--- On Thu, 10/2/08, Ben Goertzel <[EMAIL PROTECTED]> wrote:
> OTOH a global brain coordinating humans and narrow-AI's can **also** be quite 
> dangerous ... and arguably more so, because it's **definitely** very 
> unpredictable in almost every aspect ... whereas a system with a dual 
> hierarchical/heterarchical structure and a well-defined goal system, may 
> perhaps be predictable in certain important aspects, if it is designed with 
> this sort of predictability in mind...

CMR (my proposal) has no centralized control (global brain). It is a 
competitive market in which information has negative value. The environment is 
a peer-to-peer network where peers receive messages in natural language, cache 
a copy, and route them to appropriate experts based on content.

Peers have incomplete knowledge of the network, so messages may need to be 
routed via multiple hops through redundant paths to multiple experts. Each 
message identifies the sender and time sent. The receiver is responsible for 
authenticating the sender, e.g. by password and registration via an encrypted 
channel. The sender is a peer, not tied to a human. A human may manage multiple 
identities and be anonymous. Peer owners can set their own policies with regard 
to which messages to keep, route, or discard.

Initially, peers can be simple. When a peer receives a message, it matches 
terms to words in its cache, and forwards the message to the authors identified 
in the headers of the cached matches. A peer's domain of expertise is simply 
those messages posted by the author which are kept permanently in the cache. 
Peers can be more intelligent than this, of course. For example, they may match 
messages with attached pictures or video based on content.

The network's behavior can only be predicted in terms of market incentives. The 
network is hostile. Peers may be flooded with spam, so they will need some 
intelligence to decide which messages to route and which to discard. Resource 
owners (humans) compete for attention, which requires resources (storage and 
bandwidth) on other people's peers. Peers (or their owners) thus have an 
incentive to provide useful information so that they can sell advertising and 
are not blocked. Peers have an incentive to protect their reputations by 
preventing their identities from being forged. Thus, they have an incentive to 
keep passwords secret by e.g. registering with each neighbor using a different 
password.

I believe that CMR is initially friendly in the sense that a market is 
friendly. A market is the most efficient way to satisfy the collective goals of 
its participants. It is fair, but not benevolent. There is an incentive to 
cheat, but also an incentive to protect one's reputation by being honest. There 
is an incentive for peers to become more intelligent, as measured by earnings. 
Peers need to be selective in routing messages or else they will be exploited 
by spammers. Likewise, spammers have an incentive to outsmart weaker peers.

I believe that CMR becomes more dangerous as peers get smarter. We will rely on 
peers with high reputations to sort truth from lies and to rank the reputations 
of other peers. The problem is that we have to train these machines, for 
example, by clicking the "spam" button. But when machines are smarter than us, 
we can no longer make that distinction. I believe that eventually we will no 
longer know what our computers are doing as they acquire all available 
resources.

Although CMR is a specific proposal, I think it is clear that the internet is 
headed in this direction, even if it is not adopted as I described. We already 
depend on trust networks, like Google rankings alongside sponsored links, 
seller ratings on eBay, etc. Intelligent machines in any form will have to 
compete in this environment.

-- Matt Mahoney, [EMAIL PROTECTED]




-------------------------------------------
agi
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=8660244&id_secret=114414975-3c8e69
Powered by Listbox: http://www.listbox.com

Reply via email to