4P's for September 12, 1996

Progress

1. Implemented a quick, simple Soar7 web server daemon and put it out on the web.

2. Compiled and linked Soar 7.0.4 (Tcl only) in Windows 95 using Microsoft Visual C++. It doesn't run yet.

3. Had a long phone discussion with Randy last Tuesday about the sum-to-min stuff. We decided that there were two tracks to pursue:

   1. Getting a mechanism up and running that could learn to adjust preconditions. This basically involves trying to get the feature association software I'd worked on to do stuff within the context of learning about procedures.

   2. Focusing on refining the Soar model for the SUM-to-MIN shift, specifically concentrating on determining how we'd learn that it is not a requirement for the "aural buffer" to be cleared before counting out the second hand.

Plans

1. Start integrating the feature association stuff into my GIPS/BlocksWorld implementation. My goal here is to (1) have GIPS learn a new "internal" subgoal (e.g., that a block must be clear before you can place it on another block), and (2) have GIPS "unlearn" and unnecessary subgoal (e.g., that a block must be on the table before you can place it on another block).

2. Identify and document a specific mechanism for producting the SUM to Shortcut-SUM strategy shift using mechanisms available within Soar.

3. Soar project work? GUI stuff? Meet with Karen (by phone?) to discuss Tcl7.5/Tk4.1 issues? Enhancements to web stuff (e.g., with Sayan)?

Priorities

1. Formalize and document the feature association stuff.

2. Complete the implementation of the FIRST and MIN strategies. After doing the SHORTCUT-SUM, I have a new respect for the amount of energy it takes to massage these together coherently. FIRST will be hard (SAY "addend" == REPRESENT-ADDEND): MIN given FIRST should be easy (just preference on assignment). This needs to get done eventually, but we've all agreed that it'd be most productive to concentrate on the first strategy shift.)