AI Revolution 101
Pawel Sysiak
1.2K82

Nice article, but it contains some statements and assumptions that raise questions.

  1. At some point you rely on Moore’s law. As others have already commented the last few years it’s been showing signs of breaking down.
  2. Some of the properties of electronic circuits that you consider a guarantee for better performance may actually hinder better performance, .e.g.
  • Speed. “The brain’s neurons max out at around 200 Hz, while today’s microprocessors … run at 2 GHz, or 10 million times faster.”⁵¹

With increased speed comes increased energy usage (to reach higher speeds in electronic circuits you also need higher voltages, and energy consumption squares with voltage). Energy efficiency and associated heat production are a huge challenge to overcome. With Moore’s law flattening out current technology may be coming to a dead end in that respect. The current evolution is towards more parallelism (more cores). With enough parallelism the speed can be brought down, and therefore the voltage can be reduced and hence energy consumption can be reduced.

  • Memory. Forgetting or confusing things is much harder in an artificial world. Computers can memorize more things in one second than a human can in ten years. A computer’s memory is also more precise and has a much greater storage capacity.

Given the rate with which new things are happening, remembering “everything” may also make it impossible to analyze the exponential number of connections between “everything” (which presumably is needed to make sense of it all). Next to kSelective forgetting may also play an important role in keeping evolution going (e.g. because it stimulates frequent retrying of things that have failed before, which on a next iteration may turn out better because of changed environmental circumstances e.g.).

  • Performance. “Computer transistors are more accurate than biological neurons, and they’re less likely to deteriorate (and can be repaired or replaced if they do). Human brains also get fatigued easily, while computers can run nonstop, at peak performance, 24/7.”⁵²

The deterioration and less than perfect performance of biological neurons could e.g. play a central role in such phenomena as creativity. When you introduce evolution in computer algorithms you essentially turn to random number generators to emulate some of the imperfections that biology provides “for free”.

Despite the raised eyebrows, it will be interesting to keep an eye on the ongoing evolution in the field.

Show your support

Clapping shows how much you appreciated ste’s story.