(note followup updates to this post are available here: FreeLifePlanningCoachSoftwareUpdate and here: FreeLifePlanner)
This is the page for the Free Life Planning Coach Software, obviously a working name. Some other names that are under consideration are:
Action Planner Get Help Self-Reliance On Your Feet
Hard to precisely define it. This system, called Action Planner hereout, grew out of work on the SPSE2 system, a personal task management application. SPSE2 can represent Goals and their interdependencies, and can perform temporal planning and interactively walk users through the resulting plans. However, with the SPSE2 system, as many goals were being added, it became difficult to know really which of the goals that could be accomplished were more important and hence should be prioritized.
One approach is to sort the list of available goal activities by asking a series of pairwise comparisons. Another method from decision science is to ask the user to rank the goals etc. We will probably be implementing the former at some point in time. But, this system, Action Planner, grew out of a desire to have the system help in the selection of which goals to achieve, followed by their planning using the temporal planner. The idea is that if the temporal planner fails to find a plan that satisfies all goals, to know which goals to ignore first, and continue replanning (which is fairly efficient) until a plan is found that SATISFICES all of the major things. So for instance, if you had two goals, one was "learn stroke prevention" and another was "buy an HDTV", the system would be able to classify the latter as being a luxury, and the former as being part of emergency preparedness. Then, using a system to compute priorities, it would, barring unforeseen arguments in favor of the HDTV, say that the first goal was more important because it improved other goals like "Get on top of preparedness planning". The idea with allowing contravening arguments is that, while these goals have certain priorities such that certain goals override others, it is possible that there are complicated dependencies. So for instance, if the HDTV was for a person confined to hospice care with terminal cancer, such overriding arguments could be brought to bear. In AI, we call this non-monotonic, defeasible or default reasoning (if an argument can be overcome by another).
The canonical example is that you might have a system with the rule: "birds fly", and then you might add that "Tweety is a bird", and so deduce that "Tweety can fly". However, if you further add that "Tweety is a penguin", and "penguins can't fly", then the original assertion that "Tweety can fly" will be overridden.
So, as has been mentioned briefly, there will be a system of priorities, which understand the default rules regarding which goals are more important than others. This also relates to ethics, especially deontic ethics. Deontic logic is the logic of morality concerning the moral status of duties. So for instance, certain actions are prohibited, others are obligatory, some or permissible, and so on. A similar system of non-monotonic reasoning about deontics (as well as systems for virtue ethics and "the end justifies the means" ethics) will be brought in to weigh on the importance of a task from a moral perspective.
Knowledge of what happens when a goal fails to be accomplished, and the severity of the consequences will also have a commonsense AI implementation. This will be part of the effects matrix. Similarly, we will try to reason with what happens to systems involved when there are different failures, to help to develop plans that guard against these problems.
The system will have a pairwise comparison that can compare between two goals to determine which is more important. Such a capabability would be useful for solving related problems from microeconomics: the expenditure minimization problem and the utility maximization problem. (See UtilityMaximizationSystem).
That's an overview of the basic system. The idea is to get it into a position where it can assign scores to goals saying how important they are. The scores thus computed could be arrived at in a manner similar to the AI techniques mentioned in the following two papers, except that I don't mean exactly those techniques but an analogy to the general process of taking all of the facts of a situation and applying a set of processes to arrive at answers about the relative utility of a goal.
http://taeb-nethack.blogspot.com/2009/08/planar-taeb-ai.html
http://frdcsa.org/~andrewdo/archive/taeb/taeb.html
https://www.youtube.com/watch?v=2JY5-jV29hI
http://www.gnu.org/software/gnugo/gnugo_4.html#SEC39
Here is an example goal hierarchy, but unfortunately for military purposes.
http://www.isi.edu/soar/ifor/agent/
The action planner is partially implemented. So far, it consists of rules in the Expert System programming language called CLIPS. Much like the above linked GNUGo AI, the current CLIPS based Action Planner code will take a static situation, expressed as a set of facts, and compute a set of recommended moves for the agent, based on the rules. That will be the Expert System component of the move generation phase. The idea is that the situation is completely reevaluated at each "step", just like with GNUGo.
