On Thu, Dec 27, 2012 at 4:36 PM, Mike Tintner <[email protected]> wrote: > Stop with the jokes, already, Matt.
The basic function of an operating system is to allow a computer to load, run, and interrupt other programs. Not every computer has an operating system. The computers in your printer, car, TV, clock, microwave, coffee maker, etc. do not have operating systems. A phone has an operating system if it lets you install and run apps. I once programmed a UNIVAC-19 which the U.S. Navy was still using in 1980 to control its missile launchers. It had an operating system consisting of 32 18-bit words that were hard wired into the discrete diode-transistor logic to bypass the 32K word magnetic core memory. It loaded and executed a fixed sized block of memory from paper tape (normally another loader for the rest of the program). It performed all the basic functions of a minimal operating system except for stopping the program, which you did by pushing a button on the front panel of blinking lights. My proposal is for a distributed index that allows you to plug in narrow experts or peers. They communicate by sending messages to a global pool that all other peers can read. Once a message is posted, it cannot be deleted or altered. A message contains a return address (like a URL), time stamp, a body consisting of information in human readable form such as English text or a JPEG image, and a signature authenticating the sender using a key known only to the sender and receiver. When peers communicate for the first time, they establish a shared secret key through a protocol like Diffie-Hellman. Then messages are signed by appending SHA-256(message, key), which the receiver can verify. Messages are routed to "anyone who cares". It is assumed that you care about a message if there is high mutual information between it and messages you have sent in the past, i.e. you can compress it well. Peers can guess what messages might interest you by matching them to messages you sent previously. When a peer routes a message, it adds its return address and timestamp to the header, signs it, sends it, and keeps a copy. Peers are free to use whatever policy they wish for keeping, routing, discarding, and blocking messages, and they may use any algorithm they wish for matching message content. However, there is an economic incentive to do this intelligently, i.e. to provide high quality information to those who want or need it. Doing so allows the peer to command a higher price for advertising, i.e. routing certain messages for a fee. In my MS thesis I ran simulations of networks to show that a distributed index requires O(n log n) storage, O(log n) query time, and O(log^2 n) update time using a vector space model for matching messages. It is fault tolerant because there are multiple routes between peers. The current design is robust even against malicious attacks. There is no central controller which could bring the whole network down. One drawback that I mentioned would be the possibility of consolidating power in high reputation peers, but we already have that problem. We depend on Google, Amazon, eBay, etc. to screen user ratings of small merchants. So you can think of a distributed index as sort of an operating system for plugging in experts. An expert might be a calculator that only recognizes messages containing arithmetic expressions and ignores everything else. It could be a program that sends out a picture of Barack Obama whenever it sees a message with words like "Obama" and "photo". This allows a very high degree of parallelism. It allows anyone with an internet connection to add knowledge and computing power. It creates a market for peers that curate high quality knowledge and weed out spam. AGI comes from having billions of experts competing for your attention, such that a peer somewhere could answer any question you might ask. At this point, you might ask "where is my house cleaning robot?". Again, AGI is a global effort. What I am doing is designing something like the first versions of the HTTP and HTML protocols. I am not building the web. The first version of the Mosaic browser and NCSA server was a 6 week effort by one person. So Mike, what was your design again? -- -- 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
