If we make a brain by simulating neurons then making it superhuman is a matter of throwing hardware at it, and it could certainly help design better hardware. In terms of subjective time no it would not be a lot faster than a human but that doesn’t matter to the outside view.
I actually think though that apart from just more cycles the patch-work simulation brain would enjoy immense advantages in terms of access to information and computing facilities. Similar to the advantages we would anticipate from “wiring in” an organic brain but even better since it can benefit from much higher bandwidth.
If we make a brain by simulating neurons then making it superhuman is a matter of throwing hardware at it,
Not really. No matter what you do, if your simulation speed is optimized to the best of your current know-how, expanding additional hardware once you reach full simulation will only be wasted resources. And then you run into lightspeed constraints and bus/bandwidth limitations.
I.e.; if we don’t have the hardware you can’t “just throw more hardware at it”.
and it could certainly help design better hardware.
There is an underlying assumption here that a simulated brain would operate faster than a human’s. That’s by no means necessarily true. Anyone familiar with virtualization performance knows you take a performance hit as compared to bare-metal when you do full virtualization.
The speed at which it operates will have nothing to do with the speed at which human brains operate. It can be much much slower and much much faster, depending on the hardware available. Today, it would be much much slower. Yes there are theoretical limits but also neural simulation looks like a problem well optimized for parallel processing, just like our brains do. The extent to which that is practical is really the governor here since parallel scalability would really mean that we can throw more hardware at it until the cost of communicating between nodes exceeds the benefit from having one more node.
The extent to which that is practical is really the governor here since parallel scalability would really mean that we can throw more hardware at it until the cost of communicating between nodes exceeds the benefit from having one more node.
I don’t think “nodes” is an entirely accurate statement here, however. But either way—the point I was making is that the hardware is itself going to be the biggest constraint in early-period devices. Furthermore, it’s not enough to simply integrate more nodes; those nodes need to have an actual role. Otherwise, bigger brains would result in smarter individuals. Yet elephants are not smarter than humans.
If we make a brain by simulating neurons then making it superhuman is a matter of throwing hardware at it, and it could certainly help design better hardware. In terms of subjective time no it would not be a lot faster than a human but that doesn’t matter to the outside view.
I actually think though that apart from just more cycles the patch-work simulation brain would enjoy immense advantages in terms of access to information and computing facilities. Similar to the advantages we would anticipate from “wiring in” an organic brain but even better since it can benefit from much higher bandwidth.
Not really. No matter what you do, if your simulation speed is optimized to the best of your current know-how, expanding additional hardware once you reach full simulation will only be wasted resources. And then you run into lightspeed constraints and bus/bandwidth limitations.
I.e.; if we don’t have the hardware you can’t “just throw more hardware at it”.
There is an underlying assumption here that a simulated brain would operate faster than a human’s. That’s by no means necessarily true. Anyone familiar with virtualization performance knows you take a performance hit as compared to bare-metal when you do full virtualization.
The speed at which it operates will have nothing to do with the speed at which human brains operate. It can be much much slower and much much faster, depending on the hardware available. Today, it would be much much slower. Yes there are theoretical limits but also neural simulation looks like a problem well optimized for parallel processing, just like our brains do. The extent to which that is practical is really the governor here since parallel scalability would really mean that we can throw more hardware at it until the cost of communicating between nodes exceeds the benefit from having one more node.
I don’t think “nodes” is an entirely accurate statement here, however. But either way—the point I was making is that the hardware is itself going to be the biggest constraint in early-period devices. Furthermore, it’s not enough to simply integrate more nodes; those nodes need to have an actual role. Otherwise, bigger brains would result in smarter individuals. Yet elephants are not smarter than humans.