(Note - Useful ideas for what are essentials can probably be found by text mining the literature of survivalism.)
So, I need to work more on constructing the basic SPSE2 goal map that includes all the sane goals an individual might have. Some of the goals may have to be recommended based on the users situation by the move generator. In general, the move generator and SPSE2 will talk to each other using cape.sourceforge.net, and have redundant goal representations.
So, for instance, here is the file that illustrates all of the facts of a fictitious situation. Feel free to skim them.
(deffacts init (has-job John-Doe Teacher) ;; (not (disqualified-from-receiving-foodstamps John-Doe)) (address (name John-Doe) (firstname John) (lastname Doe) (address1 13-Berry-Street) (city Sheridan-Lake) (state Colorado) (zipcode 81071) (country USA) ) (has-medical-condition John-Doe Diabetes) (has-health-insurance John-Doe Blue-Cross) (has-bank-account John-Doe PNC-Bank) (father-of John-Doe Steven-Doe) (mother-of John-Doe Margarette-Doe) (brother-of John-Doe Mark-Doe) (is-dead Margarette-Doe) (relationship-strength John-Doe Steven-Doe okay) (relationship-strength John-Doe Mark-Doe good) (has-friend John-Doe Jeremy-Pierce) (relationship-strength John-Doe Jeremy-Pierce fair) (has-friend John-Doe Dina-M-James) (relationship-strength John-Doe Dina-M-James good) (person John-Doe) (person Steven-Doe) (person Mark-Doe) (person Margarette-Doe) (person Jeremy-Pierce) (person Dina-M-James) (has-automobile John-Doe Chevy-Camero) (has-laptop John-Doe dell-laptop) (has-computer John-Doe office-computer) (debt (debtor John-Doe) (amount 3000) (creditor Fannie-Mae) (type Student-Loan)) (debt (debtor John-Doe) (amount 500) (creditor Flaherty-Foot-Clinic) (type Medical)) (has-goal John-Doe "(is-self-reliant John-Doe)") ;; (not (is-emotionally-secure John-Doe)) (has-fitness-plan John-Doe John-Doe-s-fitness-plan) ;; (not (has-financial-plan John-Doe ?plan)) ;; (not (has-budget John-Doe ?budget)) )
The next step then, are the rules. Here are the rules thus added to the system.
http://frdcsa.org/~andrewdo/projects/action-planner/extension-0.clp.html
http://frdcsa.org/~andrewdo/projects/action-planner/extension-1.clp.html
http://frdcsa.org/~andrewdo/projects/action-planner/extension-2.clp
That's just some of the rules. The end result of running this is that there are a lot more facts asserted about which goals a person should have based on their situation.
And here are the example facts when run on the latest version (at time of writing) of those rules and initial facts.
