[ExI] Critical take on The Age of Em

John Clark johnkclark at gmail.com
Sat Jul 2 03:52:12 UTC 2016


On Fri, Jul 1, 2016 at 10:20 PM, Samantha Atkins <sjatkins at gmail.com> wrote:

​>> ​
>> we do know from the size of the genome so that mass of those other
>> implementation details can't be significantly larger than what software
>> engineers are already accustomed to. Of course because the code was nor
>> written by a human being it could still be hard for them to figure out why
>> it works, but as long as they know it does work perhaps they don't need to
>> know why to reverse engineer it.
>
>
> ​> *​*
> *If you arguing from the size of the genome then that is a bit odd.  The
> brain physically is produced from the genome *
>

The adult brain is different from a newborn's because
​ ​
it has learned from
​ ​
events in the environment, but so would a
​n​
AI. The question is how complicated does a machine have to be to
​be ​
able to learn and become intelligent at a human level? Well, the entire
human genome is 750 meg with about half of that used for the brain, so at
most the instructions for building a seed AI would have to be about 375
meg,  and almost certainly less than that and probably a lot less.

​> ​
>  *over many years to full maturity.*


​How long a cake needs to bake in the oven tells you nothing about the
length of the recipe. ​


​>* ​*
> *And NNs are notoriously opaque to understand the workings of quite unlike
> software most SEs deal with.*
>

​True, but once the seed AI is built humans no longer have to understand
how it's mind works, we just introduce it to the world and let it learn
stuff just as we do with a human child. ​


​> *​*
> *I could be wrong but I have long had the opinion that getting to AGI by
> uploading a human brain effectively is sort of like getting to human flight
> capability by scaling up literal birds, flapping and such and all. *
>

​I think so too but like you it's only a hunch and I could be wrong.​


​> ​
> I think that we will arrive at distinctly not human AGI by a mixture of
> methods long before we have true human upload capability.
>

​But I think once machine intelligence has been achieved, either by AI or
reverse engineering the human brain and uploading, nothing will take very
long anymore; things will start to change at a breakneck pace. And I mean
that literally, human biological necks could very well end up getting
broken.   ​


​ John K Clark​












>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20160701/2e8e5ef4/attachment.html>


More information about the extropy-chat mailing list