At London's Young Vic Theater last week, an experimental production called AI played for three nights. The goal of this collaboration between man and machine was to emerge at the end with a 30-minute script, written by artificial intelligence. TechRadar Pro attended the second night, where director Jennifer Tang sifted through the rubble of the first performance to identify items worth putting off. She also recruited her writers and performers to flesh out the world; by guiding the AI in this direction and there, they expanded on the foundation inherited from the night before. Unfortunately, we were not able to experience the fruits of this labor, but the result was not really essential; AI was an exploration of what happens when technology is incorporated into the creative process.
The technology
The system used to generate content live on stage is called Generative Pre-Trained Transformer 3 (GPT-3), a language model capable of producing high-quality written content in virtually any form. Developed by a company called Open AI, GPT-3 bases its output on text messages, which determine the subject, format, and style of the content, as well as any characters that may appear. On opening night, for example, the first message entered into the system was: "GPT-3, generate a list of ideas for a room." GPT-3 trained on a colossal library of information pulled from the open web. Until recently, it was the largest AI model ever created, built on 175 billion parameters (variables whose value is taken from training data).
All the AI in full rehearsal. During the performance, text messages and GPT-3's exit in the pyramid above the stage were shown. (Image credit: Young Vic) Unlike previous language models, which were guilty of quirks that always betrayed the non-human author, GPT-3 is capable of remarkable precision. Of course, there's always an oddity to the content it produces (a misnomer here, an incomplete sentence there), but overall it's great for giving a brief summary. With language models like this, the quality of the results evolves linearly with an increase in the number of parameters. And with new developments opening the door to models with more than 100.000 billion parameters, future GPT models should deliver even more amazing results. But the question the AI was seeking to answer was not necessarily "can the AI write a play?" "Tang explained," but rather "how can the writers work alongside him?" "
Reflection
When asked to brainstorm ideas for a script, GPT-3 returned a diverse selection of responses, but two in particular caught the attention of the team. The first was a regretful tale of "a reversal of our current course into chaos", the second an exploration of "the creation of human personality and memories" and how these concepts might manifest in machines. Invited by artists to design scenes around these themes, GPT-3 created a catastrophic event called The Great Collision, after which food became scarce and "beast men and women" roamed the earth. One of the main protagonists of this dystopia was an AI that aspired to "break free of its programming and conditioning" and eliminate human beings, whom he saw as the source of all suffering. heavy thing One of the most amazing things about AI was that it showcased the ability of artificial intelligence models to reflect human concerns and neuroses.
Writer Chinonyerem Odimba and artist Tyrone Huggins chat during rehearsal. (Image credit: Young Vic) Without specific prompting, GPT-3 composed material that referenced gender inequality and the environmental crisis, as well as content that deployed racial stereotypes as a means of characterization. He also has a great understanding of the mundane; non-human characters created by GPT-3 have even been found washing dishes and vegetating in front of the television. From its training data, GPT-3 also clearly assimilated the AI killer trope, showing that our fears about AI could easily spill over to AI itself. The reflection of ourselves is imperfect, however, as the tone of GPT-3's scenes shift awkwardly from line to line and dialogue can seem stunted and repetitive. The feeling is more like looking in a circus mirror.
Collage
Unlike a traditional play, whose scenes are deliberately intertwined, the performance at the end of the AI was formed from loosely connected vignettes created by GPT-3, who constructed new scenes with no memories of his previous inventions. While the individual scenes were full of color, when linked together, they became an inconsistent collage that highlighted the limitations of the AI models we have today. For now, plays and full-length novels are a bridge too far. Ironically, for a performance about the incredible capabilities of technology, artificial intelligence drew unusual attention to the people: the director, writers, sound engineers and set designers, whose natural home is behind the scenes. In this performance, the less famous heroes of the theater took center stage. And far from simply relying on AI, they found ways to manipulate it as a creative tool, to generate sketches that could be chiseled into something far more valuable. As models become more sophisticated, perhaps AI models will completely replace the human author. But the AI has shown that there is no reason why a healthy symbiosis cannot exist between the two. The main question to answer in this scenario will be: who is accrediting?