CLIPS> (facts) f-0 (initial-fact) f-1 (has-job John-Doe Teacher) f-2 (address (name John-Doe) (firstname John) (middle_initial nil) (lastname Doe) (address1 13-Berry-Street) (address2 nil) (city Sheridan-Lake) (county nil) (state Colorado) (zipcode 81071) (country USA)) f-3 (has-medical-condition John-Doe Diabetes) f-4 (has-health-insurance John-Doe Blue-Cross) f-5 (has-bank-account John-Doe PNC-Bank) f-6 (father-of John-Doe Steven-Doe) f-7 (mother-of John-Doe Margarette-Doe) f-8 (brother-of John-Doe Mark-Doe) f-9 (is-dead Margarette-Doe) f-10 (relationship-strength John-Doe Steven-Doe okay) f-11 (relationship-strength John-Doe Mark-Doe good) f-12 (has-friend John-Doe Jeremy-Pierce) f-13 (relationship-strength John-Doe Jeremy-Pierce fair) f-14 (has-friend John-Doe Dina-M-James) f-15 (relationship-strength John-Doe Dina-M-James good) f-16 (person John-Doe) f-17 (person Steven-Doe) f-18 (person Mark-Doe) f-19 (person Margarette-Doe) f-20 (person Jeremy-Pierce) f-21 (person Dina-M-James) f-22 (has-automobile John-Doe Chevy-Camero) f-23 (has-laptop John-Doe dell-laptop) f-24 (has-computer John-Doe office-computer) f-25 (debt (debtor John-Doe) (creditor Fannie-Mae) (amount 3000) (type Student-Loan)) f-26 (debt (debtor John-Doe) (creditor Flaherty-Foot-Clinic) (amount 500) (type Medical)) f-27 (has-goal John-Doe "(is-self-reliant John-Doe)") f-28 (has-fitness-plan John-Doe John-Doe-s-fitness-plan) f-29 (has-student-loan-debt John-Doe) f-30 (has-access-to-computer John-Doe office-computer) f-31 (can-run-get-help John-Doe office-computer) f-32 (has-access-to-computer John-Doe dell-laptop) f-33 (can-run-get-help John-Doe dell-laptop) f-34 (is-alive Dina-M-James) f-35 (should-have-goal Dina-M-James "(has-financial-plan Dina-M-James)") f-36 (is-alive Jeremy-Pierce) f-37 (should-have-goal Jeremy-Pierce "(has-financial-plan Jeremy-Pierce)") f-38 (should-have-goal Margarette-Doe "(has-financial-plan Margarette-Doe)") f-39 (is-alive Mark-Doe) f-40 (should-have-goal Mark-Doe "(has-financial-plan Mark-Doe)") f-41 (is-alive Steven-Doe) f-42 (should-have-goal Steven-Doe "(has-financial-plan Steven-Doe)") f-43 (is-alive John-Doe) f-44 (has-personal-relationship John-Doe Jeremy-Pierce) f-45 (has-personal-relationship Jeremy-Pierce John-Doe) f-46 (has-personal-relationship John-Doe Dina-M-James) f-47 (has-personal-relationship Dina-M-James John-Doe) f-48 (has-support-network John-Doe) f-49 (should-have-goal John-Doe "(has-financial-plan John-Doe)") f-50 (has-goal John-Doe "(has-financial-plan John-Doe)") f-51 (is-related-to John-Doe Mark-Doe) f-52 (is-related-to Mark-Doe John-Doe) f-53 (has-personal-relationship Mark-Doe John-Doe) f-54 (has-personal-relationship John-Doe Mark-Doe) f-55 (is-related-to John-Doe Margarette-Doe) f-56 (is-related-to Margarette-Doe John-Doe) f-57 (is-related-to John-Doe Steven-Doe) f-58 (is-related-to Steven-Doe John-Doe) f-59 (has-personal-relationship Steven-Doe John-Doe) f-60 (has-personal-relationship John-Doe Steven-Doe) f-61 (has-access-to-a-bank-account John-Doe) f-62 (must-pay-taxes John-Doe)
So, just as in a game of Go, the situation/position could be representing textually, we have a representation of the board of a person's life. Like GNUGo, we have the initial facts, and by application of the rules we have the analysis, only in this case the analysis phase is incomplete. But hopefully, it will be able to synthesize a coherent set of choices to be made and recommend the best move for the position. It seems to me silly that we have software that can play a great game of Go (computer is playing White and is clearly beating me (Black)), but the same cannot be said of a system that helps one get on top of one's life, plan a good life, cover all the bases, etc.
So, that's the status of the Action Planner at present. What is needed is to continue developing the taxonomy of needs and wants, to implement the non-monotonic logic for determining an item's importance. Knowledge management approaches should be taken. Pending approval by the Free Software Foundation, I may consider using the Java based Jena for the rules.
