A fingernail, how much area is that? A square centimeter, maybe an inch and a half? Regardless, the millimeter is not important here. Imagine that billions of tiny dots can be accommodated on an area of this size and that they are also connected with conductor tracks. Their length adds up to kilometers. Now you have a first idea of the super-tiny dimensions in which what most of us simply take for granted: computer chips.
A chip like the one in the laptop on which this column is written is not just a bunch of dots, i.e. transistors. It consists of several intricate units that can be applied to the silicon substrate in this way or in a completely different way. Even planning what goes where takes several weeks. Asked the so-called Floor planning are therefore chip developers with many years of experience and a creative spirit.
But now they could at least get help. Scientists at Google have trained artificial intelligence (AI) to address the problem. According to her, she can now at least as well, but often better than humans, at arranging the various elements and connecting them to one another in a space and energy-saving way.
Ten thousand chip designs
The training was somewhat tedious, after all, the AI was presented with around ten thousand chip designs, each of which had to be given a rating beforehand. The task of the AI was then to create designs with the best possible ratings. It is not surprising that an AI performs well here, after all, there are almost innumerable ways in which the conductor paths can run – to name just one example.
Anyone who has ever dealt with futurists like Ray Kurzweil and others may now hear the bell: Isn’t that exactly the development that people like Kurzweil have always predicted: that machines will eventually start to build ever better machines? Sounds plausible, because if the chips get faster, better and more energy-efficient, the AI will also benefit. And it can do even better chips again …
Because machines do this at a different pace than we human cramps, the evolution of technology (which is already much faster than the natural one) gets into a kind of turbo mode. Oh, what does turbo mean here, we’re talking about exponential growth, and the assumption would be that the point has now been reached at which the curve begins to rise steeply and increasingly steeply. Among other things, Boris Johnson’s insane experiment will probably provide a vivid example of this when the corona infection numbers in Great Britain skyrocket.
Physical limits
But there are also some arguments against the fact that chips are constantly developing unchecked. Previous designs are being improved bit by bit on the basis of proven models. If an AI internalizes this, really creative approaches could possibly be blocked. In addition, AI will also have to do with the physical limits with which technology is already struggling. At some point it just can’t get any smaller when the boundaries between transistors are barely larger than a single atom. Or is an AI really developing a completely new, creative idea? Rather unlikely.
Because that’s exactly what AI is missing: creative out-of-the-box thinking is not its forte. Fled through enormous mountains of data in a very short time, yes. Recognize connections that a person would not have seen, yes. Completing a symphony if it has only been given enough examples of a musician’s way of composing beforehand, including that. But it cannot really invent something new, as it is the essence of art, of human beings in general. Which is with some probability because AI comes to its results with completely different methods than humans with their 20-watt bio-calculator.
But they cannot be ignored. Fundamental rejection is just as harmful as technocratic hurray patriotism. After all, AI is also just technology, just a tool that can be used for good as well as for bad purposes.