Russ, CAIA, interesting find! I quote (from the paper):
MACISTE is a reflective system in three ways: > 1. Some MACISTE expertises have been defined so that they translate every > expertise > (including themselves) into an efficient C program. It translates its > 8,700 conditional actions into > 300,000 lines of C; not a single line of MACISTE or CAIA C programs has > been written by a > human being. This does not take into account all of the lines of C > programs written by CAIA when > it tries to solve problems using a combinatorial method; in one of its > "lives", it may generate > 300,000 more lines in these programs. The only human programs used are the > C compiler and the > Linux operating system. > 2. It can observe its behavior when it is running: it may generate a trace > as thorough as > necessary. Moreover, it can generate new expertises or new C programs and > use them immediately > without restarting the system. > 3. It can have access to the pile of all the active subprograms and to the > values of all their > variables. When there is a particular incident, it can find where it > happens and have a complete > overview of the situation. > All these possibilities are very useful for CAIA which can write and > execute new expertises > and programs, know what it has done and why it has done it, and also know > in what state it is when > a difficulty happens. > --Abram On Sun, Jun 24, 2012 at 9:47 AM, Russ Hurlbut <[email protected]> wrote: > Russell - > > Jacques Pitrat's MALICE may be a good place to start. The GPLv3 code > is at http://jacques.pitrat.pagesperso-orange.fr/MALICE.html > There is also a rather comprehensive paper that covers most of the > main concepts of his overall "Artificial Artificial Intelligence > Scientist" architecture on his home page: > http://jacques.pitrat.pagesperso-orange.fr/ > > The target problem domains of Pitrat's CAIA may be somewhat limiting > as to the usefulness for your purposes. > > As for a controlling meta language to generate compiled code, you may > also want to check out Paul G. Bassett's "Frame Technology" (not to > be confused with Minsky's Frames) > http://en.wikipedia.org/wiki/Frame_technology_(software_engineering) > Some Open Source code has been written for this ref: > http://xvcl.comp.nus.edu.sg/forum/viewtopic.php?t=106 > The FT code is also somewhat dated (has its roots in manipulating > COBOL) and many of the open source site links are broken, but the code > is still reachable. > > Combining the ideas from CAIA and FT may worth investigating. Please > share your opinion on these if you get a chance to install and try > them out. > > - Russ Hurlbut > > On Sat, Jun 23, 2012 at 2:27 AM, Russell Wallace > <[email protected]> wrote: >> The best representation of code depends on what you want to do with >> it. For efficient execution, you want machine code, or byte code that >> can be easily translated into machine code. If you want to reason >> about the code, however, it's better to use a tree or graph format >> (e.g. Lisp s-expressions). >> >> I'm currently looking at the problem of enabling heuristics to watch >> the execution of code (e.g. in order to notice when a search seems to >> have gone off down a blind alley so as to reset it with different >> parameters). This entails finding a representation of a snapshot of >> the current state of execution. Much less work has been done on this, >> and such as I'm familiar with, tends to be on the efficient execution >> end e.g. virtual machines that can dump the current state of memory >> and CPU registers; recovery of structure from that kind of flat format >> would be very difficult. >> >> Anyone know of any previous work on representations of execution >> snapshots suitable for high-level reasoning? >> >> >> ------------------------------------------- >> AGI >> Archives: https://www.listbox.com/member/archive/303/=now >> RSS Feed: https://www.listbox.com/member/archive/rss/303/18488709-43d1370c >> Modify Your Subscription: https://www.listbox.com/member/?&; >> Powered by Listbox: http://www.listbox.com > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/7190161-766c6f07 > Modify Your Subscription: https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com -- Abram Demski http://lo-tho.blogspot.com/ ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
