On Tue, Sep 30, 2014 at 5:03 PM, John Rose via AGI <[email protected]> wrote: > OK so let me figure this out I have some experience with messaging systems. > You indicate basic HTTP/HTTPS but there are issues with that basically you > need more since unsolicited messages are usually blocked (unless you are a > server) there are better protocols for this that your "AGI" could piggy back > on.
Setting up a P2P peer on your home computer or phone without a static IP is not trivial but people do it with tor and bittorrent, so it is not impossible either. You have to set up your router and firewall to allow incoming requests. But it would also work as a way for web servers to communicate. Most people would probably connect to a peer as a client and be able to receive messages over an open connection like most messaging apps. > Reading your thesis BTW which is pretty darn advanced for 1987 especially > covering distributed P2P systems with the hypercubes and all :) what happens > to all the messages that get routed to the "expert" peers? The peers answer > them and they get routed back but how does this scale up in intelligence? Is > it just that you have all these messages getting routed efficiently to the > optimal peers? The idea is that you get AGI from lots of narrow experts and a network that routes messages to the right ones. My thesis only shows that the network will scale globally, which I considered the main obstacle in the design. In the thesis I used a vector space model to match similar messages, but a more general approach is to use mutual information, since both are metrics. AGI doesn't just mean making a program that behaves like a human mind. It means doing everything that a human can do but without duplicating human weaknesses, like our poor arithmetic skills and our need to take time off work. The internet already has the right economic incentives (competition for attention and reputation) for sharing information, providing intelligent services, proving identity, and establishing reputation networks in a hostile environment. The system I described is already being built, except with more complex, ad-hoc protocols. We don't have distributed search, but now I am not so sure that would work in an environment where most small peers haven't established trust. It is technically possible now, for example, to forward queries to Google (probably with some payment) and replace their ads and sponsored results with your own. But who would use it? You have to offer something that Google doesn't already give you. -- -- Matt Mahoney, [email protected] ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
