Sounds fascinating, Alan.

Imagine, the Smalltalk written in itself teaching itself to itself.  (-:

Gary


----- Original Message ----- 
From: "Alan Grimes" <alangrimes at starpower.net>
To: <squeak-dev at lists.squeakfoundation.org>
Sent: Tuesday, August 13, 2002 3:09 AM
Subject: AI project. =)


> om
> 
> I went through all the tutorials behind the non-broken links. (time to
> do some maintainance at www.squeak.org!). I am pretty sure I have my
> bearings at this time. In this posting I am going to sketch out the
> project project I am about to attempt. If it were anything easier I
> could probably manage on my own. As AI is a really wierd area of
> software, it has some highly unusual requirements. While all the
> algorithmic work will be my own, I could use some advice about how best
> to use squeak features for this. 
> 
> The AI will attempt to simulate the cognitive processes of an immaginary
> user. It will function by analysing the pixel pattern of the screen and
> moving the mouse. It will also be able to type at an immaginary
> keyboard. Since it has an unlimited number of fingers, it will be able
> to use the entire ASCII set. =P 
> 
> The special requirements are as follows: 
> 
> There needs to be a windowed bitmapped display of a pre-defined
> unchangable resolution, say 800x600x32 bit -- 4 bit may be easier to
> analyze... 
> 
> After each display update the AI needs to receive an exact cepy of this
> data which it can check for changes and update its analysis. -- It may
> be possible to link the updating with the perception of change, but that
> may be too complex. 
> 
> There needs to be an seperate AI pointer that will be entirely
> controlled by the AI. It must have all the capabilities of the mouse but
> be confined to the common region of the AI's world. 
> 
> In this area there will be a common region and a user region. The common
> region will be some strongly bounded subset of the window to which the
> AI's pointer will be confined. Outside of that there will be a box for
> buttons, a text input line, and a column for logs and other data. 
> 
> AI is the wild unknown. It is not at all clear what capabilities (if
> any!) the system will have. It will be important to start the AI off
> with simple learning toys and gradually move it up to more and more
> complex tasks. I have been reading many things about the future of AI,
> particularly that at www.singinst.org . The predictions at that site are
> shocking to say the least. Part of my reason for undertaking this
> project is in response to what I have seen there. =\ Because of this, I
> will want to ensure that every possible step has been taken to controll
> the rate at which the AI gains power so that I can be confident that it
> will not become a risk to anyone. Basically, I want to treat the AI as a
> highly untrusted user and then gradually lift restrictions as it grows.
> 
> All of the functionality I have mentioned will be in an "AI testbench"
> package. Asside from its wierd IO requirements, it shouldn't be too much
> of a challenge, and be quite useful to many in the field. 
> 
> -- 
> Linux has more source code than my brain.
> http://users.rcn.com/alangrimes/
>