Naturally, we want as many good rules as we can think of. Please send me such rules. We don't have to worry so much about rules drowning each other out because hopefully information about the relative importance of goals will outweigh the information about the goals, so that we can always effectively know whether it is simple or not to determine which goal is more important.
Update: Sun Apr 24 16:51:09 CDT 2011
I have completed preliminary extraction of properties from texts. The method chosen was thus. I am using wikihow.com as a source of Creative Commons licensed howtos. I downloaded a bunch of these howtos. Given a single one, I run it through a script that extracts the steps and possibly tips. I extract the individual sentences and then run them through the KNext system. From there I post-process the resulting information into a form that makes good properties. To wit:
Organize Paperwork at Home
Do you have piles of paper in your home that are overwhelming you? You can
organize them so you can actually find what you need when you need it. You can
cut the clutter by understanding what it is okay to throw out.
edit Steps
1. 1
Get some file folders and a pen. You will need 1 large box and a file box
to put the files in. If you have many piles, start with the smallest.
2. 2
Eat the elephant one bite at a time. Mark boxes, TRASH, and FILE and begin
sorting just on these two categories.
3. 3
If you keep an article of paper make a file for that subject, (example:
Unpaid Bills). Put the bill in the file folder and label the tab Unpaid
Bills and set it in the file box.
4. 4
Keep making files for whatever subject you need to file labeling them as
you go. These could be anything you think is important to keep: birthday
cards, car information, checking account, dental, home loan information,
home maintenance, insurance, medical, unpaid bills, utilities, etc.
+ The best place to file something is where it is most obvious to you.
Put things where you think they should go and it will be easiest for
you to find them later.
5. 5
In another box marked "Trash" put in junk mail, duplicate bills, old papers
that you know you will never really need.
6. 6
Touch it once. Either file it or throw it out.
7. 7
Make sub files if you want to. After you file your papers you might want to
make files within files that are close to the same subject. For example,
you could divide car information into car insurance, recalls/warranties,
repairs,and copies of titles and registrations. This way you have all the
info when you need it.
8. 8
Start at the mailbox for new papers that come in. When you get your mail
out of your mailbox, go through right away and toss the junk. When you get
into the house, make it a point to place it near your file cabinet. Choose
a specific time that fits your schedule to open your mail and sort it on a
daily basis. You could put unpaid bills in a folder in the front of the
file box to help you to remember to pay them. File the important things and
toss the rest in the trash can located next to your file cabinet.
9. 9
Use your calendar. Does a particular piece of paper require action or
attendance at a later date? Write it in your calendar, then file the paper.
You can even write where you filed the paper.
I extract these sentences, and process with KNext, removing errors:
$VAR1 = [
'A BOX CAN BE LARGE.',
'A PERSON MAY HAVE SOME_NUMBER_OF PILES.',
'BOXES MAY TRASH.',
'A MALE MAY HAVE BOXES.',
'A PERSON MAY KEEP AN ARTICLE OF PAPER.',
'AN ARTICLE CAN BE CONSTITUTED_OF_OR_FILLED_WITH PAPER.',
'A FILE CAN BE FOR A SUBJECT.',
'AN UNPAID[PERSON??] MAY HAVE BILLS.',
'THINGS MAY GO.',
'THINGS CAN BE PUT.',
'A PERSON MAY THINK INDIVIDUAL -S.',
'A THING-REFERRED-TO MAY BE EASIEST -ED FOR A PERSON.',
'A PERSON MAY FILE PAPERS.',
'A PERSON MAY HAVE PAPERS.',
'A PERSON MAY DIVIDE CAR INFORMATION.',
'COPIES CAN BE CONSTITUTED_OF_OR_FILLED_WITH TITLES.',
'A WAY MAY HAVE AN INFO.',
'A PERSON MAY NEED A THING-REFERRED-TO.',
'START CAN BE AT A MAILBOX.',
'A MAILBOX CAN BE FOR PAPERS.',
'A PERSON MAY GET_OUT MAIL.',
'A PERSON MAY HAVE MAIL.',
'A PERSON MAY HAVE A MAILBOX.',
'A PERSON MAY GET INTO A HOUSE.',
'A PERSON MAY HAVE A FILE CABINET.',
'A PERSON MAY HAVE A SCHEDULE.',
'A PERSON MAY HAVE MAIL.',
'A BASIS CAN BE DAILY.',
'BILLS CAN BE IN A FOLDER.',
'BILLS CAN BE UNPAID.',
'A FOLDER CAN BE IN A FRONT OF A FILE BOX.',
'A FILE BOX MAY HAVE A FRONT.',
'THINGS CAN BE IMPORTANT.',
'A REST CAN BE IN A TRASH.',
'A PERSON MAY HAVE A FILE CABINET.'
