Important Knowledge

Listening to great thinkers can educate us not only with knowledge but visions of the future. For much of "the future" is already here; it's just unevenly distributed. Many people don't know things yet despite their existence. 

Tom Wolfe paid close attention to everything. His careful observations gird and illuminate his writing. During the last two decades of Wolfe's life (2000-2018) he became convinced that neuroscience (study of the brain) would soon revolutionize the world.

Wolfe was right, but in an unexpected way.

Artificial intelligence (AI) will change life dramatically in the next decade. We've all heard this prediction but few of us grasp what AI is, where it came from and what it's going to do.

An excellent primer on these subjects was just delivered by Prof. Geoffrey Hinton, an early AI researcher who was recently awarded the Nobel Prize in Physics. Prof. Hinton's explanation appears on Jon Stewart's weekly podcast (available on YouTube, etc.).

AI grew out of research Prof. Hinton and others commenced in the 1970s. They were studying how the human brain works and wondering if they could get a computer to mimic it. Obviously, at the time, computing power was rudimentary. Hinton, et al. developed valuable insights into the process of learning. They took lessons from human biology and abstracted them into potential mechanics. Essentially they discovered how a brain learns things and then applied that process to machines. Their discovery radically breaks from traditional computer science whose model simply instructs machines with rules to follow. No traditional computer, using the latter approach, can ever LEARN; it can merely compute.

Prof. Hinton posited in the 1980s that a machine could learn if taught how to learn and then set free to do its magic. The model -- a human brain -- reveals the learning process. Using a "neural network" the brain (or AI computer) contains trillions of individual switches that can be turned on or off. The probability of switches being turned on can be increased or decreased by what other nearby switches are doing. Once enough switches turn on, simple knowledge is obtained. As more switches are turned on, deeper knowledge is reached. But how can computers learn to do that?

By working backward. By taking actual known knowledge and reverse engineering a machine's neural network with better probabilities. Hinton realized -- in a eureka moment -- how the process of learning can be replicated in an inanimate machine.

This revolutionary discovery was knee-capped in 1986 by two things: inadequate computing power and insufficient material (known knowledge) to feed machines. Since then, of course, computing speed has increased exponentially. And content (material used to feed computers) was solved by development of the Internet.

Now Hinton's discovery, foretold by Tom Wolfe, is possible. And the race is on. Billions of dollars are being spent by dozens of large corporations to build AI machines. They differ mostly in what the computers are fed, not the computers themselves, since everyone can use the same advanced machines. AI programs differ in what their input is and how they consume it.

So that's where we are. Any questions?