Stories we tell, machines we build: a talk at the An{0}ther {AI} in Art summit in NY

I was invited to deliver the keynote talk at the An{0}ther {AI} in Art Summit in New York, at the New Museum on April 24, 2019. This is the unabridged version of my talk.

AIinArt

From left to right: Isolde Brielmaier (Westfield World Trade Centre), Zia Khan (VP Innovation, The Rockefeller Foundation),  Kamal Sinclair (New Frontier Story Lab, Sundance),  George Zarkadakis, Amir Baradaran (Founder and Lead Organizer of the Summit).

In 1649 Rene Descartes, the most famous philosopher in Europe at the time, accepted the invitation of 19-year old Christina, Queen of Sweden, to become her tutor. There is a strange and apocryphal story that recounts his boat journey from Amsterdam to Stockholm. In the story, Descartes is travelling with his charming daughter, Francine, but for some strange reason the girl never shows herself on deck, but is always kept away from everyone, locked inside Descartes’ quarters. That is enough to stir the curiosity of the crew, which turns into suspicion when the boat hits upon a tempest in middle of the sea. Fearful of bad omens, the sailors seek the young girl who has so mysteriously avoided them. Was she a witch? Did she possess some alchemical powers? Had she summoned the devils of the sea? With bloodshot eyes they descend into her room and kick their way through the locked door. But the room is empty, except for a wooden chest that is firmly shut with a padlock. Hastily, the seamen crack the chest open and, to their horror, they find inside a living doll, a mechanical automaton that moves and behaves just like a human being. The signs of dark magic are obvious for all to see; and for the captain it was an easy decision to throw poor, mechanical, Francine overboard.

urn_cambridge.org_id_binary-alt_20171013085401-80536-smallThumb-S147924431600024X_fig1t

Francine, the robot daughter of Descartes, thrown overboard.

You can read in this story what you like. For instance, how prejudice and ignorance may clash violently with scientific advancements, and often win. Or, perhaps, a prophecy for the future of Artificial Intelligence: a neo-Luddite grassroots movement rising against robots and thinking machines, the fourth Industrial revolution coming to an abrupt and unseemly end.

For me, it’s important that the story involves Descartes. He was the one who suggested that humans are, in effect, machines. He loved human-like automata, which were very fashionable back in 17th century Europe – and he may indeed have built a few himself. Automata were a splendid metaphor for his idea of what humans are. Our bodies, according to Descartes, are made up of mechanical parts that tick-tock together, like the gears of a well-tuned clockwork. Since Descartes was not an atheist, he also suggested that one said part, a small gland in the centre of the brain called the pineal gland, was the seat of the soul, the place in which all our thoughts are formed. And thus, by separating the material body from an immaterial soul, Descartes impressed upon our thinking the seal of dualism forevermore. What separates life and death is thinking: I think, therefore I am. I cease to think, and I am no more.

Around a century and a half later, on the question of animating dead flesh, Mary Shelley replaces the mechanical with the electrical. The monster is an assembly of dead bits and pieces that come alive through electrification. But does the monster have a soul? Apply Cartesian logic to the question and the answer must be yes: he thinks, therefore he is. He also loves, and hates too.

Dualism for an atheist age requires that we expunge the soul and replace it with something that “feels” less mysterious: software, for example. Our digital computers are quintessentially Cartesian. They are made of inanimate dead matter, the hardware, the silicon chips, the mesh of wires, the peripherals, etc. They come “alive” thanks to the immaterial “software”, the series of commands and instructions, the pattern of ones and naughts, the algorithm. The computer metaphor is so powerful, and so resplendently dualistic, that we see in computers a reflection of our own image. Many speak of the next stage in human evolution as the fusion of humans and intelligent machines. The only possible endgame in the story of computer technology is, therefore, Artificial Intelligence.

It is a fascinating endeavour – at least to me – to explore the complex interplay between literary narratives and technological evolution. Sometimes a literary metaphor facilitates, and sometimes, impedes, the advent of a specific technology; and often an emerging technology breeds the next literary metaphor.

metropolis_still

Making a robot in someone’s own image

ex-machina

Cinematic parallels: Ex Machine imitating Lang’s Metropolis.

