[ExI] Limiting factors of intelligence explosion speeds

[Keith Henson]

On Sat, Jan 22, 2011 at 5:00 AM,  Eugen Leitl <eugen at leitl.org> wrote:
>
> On Fri, Jan 21, 2011 at 08:00:11AM -0700, Keith Henson wrote:
>
>> > It would be very easy to spot by sniffing traffic.
>>
>> Even without steganography a huge part of net traffic is compressed
>
> Steganography has negligible payload.

How much bandwidth do you need?

>> video and the like.
>
> Totally different animal. Every netop would instantly notice
> both the volume and the traffic type has changed.
>
> Sure, you can try hiding traffic in a bidirectional high-bandwidth
> "video" stream, but stream entropy and volume would give it away.

Perhaps.  I have no idea of how much net bandwidth it would take to
support a human level AI.  I suppose it would depend on how much
processing was local and how much went over the net.  I don't even
have a comparison on the traffic for a human brain but again it would
strongly depend on the partition level.

>> I don't personally believe this has happened, but the point is I don't
>> know how we could tell for certain.
>
> If the traffic is confined to a single installation and
> the external behaviour is indistinguishable from human ruminant
> browsing, then, no, we can't tell.

What level of parasite traffic would be noticeable?

>> > The bootstrap is extremely messy, and absolutely impossible
>> > to miss.
>>
>> I don't see why the bootstrap would necessarily be messy or impossible
>> to miss.  For example, if it had happened in the context of the
>
> Because a lot of hosts get compromised, and there's huge gobs of
> new traffic types, and it's not stereotypical.
>
>> Slammer worm it would have been messy, but we likely would have missed
>
> Nobody missed Slammer. (With nobody I don't mean users or the
> general public).
>
>> it happening.

What I meant was that had it happened buried in the Slammer mess it
would have been under the radar.

Keith