because that would be different. If there’s an explosion in the scene, you’d want the character to go run to cover. But if somebody says hello to the character in the scene, you don’t want them to say right away, ‘Get to cover!’ There’s a model for each character for how they would continue speaking.
“If you imagine you always give a prompt, and then the prompt has some kind of reply. So the prompt might be somebody’s saying hello to Jack. And then Jack would respond, but then Jack’s next line of dialogue isn’t prompted by another person. It’s from his own internal dialogue. So you need a third model for that. And then the fourth model would be how a scene description would react to a character’s dialogue. So as Jack says something, what’s the next thing that would happen?”
Party of Four
Stargate A.I. Version 1.0 featured four returning characters from Stargate SG-1 and Atlantis, Amanda Tapping’s Sam Carter, Michael Shanks’ Daniel Jackson, David Hewlett’s Rodney McKay, and Jewel Staite’s Jennifer Keller – that’s 16 models as a bare minimum to generate the scenes you saw. Where it got even more interesting was the scene featuring Jackson and Keller, as the two characters never shared a scene in the series, the A.I. couldn’t be trained as it had for Jackson/McKay or Jackson/Carter exchanges.
The scene was a standout favorite from the reading, featuring a prickly Daniel Jackson shutting down a flirtatious Keller with “I could read.”
“I tried to create a specific Daniel model for responding to people who are more intellectual,” Laurence explains. “So I think I created a Daniel model for responding to Carter and McKay. And I created the Keller model for responding to McKay and somebody else. So while they’ve never been in a scene together, they’ve been in scenes with similar people.
“That was what was unique and very, very different about Stargate A.I, and we double down on that with Version 2.0. Many of the scripts are becoming a bit more coherent and the ultimate vision for this would be that if we start getting more and more focused, then we can start getting more and more value out of these scenes.”Laurence Moroney
It’s unlikely that artificial intelligence will be in a position to replace a human storyteller, even with the greater coherence in Stargate A.I. Version 2.0. The model can only react to what’s in front of it, and by adding new models you can continue giving it things to react to, and by making those models more focused and specific, the reactions can become more sophisticated, but the ceiling is low on how far the A.I. can get without direction.
“You can’t rely on a model to tell a story,” says Laurence. And I don’t think that will ever happen. I’ve been working really hard on something to drive a more coherent story, pulling from other sources, not just the scripts, pulling from general rules of storytelling, so that you have like a story arc within a scene and you have a little hero’s journey and stuff like that. So a bit of bespoke work has been done there on that. And then a lot of work on linking each description in a scene so that it could be coherent.
“There are various methods in A.I. for you to say: here’s what I currently have, give me a whole tonne of predictions, measure the accuracy of those predictions, but then also measure the similarity in that sentence to the previous one. So a model might throw out something that isn’t as well connected with a slightly lower percentage of probability, but this syntactically is more connected to the previous sentence. So I’m having it select those. As a result, there is something inherited in sentence two from sentence one. And there’s something inherited from sentence three in sentence two. To be able to do those kinds of things to push a story forward is very important.
“Then another part is to do the same thing going backward. So if you imagine the scene begins with a woman standing on a cliff, and the scene ends with her jumping, right? You’re gonna have similar sentences moving forward, and similar sentences moving back and doing a search between all of these, so you end up with something connected. So a lot of bespoke work on that kind of thing is what I’m trying to get my head around.”
This, says Laurence, is the problem he needs to solve for the work he’s done with Stargate A.I. to go from project to product.
“Imagine some kind of visual tool where you say: I’m gonna write a Stargate script today, I’m going to pull in one model for generating a story. I’m going to pull in another one for each of the characters that I’m going to use. Then I’ll write my opening line: ‘Samantha Carter is lost in the desert and the sun is shining above her. She’s sweating. She looks at her flask – no water left.’ Then I go, ‘Okay, you model tell me the next thing that will happen.’ This model tells me the thing that she would say, give me your scores, pick the highest score injected into the script, and then maybe the model that says: Samantha goes, ‘I wish I bought water’, is the highest score, and I inject that into the script. What happens next? Then the story model goes: Samantha sees a mirage of a pool with palm trees beside it. It says, ‘There’s a very high probability that would be the next thing, is that okay?’ I’ll inject that. Then you start dragging and dropping these and assembling a story.”
Stargate | How the GPT-3 A.I. Wrote its First ‘Starcake’ Script
Research scientist and author of You Look Like a Thing and I Love You: How AI Works and Why It’s Making the World a Weirder Place, Janelle Shane on her test of A.I. screenwriting.Read
We’re still some distance from an off-the-shelf suite of storytelling A.I. tools, but by going small rather than big – working on his home Linux PC with an open-source platform called TensorFlow, as well as re-using bits of some publicly available Large Language models (OpenAI‘s GTP-3 is the best known of this family) – Laurence has made the future for A.I.-assisted screenwriting look probable rather than merely possible.
