This essay was adapted from a talk developed while in residence at TED (watch).
At the first-ever film screening in the 1890’s, a few seconds of grainy, black-and-white footage of an oncoming locomotive proved to be so shocking to the audience they fell out of their seats in fear. The story is an urban legend about the 1895 film Arrival Of A Train by the Lumière Brothers. As a founding myth of cinema, it captures the dismay and panic that moving pictures initially inspired, at a time when people were still getting used to the concept of a photograph.
In the digital age, the newest medium to come barreling toward us at full speed is AI, or artificial intelligence, and we are no less terrified.
The popular narrative around AI is that computers will reach an inevitable state of complexity and eventually become superintelligent. Any day now, they might surpass us in intelligence, become self-aware, and develop feelings and intentions of their own. When they finally enslave and/or kill all of humankind, we should not be entirely surprised.
We prize our human intelligence so very much — we believe intelligence is what separates us from the animals. But even if our monopoly on smarts seems to shrink with each new leap in computing power, machines can never be poets, right? More cherished than even intelligence is our human power of imagination.
Enter Shimon, a robot that improvises music on its own and performs alongside human bandmates: Shimon tours all over the world and has an enviable career as a musician — more so than a lot of human musicians I know. The Mandarin-language chatbot Xiaoice reached a career pinnacle many poets strive for by publishing a book of poems last year. More recently, Christie’s sold an image at auction said to be ‘created by an artificial intelligence’ for an eye-popping $430K. In each of these instances, computers engaged in artmaking — with spectacular success. Yet it is exactly their success that is so unsettling.
What does it mean to have algorithms generate poetry, music and paintings? Computers can automate vast swaths of repetitive tasks now done by humans, but are we really ready to take bot poetry seriously?
When I hear a piece of music, or read a poem, or look at a painting there is humanity in these experiences. I am connected to another human being who created the art form I am enjoying. If instead the work was generated by a bot, what will I connect with? Will the experience be any less real, any less human, any less vital? Compared to the slow unpacking of emotion and intellect and experience a good reader of a poem might need to do to fully understand its nuances, what are the thousands of poems an algorithm can cheaply generate in seconds? To be on equal terms creatively with a bot is a profound diminishment.
Maybe our humanity is irrelevant now: when that purported robot apocalypse finally comes, maybe computers will be justified in just wiping us all out. Or maybe are just not seeing a new medium for what it is.
The poetry generated by a bot can seem expressive — even overwhelming — in its novelty. But so far, most talk about art works generated by AI glosses over the crucial fact that bots are themselves created by humans. The poems a bot generates are an intriguing kind of artifact, and it is the bot itself that is the real work of art. A bot is a piece of software someone has made to act in our own, human, image: a kind of puppet that performs its uniquely-programmed song-and-dance.
We focus on the outputs of a host of autonomous-seeming software in many guises — bots, robots, algorithms — without seeing them as a medium in themselves. The key to understanding this medium is puppetry. For every puppet, there is a puppet master: a human directs the puppet to move and gives it a semblance of life.
The first, apocryphal, film audience focused on the illusion of the locomotive without being able to see the film apparatus responsible for that illusion: a string of photographs sped up to simulate motion. The mechanism at the heart of the illusion of software that appears to behave on its own — known variously as machine learning, neural networks, artificial intelligence — is a similar type of speed and repetition. All software is a tool, a list of instructions a computer follows. When it is asked to repeat a simple task, the computer can appear to be moving on its own, but it never does more than what its human programmer has instructed. It is never more than a puppet.
We can deliberately make software act in our own image now. We can give a computer a large set of words, and ask it to reconfigure those words to appear as if it is chatting or writing a poem. This way of using software to model a narrow slice of human behavior is a form of portraiture. While photography similarly makes use of technology — a mechanical process of exposing photo-sensitive film to light — in order to depict a person, we don’t typically fear a portrait photograph will be so realistic we will mistake it for the human it depicts.
The terror we feel at the approach of artificial intelligence will eventually seem quaint. When bots hoodwink people into believing in machine autonomy, when they behave so realistically they seem human, we might do well to remember the early film audiences at the turn of the last century falling out of their seats in surprise. From the vantage point of the projectionist’s booth, it would have been quite a sight.
Kat Mustatea, an artist and technologist, is writing a book about the meaning of machines making art. Watch her TED talk here.