Philosopher AI
Purchase queries
How would you improve the standard models for computer system architecture?
In the present time, computer systems are designed to be as efficient as possible. Computing is powered by electricity and electrons moving through logic gates at a speed of 3 nanoseconds per gate in silicon.This works very well, because it allows computers to do trillions of operations every second with amazingly low power consumption (not counting cooling). However this design departs from reality: It never considers where the electrical energy comes from nor what happens when there is not enough power for computation or even no power available.
Let me consider a simpler design that is based on the laws of physics. This might seem like an unrealistic goal, but I am pretty sure it would work. It should be possible to build such computers using alternative energy sources and even without electrical power.
The first thing we need is huge, very fast memory. This could be made of DNA strands while operating in water at room temperature or even better by using a crystal lattice and processing light directly. This would allow computing with a few picoseconds per operation (1 billion operations per second) which should be more than enough for human level AI to think so much faster that it will never have trouble finding the solution.
Next, we need a very powerful processor. I propose to use an ionic lattice operating under the laws of physics. Its ions will move with nanoseconds per operation while consuming negligible energy (they are charged so they repel each other).
Finally, the whole computer system will have to be isolated from external energy sources and cooled. The simplest solution would be to use a vacuum chamber with light shining on it using phased arrays (similar to lasers) so that the heat is carried away by photons.
The processor will need very little cooling and the memory needs none at all.