Let me correct my conclusion. If Aaron's program was not successful after 2.5 years but mine was successful after 1.5 years then it would essentially stand as evidence that Aaron's plan to delay testing any AGI stuff until after the foundation of his program was well established was not as good a plan as mine which is to try to begin testing some AGi stuff as soon as is feasible. However, to substantiate that evidence further we would have to have some way to test those two theories of development a little further. Jim Bromer
On Mon, Jan 21, 2013 at 6:40 PM, Jim Bromer <[email protected]> wrote: > Aaron said: > Even if I don't need to extend or completely rewrite the underlying > framework due to some unforeseen issue, I predict it will take at least > another year before I start to see any fruitful results out of the system, > and more years still before the system starts to grow beyond the AGI > equivalent of infancy. I think AGI is just too big of a problem to see > instant results. Your own attempt could be seen as an attempt to falsify > that last statement. I hope you succeed. I look forward to seeing your > working AGi program in a year's time. > > ---------- > I don't want to be overly argumentative but if I was successful it would > not be a falsification of your point of view since I have spent years > working on this. I am now just reusing code that I was going to scrap years > ago but then decided to salvage. In trying to prepare for the kind of > complexity that I foresaw I became bogged down in trying to devise > complicated error handling. I wasn't able to figure that out and my > attempts to deal with text never got beyond the most basic stuff. Because I > tried to develop the program before I started to test it the program had > numerous (and frequently occurring) bugs so debugging was excruciating. So > as I am refurbishing my old code I am trying to simplify (while salvaging > the basic plan like using more than one thread) and at the same time I am > also reusing the old code to make an even simpler program to start to test > word frequency patterns as a method to distinguish broad sub-subjects as > soon as possible. Looking at this second program (the even simpler one) I > will start developing some actual AGi algorithms (very simple algorithms) > as soon as I get the file operations working again because I so totally > understand the need to develop the program incrementally. That means that I > cannot really test how objects derived from text would work without > actually developing some of the ways that the program will derive data from > the texts. --Although it might be difficult to deal with complications > that you could have foreseen, it is can be even more difficult to deal with > them before you actually see them (in real experiments). Yes there is a > point that over simplification at the start can lead you into false paths > but my experience is that trying to plan for something that you have never > encountered is itself a false path if it turns out that there were ways > that you could have started testing some of your ideas in similar > situations along the way.--- So, perhaps you weren't completely wrong. If > my project was successful in a years time then my project would be a > falsification of your method of delaying the testing of any AGi methods > until you got the foundational stuff all set. > > Jim Bromer > > > > > On Mon, Jan 21, 2013 at 4:04 PM, Aaron Hosford <[email protected]>wrote: > >> I'm going to make a prediction about my own project: It has taken me a >> year already, just laying the foundation, and I am not done yet. I am only >> just getting to the point where I can start to write and test code that >> actually does things remotely comparable to verbal/symbolic thinking. Even >> if I don't need to extend or completely rewrite the underlying framework >> due to some unforeseen issue, I predict it will take at least another year >> before I start to see any fruitful results out of the system, and more >> years still before the system starts to grow beyond the AGI equivalent of >> infancy. I think AGI is just too big of a problem to see instant results. >> Your own attempt could be seen as an attempt to falsify that last >> statement. I hope you succeed. I look forward to seeing your working AGi >> program in a year's time. >> >> >> On Mon, Jan 14, 2013 at 6:16 AM, Jim Bromer <[email protected]> wrote: >> >>> We are almost finished with two weeks of the new year. I said that I was >>> going to make a prediction about being able to get an AGi-Lite program >>> working within a year in order to demonstrate how an actual prediction like >>> this can be used as the basis for drawing conclusions of the effectiveness >>> of one's own theories. I agree that you cannot expect results in set period >>> of time but I was able to create other useful theories based on using >>> different kinds of reasoning on the prediction. The question now is whether >>> or not I can accept the results of my own experiments. >>> For instance, I said that if I was truly confident that I knew how to >>> create an AGI program (even an AGi-Lite program or AGi as I called it) then >>> I would be extremely motivated to get going on this project. So then, I >>> reasoned, if two weeks went by and I did not even have the user interface >>> done then this would be indicative that I wasn't quite as motivated as my >>> hubris would suggest. Well, I took the database definitions and the user >>> interface from the remains of an old program that I had abandoned then >>> salvaged a number of years ago and started working on it. It was much more >>> complicated than I remembered and even though I haven't been able to save >>> any data with it yet, it is slowly coming back to me. So yes, I had a >>> fundamental user interface within two weeks, and while that does not show >>> anything about whether my AGi ideas will work or not, it does show that I >>> have a fundamental enthusiasm and confidence in my theories. I have proven >>> nothing about my AGi theories, but I did take one fast indication of a >>> potential problem off the table. My conscious and my unconscious or >>> semi-conscious impressions of what I am doing are in sync. On