How we view ourselves is perhaps the most profound example of how literary and artistic metaphors modulate and transform through time in conversation with technology. Go back to the dawn of the agricultural revolution, when life was seen to grow from inside the ground, and stories were told that humans were also created of mud and spit. Jump forward into the Hellenistic era where hydraulics was used to construct the first complex machines, and medicine views humans as containers of tubes where liquids (called “humours”) flow. It was an idea so powerful, that persisted well into the early 19th century, when it was gradually replaced by the germ theory of disease and modern medicine. Descartes and mechanical automata, electricity and vitalism, are other examples in the history of literary metaphors for life.

Because, we tend to think through metaphor, it is too easy to confuse the metaphorical with the actual. This is a trap that we find ourselves today. We confuse Artificial Intelligence with Intelligence. Worse, we confuse our own brains for computers. The brain is not a computer, not a digital computer at any rate. It does not separate into hardware and software, flesh and non-flesh. It’s all flesh. My apologies therefore, to those who believe in the Singularity, that glorious moment in the future when they will be able to upload their minds in a computer and live the eternal life. I am sorry folks, but the mind is not separate from the brain, and therefore it is not uploadable or downloadable, or algorithmic, or programmatic, or made up of ones and zeros. I suspect the mind is more complicated than that, and we should wait for neuroscience to progress somewhat before we have a more learned conversation about what the mind truly is.

Let’s now turn the mirror of metaphor round….

Now it is the machine that is looking in the mirror, imitating us. In doing so, machines have started spinning their own stories. Here’s a story that a machine recently came up with:

Dr. Jorge Pérez, an evolutionary biologist from the University of La Paz, and several companions, were exploring the Andes Mountains when they found a small valley, with no other animals or humans. Pérez noticed that the valley had what appeared to be a natural fountain, surrounded by two peaks of rock and silver snow.

Pérez and the others then ventured further into the valley. “By the time we reached the top of one peak, the water looked blue, with some crystals on top,” said Pérez.

Pérez and his friends were astonished to see the unicorn herd. These creatures could be seen from the air without having to move too much to see them – they were so close they could touch their horns.

While examining these bizarre creatures the scientists discovered that the creatures also spoke some fairly regular English. Pérez stated, “We can see, for example, that they have a common ‘language,’ something like a dialect or dialectic.”

Dr. Pérez believes that the unicorns may have originated in Argentina, where the animals were believed to be descendants of a lost race of people who lived there before the arrival of humans in those parts of South America.

While their origins are still unclear, some believe that perhaps the creatures were created when a human and a unicorn met each other in a time before human civilization. According to Pérez, “In South America, such incidents seem to be quite common.”

Thus spoke GPT-2, a “model” (shorthand for an artificial neural network), trained to predict the next word given the previous words in a text. OpenAI, the non-profit organization, that trained the model were both elated and horrified by their achievement. So much so that, despite the “Open” in their name decided to keep GPT-2 firmly “closed” and “restricted”, lest it was released in the digital wild adding more fake stories to the ones already peddled by humans. I am horrified too, but for two completely different reasons. Horror 1: we now have scientific proof that creative writing can be mindless and stochastic, a probabilistic process that has no need for self-indulgent fluff such as “inspiration”, “imagination”, or “consciousness”. Horror 2: I really liked the story. I might even put my money up and buy the whole unicorn discovery book, if GPT-2 ever gets the permission to write it.

I won’t be the only one who’s ready to pay for computer-generated creativity. Last October, Christie’s sold “Portrait of Edmond de Belany”, an algorithm-generated print in the style of 19th century European portraiture for the amount $432,500. The AI art gold rush is here! In New York!

The artificial artists are called GANs, with an A, acronym for Generative Adversarial Networks. It’s interesting how they work. They are made up of two artificial neural networks, working against each other. One network constantly creates Fake images, starting with white noise. Let’s call that network “Donald Trump”. The other network – call her “Democrat Opposition” – takes two inputs: the fake image input from Donald Trump, and a real image from a human trainer. The Democrat Opposition compares those two inputs and calls BS, when it discovers that Trump has fed it with a fake image. But – and here’s the genius of the system – Trump takes the output of the Democrat Opposition’s judgement and uses it to improve the fake image. Do this a few thousand times and Trump ends up creating fake images that the Democrat Opposition cannot tell the difference from the real ones. GANs are the ultimate content machines: journalists, copywriters, graphic artists, photographers, and assorted media and advertising creatives, your days are numbered!

