Essay by Eric Worrall

The real reason Silicon Valley silent quit on the Democrats, and quietly backed President Trump’s America first energy policies.

Let’s dive straight in. What you are seeing below is (hopefully) a real neural network running in your web browser. The neural network wants to learn how to drive.

An explanation. Modern artificial neural networks usually start by generating a set of random neural networks.

As you can imagine, a randomly generated neural network is unlikely to be a good solution to the problem. But by chance, some of the randomly generated networks will be slightly less awful than their siblings.

Those slightly less awful neural networks are used as a guide to creating the next generation population of neural networks. The actual method of extracting the goodness from the previous generation differs, but it can involve backpropagation (mathematical optimisation for expected results), genetic algorithms (“breeding” the best networks like you would breed cattle, in the hope the child networks will inherit the best from both parents), and mutation – photocopying a parent neural network but randomly changing a few parameters to see if those random changes result in a better neural network.

Eventually if you continue long enough, some real skill should begin to emerge.

Evaluating the performance of neural networks is itself a hot topic of research. The following article discusses some of the implications of different approaches to evaluating performance.

Why do neural networks require so much power?

The reason is the software artificial neural networks execute is trash. AI researchers are nowhere near unravelling all the shortcuts and hacks our human brains use to discover solutions.

But this lack of quality can be compensated by quantity – using acres of computers burning hundreds of megawatts of power, to do what our human brains can do with less power than it takes to run your TV.

If you press “high powered computer” on the demo above, you can see a glimpse of how lots of computing power can help. Instead of testing one neural net at a time, in “high powered computer” mode, the neural network above evaluates the driving ability of all 10 neural networks in a single generation simultaneously.

In a similar way, evaluating millions of neural networks containing millions of neurones simultaneously on gigantic data centre computers can compensate for the poor quality of the primitive software which drives the individual neural networks.

What is the point of spending all this cash and pouring all these resources into neural networks?

Have you ever tried playing the chess app on your computer? Unless you are a very good chess player, that simple chess app will kick your butt every time, because the chess AI always knows the right move.

My belief is what that chess app does to you when you play is what AI research could do to the entire planet. Advanced neural networks will help tech giants to always make the right move. They will always know exactly what to put on that teleprompter to persuade their audience.

Tech giants may also be spending big on medical life extension, Google hiring Ray Kurzweil in 2013 to a senior position raised eyebrows in the industry. In addition to being a tech genius, Kurzweil is one of the world’s leading proponents of using AI technology to research life extension and medical immortality.

All that a man hath he will give for his life.

Tech giant top executives want to win, and keep on winning forever.

But to win, tech giants have to ditch green energy. To compete with Asia, which is also going big on AI technology, they need rock solid reliable energy supplies at a price comparable to what Asian tech giants pay. Which is why tech giants now want their own in-house nuclear reactors.

I like to think I’m making a difference writing for WUWT, but when the dust settles it won’t be anything I ever wrote or said which kills the green energy movement.

The people who deliver the final death blow to the green energy movement will be those who were once its strongest proponents.

But there is a dark lining to this silver cloud. If tech giants succeed, if they gain the ability to always know the right move to advance their goals, personal, political, social, and of course financial, perhaps in the future we shall come to miss the good old days when all of our opponents were human beings like ourselves.