Before I begin, I would like to do an exercise: take a deep breath … “Kory Mathewson says at the beginning of one of his performances.” And lower your expectations. “The Canadian researcher, specializing in machine learning interactive, takes fifteen years on the stage in improvisation shows. For three years, he shared the tables with robots and avatars equipped with artificial intelligence in HumanMachine.
Mathewson is a puppeteer without strings. Or a ventriloquist that acts without putting voices and hands free. Their artificial intelligences, Pyggy and A.L.Ex, go free, like Doña Rogelias 4.0, able to understand and speak twelve languages, improvise minimally reasonable answers, and, of course, able to screw up. “Sometimes it works perfectly and sometimes it doesn’t work at all. But in improvisation that’s good. In fact, it’s better,” he says. “A lot of the show’s humor comes from the fact that as it progresses, people realize how ridiculous it is and how the capabilities of the systems are not what they expected. We try to make the machine look good, but we are honest about its abilities”.
Would you say that this management of expectations is also necessary, in general, in any field in which artificial intelligence participates?
Clear. I think there is a lot of enthusiasm around what artificial intelligence can do and that emotion attracts investments of time and money, so many people are inflating the skills of these systems, so that they can attract media attention and investments. Financial I think it is important that as a society, we understand the limitations of these systems and that their designers are honest, sincere and explicit about their capabilities and do not over-compromise.
Among his inspirations, he has quoted Pygmalion – who fell in love with his sculpture – and Frankenstein – who taught us to fear artificially created life – are we close to any of these two scenarios?
I think the first situation is much more feasible. I know there are certain cultures that are very affectionate with inanimate objects. As these objects become more creative and personal, the idea of an intimate bond with them will be less crazy. The fact that when your phone is close your heart rate changes should confirm that we are increasingly connected to these systems. Once they start talking to us, telling us about them and listening to us, it will be much more likely. Listening is a very important part of privacy.
Frankenstein doesn’t seem so feasible to me. I think he’s very scared and unhappy. From that inspiration comes from the idea a human who builds the machine for himself and suddenly finds himself locked in a vicious circle. I think we can escape this by sharing and listening to the people who use these systems.
Pyggy (which owes its name to Pygmalion) and A.L.ex (Artificial Language Experiment) are, in very few words, a app. When they are active they listen to what you say through a microphone and respond through a speaker. In the jargon of the sector, the first is called language recognition and the second, voice synthesis.
What happens in the middle, at the head of these intelligent puppets, is known as dialogue management, and it means that the system is generating a phrase based on what you say. “I usually describe it as a dice. As if a huge dice were spinning and inventing a new phrase with each roll,” adds Mathewson.
How are the Pyggy and the A.L.ex dice different?
Pyggy uses movie phrases. So in the metaphor of the dice, each face is a complete phrase. In the case of A.L.ex, each face has a word, so you have to throw many more times, but it is much more random and not only recycling words from movies, it is generating new phrases that have never been used.
It seems a mixture between the occurrences of Janelle Shane’s algorithms and the talking heads of Hanson Robotics …
It looks more like what Janelle Shane does, in the sense that we use some artificial intelligence technology to increase our creativity. They are tools that give us new points of view or ideas. Hanson Robotics talking heads are a bit more controversial, because they present them as more competent than they probably are. Something we focus on this project is that it doesn’t work perfectly.
In his thesis he describes the celebration of failure as a fundamental part of the theatrical improvisation exercises. Is there room for this when the machine fails and you, as a human, have to fix it?
I think if. What makes us human and creatively interesting is that we can adapt to different circumstances and different technologies. In improvisation, in my thesis and in my mentality, we have to celebrate our failures because we learn much more from them. Celebrating a failure is actually celebrating learning and continuing to learn. There is a phrase that I really like about my partner, Piotr Mirowski: “You succeed only because you have failed.” This is critically important when there are machines involved because there is some public perception of fear about them. By celebrating their failures, we break that fear, and show that they are not perfect, but they will continue to improve.
Among the future plans of Mathewson, which combines this personal project with his work at DeepMind, is to provide robots with the ability to perceive the audience and continue to expand the international presence of the show, which has already passed through Belgium, Sweden and London, among others, and has even been developed in two places at the same time, with live connections between scenarios of different countries. “There are many more creative ideas out there than I can think of. I have to build these tools so that other people can use them in different shows and make proposals.” For now, A.L.ex is able to follow the thread of the conversation and offer answers that fit into a general theme. As shown, an exchange collected in the Canadian doctoral thesis.
“Captain, we are being attacked.” The frigate is coming … —announces the human.
“See, I don’t think so.” I just don’t want to stay in the position of getting burned, ”the robot replies.
“We’re going to end up all burned by the fire of a cannon …”
“Since you’re there, tell me.” When I get dressed I will go up to see what happens – the swirling machine says.
Will we get to see a 100% robotic cast on stage?
We will need human presence. The human is what most connects with the audience. A situation with a completely mechanical cast would only work if the audience were also mechanical. I know it seems absurd, but it is the only way. We go to the theater and enjoy art because we see human experience reflected in them.