the other >>> hand, since I did just grab a program out of the attic I should have made >>> more progress than I have. Two weeks is1/26 of the way to my predicted >>> goal. I also realized that I could begin making some very basic AGi >>> experiments using the text on the web and I feel that I should have started >>> that by now. >>> So prediction - (including prediction as a nexus of potential progress >>> overlayed with the nexus of dynamically developing plans) -is useful to me. >>> I can now use the goals - the original one and the new one produced by a >>> realization that I could start some initial testing using the web - to >>> create a new schedule that will provide me at a tiny bit more insight into >>> how my plans are starting to unfold. At this stage I haven't gotten any >>> results on any AI / AGI theories but if I am able to start testing one or >>> two of my ideas within the next two weeks I should have some kind of >>> results to examine. One thing I did learn was that to get an advantage on >>> the preliminaries it is nice to have something - that had been salvaged for >>> just such a situation - to use to jump into the fun part of the puzzle a >>> little faster. And that is a strategy that I can use in the next stage of >>> my planning. Instead of writing a web-crawler (which is what I would like >>> to do) I can just copy some text from various web pages and then paste them >>> into the user interface on my salvaged program and test some elementary >>> text searches to see if one of my ideas can actually be made to work. So >>> based on my experiences during the first two weeks of working with a >>> schedule I have developed a more efficient method of getting to the lowest >>> levels of the game. I still want to write a web crawler but I can do that >>> when I learn more about it. >>> So the next two weeks: That will be 1/13 of the way to the end of the >>> year. I better get going on this. (That is a familiar example of how >>> prediction and using the actual results of your efforts can drive insight >>> by the way). So I will like to get the bugs out of my salvaged program and >>> begin testing the database operations with automated methods during the >>> next 2 weeks. And I would like to conduct my preliminary text searches on >>> text that I can take from selected web pages to test my theories about the >>> relative words and count of those words associated with some key words >>> based on whether or not the key word was (according to my opinion) a >>> primary subject word or not. For instance, the word "flight" is found in >>> many types of texts, but by developing frequency of use records of the >>> words used along with the word "flight" I should be able to identify >>> whether the text is primarily about birds or about about airplanes or about >>> other sub-categories. (Although this is not a very exciting AI theory in >>> this day and age, I am using it as an example of how I am able to jump into >>> AI testing even though my program will take a some months to get up to >>> speed.) >>> Explanation of the Project >>> I got so tired of hearing people use the word "prediction" as a basis of >>> their AI/AGI theories that I decided to try using prediction in real life >>> to see if was an effective method. I found, that like most other AI >>> theories it worked really well in a few cases and not so well in most >>> cases. However, the use of formal (declared) predictions gave me a >>> surprising ability to crystallize new insight around the predicted events >>> when they were compared to the event. So I tried to get the guys who like >>> to use the term "prediction" in their discussions of AI to try this >>> experiment themselves. Of course they could not be bothered with such a >>> mundane experiment. So I challenged them: Why not use your predictions >>> about your own AI/AGI projects as the basis for developing new insights >>> about your theories? They would have to be able to accept the results of >>> their experiments regardless of how well it worked out for them because it >>> just does not make sense to believe that you have it all figured out if you >>> cannot get your ideas to work year after year after year. But even with >>> this aggressive challenge they still could not be bothered. So I decided to >>> try it myself to show them how it might actually work. I predicted that I >>> would be able to get a limited AGI program (I called it AGi) working within >>> a year. I pointed out that some skeptics would not be convinced no matter >>> how the program turned out but that many of my peers (like the less >>> delusionally narcissistic guys in these groups) would be able to see that >>> it was working if I was actually able to get it working. On the other hand, >>> I explained, if I could not get my program working then I would have to >>> accept the results of my experiments and recognize that I did not and do >>> not have it all figured out. I do agree that an experimenter has to be >>> given some leeway and there is always the possibility of a life-changing >>> event interfering with the goal, but if I still have nothing to show after >>> 2 years then I have to accept that there must be something important that I >>> haven't figured out. It was explained to me that the concept of >>> "prediction" is used in different theoretical ways in this group, but I was >>> already aware of that. If you are using "prediction" as a basis of any kind >>> of AI/AGI learning models, then you have to be able to know how to accept >>> the results of an actual experiment to compare it against the predicted >>> experiment. In an advanced AGI models (including the use of Bayesian Nets >>> which I consider to be mundane) the potential for generating new structured >>> relationships based on actual experiments is essential to the effective >>> utilization of the model. This kind of effective utilization is directly >>> related to my personal experience of having new insights crystallize around >>> the predicted event when compared with an actual experiment. My interest >>> though, is not in prediction per se but in how new structured insights >>> crystallize. >>> Jim Bromer >>> >> ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
