Re: Uploaded memories

Anders Sandberg (
08 Dec 1999 15:42:21 +0100

"Robert J. Bradbury" <> writes:

> Seriously (can I do that here?), though I can't remember the paper
> reference (I do have it, but I have so many papers...) research at
> Bell Labs seems to indicate that your computer *could* hold a
> large fraction of your memories. The experiments they have done
> seem to indicate you store only a few bits a second.

This is likely your *conscious* bandwith. Unconscious learning can be orders of magnitudes more powerful, increasing storage demands a lot. Still, the total memory capacity of the brain seems to be within reach within a few decades at most.

> As I've mentioned in another note, the question remains open
> whether recall capacity and recognition capacity are the same.
> As someone, Hal perhaps?, pointed out, there may be a fair
> amount of compression occuring. I don't have to remember
> walking to school everyday, I only have to remember odd
> events that have occured on a few of the days that I
> walked to school on top of a general pattern of walking
> to school.

Yes, there is plenty of compression, abstraction and plain loss in memory. Recall capacity in general seems to be smaller than recognition capacity (because the later doesn't have to reconstruct a whole scene, just check if the network has a reasonable attractor corresponding to a memory).

> I don't think this is proven. More likely, as I think Anders has
> pointed out, you may get loss of memory because you don't exercise
> the synaptic connections and they gradually lose strength relative
> to other synaptic connections. You may occasionally "recover" a
> memory if you happen to trigger a global pattern that pieces
> together a memory from widely separated subfragments across the brain.

True. But most forgetting seems to be overwriting or interference between overlapping memories, or the gradual distortion of memories as you remember more and more your reconstructions of them and less and less of the original situation.

The problem is the time constants of learning: too short, and you learn and forget quickly, too long and you cannot learn quickly and run the risk of catastrophic forgetting if you overload the whole system (not observed in any brains so far, though). Another reason we already seem to have different memory systems with different learning rates.

Anders Sandberg                                      Towards Ascension!                  
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y