];
And then finally convert these with some regexes and conditionals into this (note proper conjugations haven't been completed for most verbs):
(is-large ?box) (has-some_number_of-piles ?person) (trash ?box) (has-boxes ?male) (keep-an-article-of-paper ?person) (is-constituted_of_or_filled_with-paper ?article) (is-for-a-subject ?file) (has-bills ?unpaid[person??]) (go ?thing) (is-put ?thing) (think-individual--s ?person) (was-easiest--ed-for-a-person ?thing-referred-to) (file-papers ?person) (has-papers ?person) (divide-car-information ?person) (is-constituted_of_or_filled_with-titles ?copy) (has-an-info ?way) (need-a-thing-referred-to ?person) (is-at-a-mailbox ?start) (is-for-papers ?mailbox) (get_out-mail ?person) (has-mail ?person) (has-a-mailbox ?person) (get-into-a-house ?person) (has-a-file-cabinet ?person) (has-a-schedule ?person) (has-mail ?person) (is-daily ?basis) (is-in-a-folder ?bill) (is-unpaid ?bill) (is-in-a-front-of-a-file-box ?folder) (has-a-front ?file-box) (is-important ?thing) (is-in-a-trash ?rest) (has-a-file-cabinet ?person)
Finally I will either manually prune them or work further to extract rules in addition to properties.
So far, I have progressed with the extraction of properties. Unfortunately, the extracted properties are not as good as I would like, and there are tons of them that don't really make sense. But by a few heuristics I think I will be able to get to the good rules rather quickly.
Here is a sample of the output data.
?computer
(has-computer ?camera ?computer) A CAMERA MAY HAVE A COMPUTER. (PC SOFTWARE CAN OFTEN SMOOTH NOISE IN POST PRODUCTION COMMA WHILE PRESERVING DETAIL BETTER THAN TH\
AT SUPPLIED IN THE CAMERA S INTERNAL
COMPUTER COMMA AND YOU CAN REVERT TO YOUR ORIGINAL IMAGE IF YOU DECIDE YOU DID N T WANT SO MUCH SMOOTHING COMMA OR EVEN ANY AT ALL)
(has-computer ?individual--s ?computer) INDIVIDUAL -S MAY HAVE A COMPUTER. (THERE ARE PEOPLE OUT THERE WHO WILL DOWNLOAD PICTURES OF YOU ONTO THEIR COMPUTER AND PORE OVER THE\
M)
(has-computer ?person ?computer) A PERSON MAY HAVE A COMPUTER. (IF YOU HAVE A COMPUTER ON COMMA AND IT S NOT THE SOURCE OF LEARNING COMMA TURN IT OFF)
(has-desktop ?computer ?desktop) A COMPUTER MAY HAVE A DESKTOP. (DO N T HAVE ANY ICONS ON YOUR COMPUTER S DESKTOP)
(has-keyboard ?computer ?keyboard) A COMPUTER MAY HAVE A KEYBOARD. (LABEL-MAKERS THAT CONNECT TO COMPUTERS SAVE A LOT OF TIME BY ALLOWING DATA ENTRY ON THE COMPUTER S\
SUPERIOR KEYBOARD COMMA OR EVEN FROM A
DATABASE OR OTHER COMPUTER FILE)
(has-outs ?computer) A COMPUTER MAY HAVE OUTS. (YOUFLYERSPUSH THE HTML BOUNDARIES LEARN THE INS AND OUTS OF YOUR COMPUTER)
