om In this posting I will argue that Go is an instance of the translation problem and hence a good computer Go player will be 80% of a sucessful AI.
om The first stone placed on a full-size 19x19 Go board can be placed in any one of 361 locations. The next stone picks from any of 360 places, etc... In this manner the number of possible two move games is 361*360*359*358, a combinatorial explosion. While it is true of chess, it is even clearer in Go; we are not looking at something where simple concepts can be applied. The pattern of stones on the board create a *concept space*. Go is intractable to a brute force approach. Instead we must create a software system that can 1. Apply abstract concepts to the pattern of stones, and 2. reason about these concepts in a manner sufficient to select an appropriate strategy. What are these concepts? In terms of position there is the concept of the _teritory_. A teritory is a region composed of stones of a given color _AND_ empty spaces such that no stone in that teritory can be captured by the other player. This is not a specific part of the game rather it is an *emergant concept* of the game. In the context of choosing the next moove there is the emergant concept of offence and defence. An offensive move is defined as a move that primarily weakens your opponent's position. A defensive move is defined as a move that primarily strengthens your own position. Auxilary to these you have concepts of strength and weakness where you apply your offensive strateies to your opponent's weaker positions and apply defensive moves to your own weaker territories. Other such concepts are listed in the varrious Go tutorials around the net... Remarkably (sufficiently to motovate this post!) is that this concept of concept spaces that we find in Go is exactly the same idea that leads us to declare a pattern of attoms to be a mouse or a diskette. _EXACTLY THE SAME!!_ The Go problem is a problem of _conceptualization_ or, "language generation". An ideal AI Go player will be able to recognise emergant concepts from the pattern of stones and then generate a language to describe this concept space. From there it is only a single step to apply cyc-like reasoning to the problem and arrive at a solution in short order. So how does this connect to the translation problem? The translation problem is just like the Go problem but it works in both directions, both recognising and generating patterns. In another one of my loosleaf notes I described the process of translation as follows, augmented with a relevant connections to this article. Lets consider the problem of the human translator who must listen to a verbal message and then repeat it in a different language. 1. The words are heard and then they are _ENCODED_ into an abstract representation of those words. That is, the EMERGANT CONCEPT of words is detected and analyzed in the auditory signal and an internal language is extracted. 2. This literal message is _DECODED_ into a pattern of sense perceptions or abstract pattern-ideas in the translator's mind. 3. The process of _ENCODING_ , just like the process required to play Go properly, is applied to these mental representations, extracting the abstract encoded representations of the words in the target language. 4. And finally, the words in the new language are _DECODED_ into the form of muscle contractions which result in the correct vocalization of the translation. This, according to Hofstadter, is the only problem in AI. Solve it and you take home the Big Trophy. om AIEEEE!!! The High Rabbi of the Singularity is on this list and this posting just might point him in the right direction. =(((( I think, by pushing the send button, I just seriously shot myself in the foot!! I'm so stupid!!! =((((( -- I WANT A DEC ALPHA!!! =) 21364: THE UNDISPUTED GOD OF ALL CPUS. http://users.rcn.com/alangrimes/ [if rcn.com doesn't work, try erols.com ] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]