Uploads: simulations & ethics [was AI motivations & Self-rewiring]

Robert J. Bradbury (bradbury@www.aeiveos.com)
Sun, 31 Oct 1999 09:34:54 -0800 (PST)

On Sun, 31 Oct 1999, Robert Owen wrote:

> phil osborn wrote:
>
> >
> > What no one here seems to understand is that the "mind" is much more
> > than the wiring. I'm talking about hormones, thousands of different
> > neurotransmitter and modifier substances, all of which are released both
> > generally and in specific areas of the brain and are essential to mental
> > focus, motivation and action. The mind is not a logic engine; it is part
> > of a living system.

True. The question comes down to whether the living system is deterministic, random or chaotic and the "information content" of the system. I'm fairly partial to the determistic camp, but would be willing to tilt my hat that some of what goes on in the mind can be attributed to either random or chaotic effects (at a low level).

Now as for the "information content", theors for this have been extensively developed based on the work of Shannon and others. Whether you are encoding the information in neural impulses (in the time (frequency) domain or in the amplitude domain, or both), or in the quantity and effectiveness of neurotransmittors diffusing across synaptic junctions or hormones being received from the blood -- *information* is *information*.

>
> But Robert's claim that neuron replacement is feasible I find reasonable.

Moravec & Minsky are the authors of these ideas, I'm simply observing that I find them consistent with my knowledge of physics, chemistry and biology.

> So do you regard the synthesis of neurotransmitter chemistry impossible?
If you mean the "manufacture" of neurotransmitters, then clearly it is possible because our brains do it. But on an alternate computing platform you probably wouldn't want to do that.

> The cybernetic aspect of e.g. signal control requires the operation
> of antagonistic agents -- is it possible to simulate the control
> mechanisms which regulate the release and re-uptake of, say, serotonin
> by monoamine oxidase chemistry? Or do you think these feedback
> processes are so complex that we simply cannot artificially replicate
> them?

All you have to deal with is the information transfer problem. You do have to know fully and completely the information that is being transfered. Then you simply encode it in the most effective way on the simulation machine.

I could envision 3 different levels of uploads: (a) A molecular level where you run the simulation on a (very)

      big cellular automata that is simulating precisely
      the quantum interaction between *all* of the atoms in the brain.
  (b) A biochemical level where you simulate the activities of all
      of the genes & proteins in the cell.  Genes being turned on, off
      proteins catalyzing reactions with various molecular concentrations,
      contractions controlled by statstical diffusion or active transport,
      etc.
  (c) An information flow level where you simulate simply the information
      being transfered.  I can reduce the information to packets --
      a neural net information packet with frequency & amplitude
      information, a hormone packet providing concentration & state
      information, neurotransmitter packets that are triggered
      by the neural net information packets, reuptake packets
      managed by cell state or net weighting state information).

I suspect that you will lose a little of a typical mind as you move up through each of these levels, but you probably run faster and faster because less processing power is required. Given equivalent levels of computing capacity (a) probably runs much slower than real-time, (b) in approximately real time (+/- a few orders of magnitude) and (c) probably faster than real-time. If it turns out there are random or chaotic elements that would be accounted for in (a) but not (b) or (c), then you need to add these to the simulation in some way.

You still need the inputs (or the simulation equivalents of those inputs) that the body is normally going to be giving the brain (O2, glucose, heart rate, temperature sense, sight, smell, taste, touch, etc.)

You only get things that are potentially really different from the uploaded mind if you give yourself the ability to edit the source code. Don't like the fact that Nitric oxide makes men passive and women aggressive -- well delete that molecule from (a) or those genes & enzymes from (b) or NO packets from (c) and you *certainly* have a different mind. I would envision that changes within the realm of human variation (of which there is quite a bit) or those we have explored in laboratory animals (as is the case with deleting the NO enzymes) will be fairly acceptable/explored. When you say, lets delete all the packets involved in mental processes when you are asleep and end up with a memory-less psychotic then you would be well advised to have a backup copy on hand.

Interesting morality problem... Should you have the right to kill (due to poor reprogramming) yourself (that really isn't yourself once you have forked a copy) again and again and again.... "Mind ethics" is going to be a bear as we have pointed out previously.

I wonder if you can solve the problem by saying "forking" isn't allowed. You can make a change but you have to save all of the subsequent state information in such a way that you can "undo" (editor terminology) or rollback (database terminology) the changes. If you decide you don't like how things turn out, you rollback to your original state (presumably with a message from your former self explaining why it didn't turn out very well). You have a limited amount of material & energy in which to keep your rollback information (and the possible rollback information of everything/one you interact with), so you can only go forward along a path for just so far before you have to make a decision whether or not to "commit" to that reality. Presumably you have auto-rollback "on" unless a conscious commit occurs, preventing you from staying locked in a psychotic state.

Not as much flexibility, but much simpler from an ethical standpoint.

Robert