I just stumbled on this old post of mine from 2008 where I predicted that a supercomputer would be faster than a human brain by 2012.

This was based on Hans Moravec’s suggestion that the human brain has a processing capacity of 10 quadrillion instructions per second (10 PFLOPS).

At the time I said:

In comparison, it was announced today that the fastest supercomputer in the world, called Roadrunner and devised and built by engineers and scientists at I.B.M. and Los Alamos National Laboratory, is capable of handling 1.026 quadrillion calculations per second (1.026 PFLOPS).

As of 2012, the world’s fastest supercomputer was the “Titan,” a Cray XK7 system installed at the U.S. Department of Energy’s (DOE) Oak Ridge National Laboratory.

Titan achieved a performance level of 17.59 petaFLOPS (quadrillions of calculations per second). So I was right – it was almost twice as fast as the estimate of the human brain.

But compare that to the fastest supercomputer in the world right now which is the Frontier system out of Oak Ridge National Laboratory (ORNL) which can achieve 1.194 Eflop/s (Quintillions of FLOPS).

Both terms, PFLOPS and Eflop/s, refer to a unit of computing performance. The acronym FLOPS stands for “FLoating point Operations Per Second,” which is a measure of a computer’s performance, especially in fields of scientific calculations that make heavy use of floating-point calculations.

“P” in PFLOPS stands for peta, which is 10^15, and “E” in Eflop/s stands for exa, which is 10^18. Therefore, 1 PFLOPS equals 10^15 FLOPS, and 1 Eflop/s equals 10^18 FLOPS.

So, if we translate these units:

  • 10 PFLOPS = 10 * 10^15 FLOPS = 10^16 FLOPS
  • 1.194 Eflop/s = 1.194 * 10^18 FLOPS

Therefore, 1.194 Eflop/s is significantly larger than 10 PFLOPS, more precisely it is 1.194*10^2 or about 119.4 times faster than the human brain.

Of course, we’re talking about supercomputers here, but today a single Nvidia GeForce RTX 4090 chip (retails for about AUD$3000) can achieve a performance of 69.7 teraflops (TFLOPS), which makes the human brain about 143 times faster than a single 4090 – in terms of pure processing speed. But string tens of thousands 4090’s together, and you get ChatGPT.

I went on in my old post to wonder why there wasn’t more talk about AI in the mainstream media and by world governments. Then I said

It reminds me of a chat I had with Australian SF author Damien Broderick over dinner about ten years ago. I asked him when he thought these subjects would be discussed by the general populace. He replied “when it’s way too late to do anything about it”.

And look at us now, running around like chickens with our head chopped off trying to work out how to regulate AI. Don’t say I didn’t warn you.