[ExI] after upload, what?

[Travis Porco]

> From: Adrian Tymes <atymes at gmail.com>
>
> > Forgive me for delurking,
>
>
> No forgiveness needed.  Welcome to the chat!
>
>
> > but perhaps the time has come to ask *how*
> > to do an upload or upload-like activity right now, today, with
> > existing technology...extracting enough memory and identity to
> > generate an authorized continuing agent for yourself, without being
> > distracted by worries about whether it is "really you" and so forth.
> > As I follow these discussions I find the philosophical worries merely
> > lead to inaction, and in some cases, contain elements dismissive of
> > the value of lives.
> >

> Well...there is one option with today's technology, but it'd be heck of
> impractically expensive for a human brain, though I hear something like
> this might be being done with insect brains.

Yes--I hear they have mapped out the "connectome" of an insect brain
(Winding et al, Science, March 2023),
but apparently they don't know whether the connections are inhibitory
or excitatory; it's just structural so far.

> Map out and simulate each neuron, one by one.  Make a good enough (which
> likely requires near-perfect in practice) software simulation of the
> neuron, including its synapses (its inputs and outputs).  Connect these
> neuron simulations together, in the exact same fashion as the actual
> neurons are connected.  Run the whole thing on a powerful enough computer -
> which involves massive parallelism, even for insect simulations, thus the
> expense.  There also needs to be blank/spare/extra hardware to simulate
> growth of new neurons over time (not as many as already exist in the
> running brain, but a nonzero amount, with a potentially indefinite cap if
> the simulated brain lives forever; more hardware can be added over time to
> support this, but this might put a cap on the practical maximum speed-up if
> the uploaded brain is to run significantly faster than the original).

> This misses input from the rest of the body, but it's a start (and can help
> narrow down exactly what the rest of the body's inputs are).

This is what I'd call a "maximalist" approach, aiming to do it right.
I'm more looking for
"good enough, right now". I'd like to see as much of the 'narrative
self' preserved, in any format, using
any substrate, with a hope of further growth or regrowth later. What
are the limits on the "minimalist" side?

One could use an authorized biography, images, notes, etc.--as much
information as you can download or
extract from the brain using the "built-in" I/O if you will. (I know
not everyone loves brain-as-computer metaphors). Use
a fine-tuned LLM or a collection of them, using some "agent"
framework. There has been some discussion out there of these
so-called "digital ghosts" or "memorial chatbots", usually derogatory,
but I think the discussion has been shallow and dismissive. But
certainly such efforts could hardly be called "uploads".

What you would get with March 2024 technology would be a far
cry from ideal...but I'm interested not in whether it is a desirable
final endpoint. Like the cryonicists say, it only has to beat
the alternative: guaranteed oblivion!

Long term one would need more capable agents and models--today's are
too shallow and incapable of genuine autonomy. Longer
term still, one needs a better hardware. I do not for a moment think
that the sort of chat-agent we could make today is
conscious, and so the chat-agent needs to have this as a goal. I
suspect that the insect connectome work will lead someone in
the right direction.