Search                        Top                                  Index
HELP RMS                                         Bob Searle and Chris Mellish

LIB RMS provides the basic facilities of a "Reason Maintenance System", as
described in:

     "Truth Maintenance Systems for Problem Solving"
     Jon Doyle, MIT AI-TR-419, January 1978

(It has nothing to do with Record Management Services). This brief HELP file
does not describe what a Reason Maintenance System is, but simply summarises
the procedures provided by LIB RMS and LIB RMSAPPS.

---- LIB RMS ---------------------------------------------------------------

The RMS deals with information in the form of NODES, which represent facts
that are believed (IN) or disbelieved (OUT). The procedure for creating a new
node is MAKENODE, which takes as its argument any structure which when printed
out gives a readable rendering of what the fact is. For instance:

    makenode([nobody likes me])=>
    ** <node1>

When printed out, for brevity a node comes out as the word "node" followed by
a unique number. A fuller printout is obtained by use of the procedure
SHOWNODE.

If a node is to be believed, it must be JUSTIFIED. The procedure JUSTIFYNODE
takes a node and two lists of nodes (the INLIST and the OUTLIST) and creates
an SL justification for the node. The idea is that the node will be believed
as long as it has a justification where everything in the INLIST is IN and
everything in the OUTLIST is OUT.

    vars sun, nosun;
    makenode([the sun is shining]) -> sun;
    makenode([the sun is not shining]) -> nosun;
    justifynode(sun,[],[^nosun]);
    shownode(sun);

    <node1> [the sun is shining]
            IN      <sl1>   []      [<node2>]

    shownode(nosun);

    <node2> [the sun is not shining]
            OUT

SHOWNODE gives the node number, human-readable form, and an indication as to
whether the proposition is IN or OUT. If it is IN, it prints the "name" of one
of its valid justification (here, "<sl1>"), together with the relevant INLIST
and OUTLIST.

The procedure RMSNODEISIN tests whether a node is IN or not. Eg. in the last
example:

    rmsnodeisin(sun)=>
    ** <true>

It may be useful to mark a node as CONTRADICTORY, using the procedure
RMSCONTRADICT. Such a node should never come IN. Note that the system will
only pay any attention to this information if it attempts to bring the node
IN. If the node is already IN, nothing happens immediately.

   rmscontradict(nosun);
   shownode(nosun);

   <node2> [the sun is not shining]
           OUT
                   ***contradiction***

If the system is ever about to bring IN a contradiction, it must sort things
out so that, by suitably changing the status of assumptions made earlier, the
contradiction is no longer believed. The procedure RMSFIXCONTRA is called
automatically for this. The procedure takes as its argument the contradictory
node. The default value of this procedure is RMSDEFFIXCONTRA, but the user may
assign a new value to it. Defining such a procedure appropriately would
require some study of the RMS code, however.


---- LIB RMSAPPS -----------------------------------------------------------

LIB RMSAPPS is not strictly necessary, but packages up some of the RMS
facilities in some useful ways. The facilities provided are:

   NODES - a global variable, holding a list of all the RMS nodes created
           by the following procedures:

        DECLARE - takes one argument, calls MAKENODE on it, returns the node
                  created and puts it on the list. The node is initially OUT.

        PREMISE - as DECLARE, except that the node is made IN by giving it
                  a justification with INLIST and OUTLIST []. Returns the node
                  as its result.

        ASSERT  - takes three arguments. Calls DECLARE with the first and then
                  justifies the resulting node with the other two. Returns the
                  node as its result.

   SHOWINNODES     - Three procedures for displaying the nodes in NODES. The
   SHOWOUTNODES      first two call SHOWNODE on only IN and OUT nodes
   SHOWNODES         respectively. The third calls SHOWNODE on all nodes.


--- EXAMPLES OF USE OF THE RMS SYSTEM --------------------------------------

(1) MIDSUMMER PROBLEM

;;; this scenarion demostrates the use of dependancy-directed backtracking
;;; to resolve 'inconsistencies' in belief systems.
;;; thses inconsistencies are declared by the particular belief system and
;;; are enforced by rms.
;;; this example is based on Jon Doyle, (MIT AI-TR-419, January 1978)
;;;     "Truth Maintenance Systems for Problem Solving"
;;; the story is that of Hermia, Lysander, Helena, and Demetrius from
;;;     Wm. Shakespeare's "A Midsummer Night's Dream"


;;; get rms primed up to go

lib rms;
lib rmsapps;

;;; initialise list of interesting nodes

[] -> nodes;

;;; declare some variables

vars node, tragedy, hermia, lyshermia, demhermia, lyshelena;

;;; first declare the loves of the women as unchangeable

premise([loves hermia lysander]) -> hermia;
erase(premise([loves helena demetrius]));

;;; now declare the loves of the men as assumptions
;;; demetrius starts out loving hermia, but if this turns out
;;;     to be unfortunate, he will turn his attentions to helena

declare([not loves demetrius hermia]) -> node;
assert([loves demetrius hermia],[],[^node]) -> demhermia;
erase(assert([loves demetrius helena],[^node],[]));

;;; similarly, lysander loves hermia,
;;;     but will settle for helena if required

declare([not loves lysander hermia]) -> node;
assert([loves lysander hermia],[],[^node]) -> lyshermia;
assert([loves lysander helena],[^node],[]) -> lyshelena;

;;; unfortunately, lysander is a jealous man,
;;;     and wants hermia for himself

premise([jealous lysander]) -> node;

;;; being violent, this leads him to want to kill demetrius,
;;;     because of both of their loves, and lysander's jealousy

assert([kills lysander demetrius],[^node ^demhermia ^lyshermia],[])
   -> node;

showinnodes();          ;;; display what is currently believed

;;; since this is supposed to be a comedy, and killing isn't funny,
;;;     the attempted murder is declared to be a tragedy
;;; first - we declare that a tragedy exists

declare([tragedy]) -> tragedy;

;;; second - tell rms not to allow this to come about

rmscontradict(tragedy);

;;; and, third - connect the attempted murder to the tragedy

justifynode(tragedy,[^node],[]);

;;; display rms' resloution of the conflict

showinnodes();
showoutnodes();

;;; unfortunately, hermia is taking this situation badly, and is
;;; about to commit suicide from unrequited love, because her
;;; chosen one, lysander now loves another

assert([suicide hermia],[^hermia ^lyshelena],[]) -> node;

;;; suicides aren't any better in comedies than murders, so this
;;; must also be declared a tragedy, and resolved by rms

declare([tragedy 2]) -> tragedy;
rmscontradict(tragedy);
justifynode(tragedy,[^node],[]);

;;; this brings the story to a successful conclusion, viz:

showinnodes();

;;; the items not currently believed are:

showoutnodes();

(2) DETECTION OF CIRCULAR REASONING

;;; demonstration of circular reasoning,
;;;     and the detection of such by rms

lib rms;
lib rmsapps;

vars node1, node2, node3, assump;
[] -> nodes;

premise('x+y=4') -> node1;
declare('assume 1') -> assump;
assert('x=1',[],[^assump]) -> node2;
assert('y=3',[^node1 ^node2],[]) -> node3;
shownodes();
justifynode(assump,[],[]);
shownodes();
declare('assume 2') -> assump;
justifynode(node3,[],[^assump]);
justifynode(node2,[^node1 ^node3],[]);
shownodes();
justifynode(assump,[],[]);
shownodes();