perm filename FUTURE.SUM[AP,DBL] blob
sn#114155 filedate 1974-08-05 generic text, type T, neo UTF8
00100 Future work, as forseen on 8/5/74 Doug Lenat
00200
00300 This is a brief sketch of future work, to be incorporated into the
00400 progress report if desired.
00500
00600
00700 Lenat is currently experimenting with the PUP5 system. By having it write
00800 more programs, one gains in the following ways:
00900 (i) More programming knowledge is accumulated
01000 (ii) The knowledge which is task-dependent becomes more distinct
01100 (iii) The task-independent knowledge is forced into more general forms
01200 (iv) Problems with the dialog style itself become apparent
01300
01400
01500 Also under consideration is the possibility of building a proof plan
01600 by automatic programming. As currently envisioned, this would entail
01700 sifting through a text, e.g. Topology, and letting the system write
01800 programs (and data structures) to capture the intent of each new concept.
01900 When a new statemnt is encountered which involves only old concepts, the
02000 system would "run" those programs; the result would either point to a
02100 counterexample or to a proof. This is analagous to mathematical
02200 intuition. Notice that the type of programs being written by the system
02300 are quite alien; one imagines vast amounts of tedious code being
02400 generated. This might be the first type of application of automatic
02500 programming: Just as the early uses of computers enabled vast amounts
02600 of tedious calculation to be performed automatically, and compilers
02700 allowed vast amounts of trivial code to be generated from smaller
02800 masses, so automatci programming could allow vast amounts of high-level
02900 code to be written.