Re: Paths to Uploading

Anders Sandberg (asa@nada.kth.se)
05 Jan 1999 16:29:44 +0100

"Billy Brown" <bbrown@conemsco.com> writes:

> Anders Sandberg wrote:
> > I don't think you can get uploading without fine-grained models. How
> > do you make a higher-level model of a mind without scanning it on a
> > low level? Remember that the only thing we know are possible to scan
> > are the actual physical processes going on in the head, not the
> > information processes. And these scanable processes are low-level
> > stuff like neural firing, neuron types, connectivity and so on, highly
> > variable and individual.
>
> Yes, but you don't need to be able to sim the entire brain at this level.
> You can build an abstract, generalized neuron simulator without ever
> simulating more than a small cluster of neurons at the molecular level. Set
> up one instance of the simulator for every neuron in the brain, and you've
> got an upload.

HAH! Not so fast!

Neurons are complicated, and have a lot of individuality. First, they come in many different types, each having quite different properties; both shape and size, but also electrochemical properties, receptor types, ion channels and behavior that interact in a very nonlinear way. Then each neuron may have tricky internal states, both in the synapses (concentrations of calcium and at present little understood proteins and other chemicals that affect their strength) and inside the cell body. Add to this the individual variations in geometry and perhaps chemical properties, and you get a very confucing picture.

This is what you need to simulate to make a somewhat realistic neuron. Note that we are not talking about actual molecular scale simulations, just things like calcium concentrations in dendrite <1 2 1 3> or the membrane potential at the soma. So to do an upload we will need to set up an instance of this kind of simulation for each neuron. It is not that impossible really, except that we lack the necessary data for a truly complete neuron model (membrane conductivity is reasonably well understood, synapse behavior a hot topic, plasticity controversial and the more subtle or complex interactions between activity, modulation, geometry and genetic switches are completely unknown), and that we have too little computing power yet (both the computers and simulators need to get better). But it is in principle doable, and I hope to be there when it happens.

Can we simplify neurons? Probably; this is just how nature made them, and I think it is possible to create more computationally efficient input-output mappings retaining the same properties. This is an interesting problem, both from a research, a theoretical and an uploading standpoint. What can we do away with without losing the important? How much detail is needed? Could you upload yourself into an integrate-and-fire model? Could this process be continued on higher scales?

<preening my computational neuroscientist feathers>

> Of course, you still need to scan at the molecular level to get the data to
> initialize those simulators. The advantage is that you reduce the
> computational burden by a gigantic factor. The necessary knowledge of
> neurobiology should already exist by the time we can run even this
> simplified sim, so this approach should be feasible long before we could run
> the brute-force sim.

The technology for doing uploads will likely at first be developed just for getting the data for advanced neuron and brain simulations.

What we need now is better data, better models and better computers. 

-- 
-----------------------------------------------------------------------
Anders Sandberg                                      Towards Ascension!
asa@nada.kth.se                            http://www.nada.kth.se/~asa/
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y