But what about art? Will the museums of the future hang on their walls artwork from the grand AI masters of the 21st century? Well, Marchel Duchamp showed that anything can be art. “Art is what you get away with”, added Andy Warhol. GANs can do urinals, Campbell soup cans, and Renaissance portraits at a blink of an eye, and you won’t be able to tell the difference. Trained, as we humans are, to welcome the new and shameless with gasps of awe and adulation, GAN art passes the Turing test with flying colours. The only problem is that GANs are trained in art that already exists. AI-generated art is trapped in the Past. With all the hype surrounding it AI-generated art may feel novel and exciting today, but will soon get tiresome, repetitive, familiar and uninteresting. You see the AI we currently have at our disposal, impressive as it is, is “narrow” AI. It is only good – very good actually – within a narrow domain determined by specific goals set by humans. Current AIs lacks the will to deny, doubt and challenge, the essential virtues of human creativity. Effectively, we have built useful artificial savants. Those savants, when working together with humans, can act as cognitive and creative multipliers. AI can augment artistic and scientific endeavour, it can be a powerful tool in our ever-expanding technological toolbox for creating, exploring, discovering, and profiting. And that’s quite a wonderful thing we should all look forward to.

But, human and artistic augmentation may not be the end of the story with AI. And since we are all gathered here today united around the question of “what’s next?”, let me speculate about the future of Artificial Intelligence.

Literary engine

Depiction of the Lagado word machine, eerily looking like a neuromorphic chip.

It is foretold in Jonathan Swift’s Gulliver’s Travels. In Book III, Gulliver is abandoned by pirates on the continent of Balnibarbi. After a visit to the flying island of Laputa, Gulliver is taken to the Academy of Lagado, where “useless projects” are undertaken. There, he is given a demonstration of a word machine, which is nothing less than a giant mechanical computer used for making sentences and books. The wise men of the Academy pride themselves for discovering a machine that renders obsolete any study or expertise; for an absolute idiot can now write a masterpiece by virtue of cranking the machine. The GPT-2 is not there yet, and it will never be, given the statistical nature of today’s machine learning systems. There are limits to using statistics in order to predict language sentences without understanding meaning. But what is meaning? And can a machine understand the meaning of things, of words and ideas, of paintings and music? Can a machine be, or become, conscious?

If we assume that there is nothing magical about the human mind, and that it’s a product of natural processes and interactions, then there is no reason to deny that a non-biological, artificial entity may gain consciousness. The real question should be how. To answer the question we need two things: a general theory of intelligence, and some kind of idea of how to build computers that are not digital, or Cartesian, or dualistic, if you prefer.

General theories of intelligence exist and are quite mature. Suffice to say that the most interesting of them looks into how thermodynamics play out in biological systems. A key idea here is entropy, the degree of disorder. Living things, brains too, continuously fight against the tendency of the universe towards chaos. Information and thermodynamics have a very interesting relation, conceptually as well as mathematically, and replacing statistics with physics seems to me a much more promising path towards General Artificial Intelligence.

But what about computers? Well, life and brains are not digital but analogue. A good non-digital, analogue computer architecture goes by the name “neuromorphic”. Essentially, it is hardware that directly emulates how biological neurons function and communicate. There are currently several experiments around the world where teams of scientists and engineers are trying to build General Artificial Intelligence systems. One of the most interesting projects is called Spinnaker, and is run at the University of Manchester, in UK. There, a neuromorphic supercomputer uses a million processing units to emulate the internal workings of up to a billion neurons. The machine is currently simulating the 100 million neurons inside a mouse’s brain. The human brain, with 100 billion neurons, is next…

So within the next 5 years expect some very interesting publications that go beyond the statistical and the digital, and into the physical and the neuromorphic. General Artificial Intelligence is theoretically feasible and technologically, not too far away. When it is finally delivered we will have created a new intelligent being, a new “I” in our social and cognitive universe.

Will we have delivered a “useless project” then, much like the fictional academics of Lagado? Will that historical moment be one of triumph, of tribulation, of both? One thing’s for certain: that the ethical, legal and social challenges that will derive from such a profound act of creation will be nothing less than Promethean.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s