“There are musicians out there who use A.I. to help compose, there are artists out there who use A.I. to help, you know, make their music videos or make their paintings, and there can be writers out there who are using AI to maybe help inspire their stories or drive them along.”Laurence Moroney
If you use Gmail, you’ve probably noticed Smart Compose, launched in 2018 at that A.I.-led Google I/O conference. You begin typing a familiar greeting and the phantom conclusion appears, predicting the most likely outcome based on your greetings to date. Jab the tab button to use the suggestion, or try to outwit destiny by continuing anew. This is machine learning as a tool for writers right now.
“Gmail is a great example,” Laurence says. “If you took that to the next level, then how do you sign off your email? People in the UK might sign off ‘cheers’, right? People in the US might sign off differently. Or you might have your own particular vernacular for how you sign off. If they can learn that from you, [they can] then predict that you’re coming to the end of your email from the tone of your text, and then automatically generate the sign-off in your vernacular. I could see where this could be very useful to writers, particularly for long-running series, where it’s being trained on the vernacular of different characters, and then to help writers find the voice of that character, predicts the text for them, or analyzes the text that they’ve already written. Maybe Jack O’Neill never says, ‘bollocks’, because he’s not English. But he would say ‘crap’.”
Where large numbers of creatives are working at a remove from a source, whether writing power rankings for the back of a trading card, creating a proof of concept model kit, or writing licensed fiction for comics, books, and videogames, this sort of tool could prove invaluable. It could free up the creatives to do what they need to do when they reach the edge of the brand guidelines and drastically shorten the laborious approvals process.
Once upon a time, writers new to the Star Wars Expanded Universe were sent a crate of the RPG sourcebooks produced by West End Games, so comprehensively had they collated material from across the series – even inventing their own names for ships, weapons, and species, to fill the gaps. They were the exception rather than the rule in the land before Wikipedia, and even now, where a setting is entirely in the mind of a single creator – such as George R.R. Martin with A Song of Ice and Fire – you’re only going to get an answer when they get round to it.
Every rights holder, creative project, and tangle of responsibilities are different, but one thing they have in common is just how labor-intensive they are. Even now, in what is a golden age for convenience, assume you’re writing a Star Trek tie-in and the answer you’re looking for isn’t a Google away, you fire off an email to your editor, he emails someone else, and eventually you might get the answer you needed. Or you might have moved on, deleted the whole problematic line of dialogue about shore-leave gone awry, and who cares what plant life can be found on the surface of Betazed?
In another world, Laurence Moroney opens up a chat window and asks the continuity bot: what plants grow on Risa? The answer – scraped from every word ever written – comes in seconds. Maybe it’s filtered, so even if you get contradictory answers – you know one is from a comic and the other is from an episode of The Next Generation, and you can prioritize the latter. Or maybe the A.I. has been trained to make the call for you because it knows you’re working within the comic-book continuity.
“Wouldn’t it be nice if you had A.I.-based models trained on your existing work to help you? Like, ‘What color was the sand on Abydos?’, or ‘What happened the last time we were there?’ Particularly if you’re writing a long-running show like Stargate models could be built, chatbots could be built. Those kinds of things really ease the way for writers and then maybe ultimately end up helping them generate text. There might be people like Brad who understands Jack’s vernacular, but somebody in the writers’ room who’s recently been hired doesn’t, wouldn’t it be really nice if that person could write a line of dialogue and get it scored? There’s an A.I. trained completely under Jack’s vernacular and then it says, ‘There’s 70 percent chance Jack would say it like this, but maybe you should use this word instead of this one.’ That’s the long-term vision I see for this.”
Video | SG-1’s Richard Dean Anderson in Conversation with Brad Wright
Rick’s last convention appearance was in 2019, so enjoy these two Stargate veterans deep in discussion about a show – and character – that means the world to so many of us.Watch
Wraith to the Finish
Things quickly come full circle, as they have a tendency to with Laurence Moroney.
“Let me tell you how I first got involved in the comics. When they were being created, they were like, ‘Okay, we don’t know how good you are at telling a story. Take the crew of Stargate Atlantis, put them on vacation, and give me an adventure that they would have when they’re on vacation.’ I spent a lot of time coming up with an idea. The folks from Atlantis go to Ireland on vacation. It’s a classic kind of quiet man story, where they want to get away from all of these big galaxy-spanning epics, and just sit in a quiet pub in Ireland and have a drink and enjoy the scenery. In Ireland, there’s a legend of something called the Dearg Due, which is a Red Woman – almost like a banshee. When you think about the female Wraith in Atlantis, there’s one with the red hair.
“I’m like, ‘What if there was a Wraith buried in Ireland that led to this legend, and it’s waking up right now?’ And the crew just happened to be there and we have an adventure of them dealing with villagers and going into the woods. It was a whole lot of fun and it took me a long time to put it all together and write it all up. And then I submitted it to them. And they loved it. And they thought, ‘Okay, we’ll bring you on board as a comics writer and we might even do this one as a one-shot.’
“They submitted to MGM, and then MGM came back and said ‘No Wraith on Earth.’ Right. You could have told me that beforehand, you know?”Laurence Moroney
Whether the power of small but focused artificial intelligence or a reply to a tweet – it’s often the smallest of things that have the most meaningful impact.
James is editor of The Companion. He has been “working in publishing” since the early 1990s when he made his own Doctor Who fanzine to sell in the school playground.
You can find him on Twitter @JDHoare