My son saw a shark today.
In his two short years, this was the first time he beheld the creature. His eyes went wide. Would he fear the thing?
His head tilted a few degrees left, betraying a smile. Not knowing what it was or its name, he christened it himself. Pointing, he proclaimed, “Agu dalpha!”—his toddler parlance for “alligator dolphin.”
As a linguist and translator, I was charmed. What a peculiar name. What a sensible name.
He saw the creation, and he named it, tapping into that ancient stream of humanity flowing back to our first ancestors.
In mere fractions of a second, his mind sifted his entire knowledge base. Visual, motor, linguistic, and associative information danced across the trillions of synapses in his brain, pivoted in his frontal cortex, then flew back through the Wernicke’s area, arcuate fasciculus, and Broca’s area and was ultimately heralded out by the motor cortex controlling his speech articulations.
These so-called “basic” mental processes are extremely complicated. Legions of academics and corporations toil round-the-clock to teach computers to approximate just one of these steps. The who’s who of star tech companies together spend billions to crack even a sliver of the great problem of true intelligence.
But my toddler, unbidden and unsupervised, does this entirely of his own agency. He is an analog masterpiece. And he is not unique in this. In his young mind, unsupervised learning meets unsupervised creation. These are hallmarks of human intelligence.
Currently, we are the only created beings in the universe that can do these things—at least that we know of. Computers cannot perform unsupervised learning like we do. Even their supervised learning is shaky.
But every day, artificial intelligence gets closer. Will they ever match our curiosity? Our creativity?
Though we take it for granted, the method humans employ for learning is quite remarkable. Our learning is unstructured, unsupervised, creative, intuitive, and dripping with common sense. For example, most of what we learn in our lives is untaught. Ninety-five percent of our vocabulary is plucked out of the ether. Nobody taught it to us, we simply learned it along the way. We were encouraged to walk, to catch a ball, to make friends. We were given opportunities to try and fail, and then try again, but we were never strictly taught these things. The mind can never be taught howto catch a ball. We can learn where to place our hands or how to catch with better form, but there is simply no equation. Imagine explaining the midair kinematics to a toddler. The mind’s hidden algorithms must simply pick it up through repetition and minimizing prediction error. A robot must have the numbers.
We are intuition machines, remarkably accurate in wild guesses with very little evidence behind our intuitions.
The way a computer learns is, shall we say, less poetic. In supervised learning, a computer is told, “This is a shark” and then shown millions of pictures of sharks. In the space of a few seconds, a computer analyzes more sharks than we could ever lay eyes on in a lifetime. The computer is then asked to identify sharks in unlabeled pictures and corrected by a human or a human-written program when wrong. And still they’re quite awful at it.
But by playing to their strengths, they’re improving. Through something called generative adversarial networks, computer programs are made to compete against each other in a sort of war games scenario. They each attempt to fool the other, sometimes millions and billions of times, and then they sharpen each other based on the results—all while their programmers sleep in their beds. This can be done with any set of data. Facial recognition, biometrics and DNA, or pictures of sharks.
Click here to read more.
SOURCE: Christianity Today, Jordan Monson