(has-security ?computer ?security) A COMPUTER MAY HAVE A SECURITY. (PROGRAMS COMMA PLUG-INS OR GAMES COMMA THREATENS YOUR OWN PRIVACY AND THE SECURITY OF YOUR COMPUTE\
R)
(has-time-system ?computer ?time-system) A COMPUTER MAY HAVE A TIME SYSTEM. (GO INTO YOUR COMPUTER S TIME SYSTEM AND SET THE TIME FORWARD COMMA INTO THE FUTURE)
(has-wordpad ?computer ?wordpad) A COMPUTER MAY HAVE A WORDPAD. (USE PAPER AND PENCIL OR TRY RECORDING ON THE WORDPAD OF YOUR COMPUTER)
(order-files-by-name ?computer) COMPUTERS MAY ORDER FILES BY NAME. (WHEN YOU TRANSFER YOUR PHOTOS FROM YOUR CAMERA TO YOUR COMPUTER COMMA IMMEDIATELY PUT THEM INTO A \
FOLDER ON YOUR HARD DRIVE -- NOT JUST
MY PICTURES COMMA BUT CREATE A SUBFOLDER BY DATE -LRB- USE REVERSE DATE FORMAT E.G. 2007-06-26 WHICH IS LISTED BETTER BY COMPUTERS
ORDERING FILES BY NAME -RRB- COMMA EVENT NAME OR BOTH)
(use-a-firewire ?computer) A COMPUTER MAY USE A FIREWIRE. (IF SO COMMA CONNECT THE CAMCORDER TO THE COMPUTER USING A FIREWIRE OR USB CABLE)
?computer-program
(has-computer-program ?person ?computer-program) A PERSON MAY HAVE A COMPUTER PROGRAM. (ALTERNATIVELY COMMA YOU CAN MAKE A COLORFUL CHRISTMAS THEME ON YOUR COMPUTER PROGRAM THAT \
WILL PRINT DIRECTLY ONTO PLAIN PAPER)
?computer-system
(has-limitations ?computer-system) A COMPUTER SYSTEM MAY HAVE LIMITATIONS. (ANY SPECIFIC COMPUTER SYSTEM THAT IS SOLD AS A PC COMMA NO LONGER HAS THE SAME LIMITATIONS OF PCS \
You can see all of the extracted properties here:
http://frdcsa.org/~andrewdo/projects/action-planner/all-properties.txt
And here are the properties selected for inclusion so far:
http://frdcsa.org/~andrewdo/projects/action-planner/chosen-properties.txt
These properties will be useful in doing positional analysis of a persons life. I will also process Sensei's net to extract Go knowledge, and also recipes to extract recipe knowledge. In fact, the system is a (rather slow) ontology development system, most of the work being done by KNext, but I will be making some additions that extend that core functionality.
Another aspect of the life planner, once the positional attributes of the "game" of life are known, will be actually developing a strategy. Classically this might involve running a PDDL planner on a custom domain. However, another tack I'm investigating is the use of so-called General Game Playing software. Fortunately, a very winning program has been released, called CadiaPlayer, which is available for download here:
http://cadia.ru.is/wiki/public:cadiaplayer:main
I will be experimenting with generating a game which simulates aspects of life, written in the corresponding GDL (Game Description Language?).