On Mon, Jul 29, 2019 at 7:10 AM Bruce Kellett <[email protected]> wrote:
> On Mon, Jul 29, 2019 at 8:49 PM Jason Resch <[email protected]> wrote: > >> On Fri, Jul 26, 2019 at 8:09 PM Bruce Kellett <[email protected]> >> wrote: >> >>> On Sat, Jul 27, 2019 at 7:10 AM Jason Resch <[email protected]> >>> wrote: >>> >>>> On Fri, Jul 26, 2019 at 12:44 AM 'Brent Meeker' via Everything List < >>>> [email protected]> wrote: >>>> >>>>> >>>>> On Fri, Jul 26, 2019 at 2:42 PM Jason Resch <[email protected]> >>>>> wrote: >>>>> >>>>> For example, you could, over time, change neuron by neuron, until you >>>>> looked like and had the mind of Julius Caesar. >>>>> >>>>> I think these thought experiments need to be more carefully >>>>> considered. I don't think it is nomologically possible to give you the >>>>> mind of Julius Caesar by transferring on neuron at a time. That would >>>>> entail intermediate stages in which neurons were connected neither as >>>>> yours >>>>> were nor as Caesar's were, and less obviously the same goes for the >>>>> connections of the body cells. It is too cheap to just say "at the >>>>> appropriate substitution level". >>>>> >>>> >>>> It's a given that the intermediate stages are neither like you nor like >>>> Caesar. But if you can remain conscious/alive during the process, then >>>> "what happened to you", "did you die in the transformation?", etc. >>>> >>> >>> That's rather a big "if". It seems to me that one important difference >>> between your mind and that of Julius Caesar is the connections between >>> neurons. Just replacing one neutron at a time is not going to create/change >>> the necessary connections. Besides, do you have any evidence that any two >>> minds have exactly the same number of neurons? Piecewise replacement of >>> neurons will almost certainly destroy consciousness, even life -- the >>> intermediate stages will not correspond to any conscious or living person. >>> >> >> The neuron replacement includes creating the appropriate connections (as >> well as adding or subtracting neurons if necessary). >> > > That is not a well-defined procedure -- too many ambiguities remain. > > What is ambiguous? One physical state "your brain" is adjusted gradually until it is equal to the brain of someone else. You agree this is physically possible, right? Do you think the path taken in this gradual adjustment is relevant to the thought experiment? If so can you explain two cases that show different consequences resulting from how the transition is performed? > Over time you can slowly morph one person's brain into another, if the >> brain is a just a physical object, then physical objects can be changed, >> sometimes radically. We may lack the technology now, but this is already >> possible today in software implementations of neural nets. >> > > That does not demonstrate that the intermediate stages correspond to > anything sensible. > I don't see that it is necessary. Though I think we have strong evidence that it can be done. Given brain bisection cases (you can remain conscious with only half a functioning brain), it seems likely that your consciousness can survive much smaller and less destructive changes to one's brain organization. > > >> Nothing prevents making adjustments to such a software neural net one >> connection at a time, until one "mind" becomes a totally different "mind". >> In fact alphago began with an entirely randomized neuron net, which though >> training was adjusted one neuronal connection weight at a time. >> > > Are you sure that it was just one connection weight at a time? > > In any computer, operations are serialized such that modifications to something in memory (e.g. AlphaGo's neural net) are performed incrementally (one bit or one register at a time). However in practice this doesn't indicate anything significant. To save time in training it is almost certain that multiple neuron weights were adjusted before running the neural network again. From: https://deepmind.com/documents/119/agz_unformatted_nature.pdf "First, the neural network is initialised to random weights θ_0" "Specifically, the parameters θ are adjusted by gradient descent on a loss function l that sums over mean-squared error and cross-entropy losses respectively," Jason -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/CA%2BBCJUj84tzJx4S88j4GXagUbJpASkG8tZ4UxSaeqQ1bWjopcA%40mail.gmail.com.
