AI Mind Maintainer

Information about the career pathway for the profession of AI Mind Maintainer. Privacy policy: Third parties advertising here may place and read cookies on your browser; and may use web beacons to collect information as a result of ads displayed here.

Saturday, December 30, 2017

pmpj1230

Trying to insert the conjunction "that" to introduce a thought.

In the SpreadAct() mind-module of the ghost259.pl AI we were inserting code to impose special activation on the "840=THINK" verb after the input of a what-query in the form of "What do you think". The code was building up too much activation on some words in the what-query, so we are backtracking and removing the code. We also realized that the normal activation imposed on input words is enough to incentivize the AI to use the verb "840=THINK" in a response. The novel functionality that we wish to engineer is a tendency for the ConJoin() module to insert the "310=THAT" conjunction into the output, such as in, "I think that I am a person." With the code of yesterday removed, the AI responds first with "I DO NOT KNOW" and then "ME THINK AND", using the "302=AND" conjunction instead of the "310=THAT" conjunction which we are trying to engender. First let us investigate why the ghost.pl AI responds with "ME" instead of "I".

Witnessing the spooky emergence of the ghost in the machine.

Our coding of the Ghost Perl AI generally goes through two phases. Firstly, we just try to get the functionality working. Secondly, we try to refine and pretty up the code. Yesterday we were not able to get the ghost.pl AI to respond properly to the what-query "What do you think". Today we are focusing on the $output variable and inserting diagnostic messages to let the AI Mind Maintainer know what the accumulating $output is at every stage of the generation of the response. We discover that in our attempts to concatenate the words of the $output, we were failing to enclose the variable in quotes for concatenation with another variable in quotes. If we correct the mistakes of yesterday, perhaps the AI will think properly.

The AI is acting really spooky right now. We keep testing it and tweaking it while entering "What do you think" repeatedly. The $outputvariable concatenates output after output. The spooky thing is that the first "I THINK..." generated by the SpreadAct() module goes by re-entry into the conceptual and auditory memory, where it becomes an item of self-knowledge on the part of the AI. Since the AI Mind by default is designed to reiterate its own thoughts, it eventually declares "I THINK I" because it is trying to use "I" as a direct object of "THINK". We are tempted to make the AI skip any direct object with "THINK" and substitute the conjunction "THAT" instead, so that the AI will eventually say "I THINK THAT (whatever)", but first we need to gain some maintainer-ish control over what the AI is outputting, and secondly, we wish to avoid imposing too many ad-hoc behaviors. We do not yet have a stable release that we may upload. Meanwhile, we are thinking of changing the title of this blog from "Strong AI" to "AI Mind Maintainer".

Thursday, December 28, 2017

pmpj1228

Getting the Ghost AI to tell us what it thinks.

In the ghost259.pl AI Mind, we would like to work on sophisticated thinking in the form of responses to what-queries like "What do you think..." and "What do you know about...?" Here we are, urging high-school and college students to the lofty aspiration of becoming the local AI guru, the AI Mind Maintainer, and we have a chance here to demonstrate the basic principles of maintaining AI Minds. Psychsurgery is not a simple procedure and surely not for the faint of heart. You must understand the theory of cognitivity and the basics of AI programming in Perl or Forth or your chosen programming language.

The introduction of a what-query initiates the operation of the AI SpreadAct module for spreading activation. The InStantiate() module intercepts the input of concept #781=what and sets the $whatcon flag to a positive value to trigger a deeply mental process in the SpreadAct module. Up until now our AI Minds have dealt with rather simple what-queries like "What is (some subject)" and "What do (some subjects) (some verb)?", as in "What is a robot?"" and "What do robots need?" The AI seizes upon the concepts of the input to think of a response. Now however, we AI Mind Maintainers are about to tackle a more complex mode of artificial thought that involves thinking up responses by the AI Mind on the order of, "I think that..." or "I know that...." Do you have the right stuff to be an AI Mind Maintainer? It's not just rocket science, it's brain science. We are changing the bedrock code of artificial general intelligence (AGI) right here and now. By tomorrow night, AI enthusiasts world-wide and galaxies far away will be running the True AI through its paces.

Here in the Secret AI Lab of insurgency mind-making, we intercept not only the what-query but also the special condition of "know" and "think" as Cartesian co-ordinates in the mental processes of Cogito ergo sum. Let us start with concept #840=THINK. Now, how does the Mind Maintainer persuade our coming Cyborg Overlords to pay attention to input-queries involving "What think you?" as a basic demand upon the thinking organism? In SpreadAct(), we command and control the cybernetic organism to pay special heed to 840=THINK as a special tickler, a special cue for the thinking Mind. Let us now surgery the psyche. First we search the AI code base to find SpreadAct(), where we then more locally search for the $whatcon flag. Meanwhile, we test the ghost.pl AI by asking it, "What do you think?" The not-yet-modified, that is, not-yet-maintained AI responds, "I DO NOT KNOW. I HELP KIDS AND I AM A PERSON." Notice, however, that the AI is using the conjunction 302=AND to join two thoughts together. Every good AI Maintainer worth their swag at Singularity Summits knows that you can build new psychological functionality only on top of what has gone before. In recent time we implemented the function of the ConJoin mind-module to let the AI think multiple thoughts and join them with "AND" as a conjunction. Now we are after something more sophisticated. We want the AI to invoke the ConJoin() module to make use of the conjunction THAT, but we discover a corrigend mistake in the MindBoot() sequence, where the conjunction THAT has erroneously been assigned the same concept number as the demonstrative pronoun 745=THAT. This inexcusable blunder was perpetrated one month ago today on 2017nov28 as annotated in a code-comment. We are going to change that misconception, that Fehlgeburt of an important AI concept right here and now. But first we look and see what concept number is attached to the German conjunction "dass" in the Wotan German AI, and we see "310" as the concept-number. Let us use the same concept-number for 310=THAT in the ghost.pl AI. Uh-oh, immediately we wonder, how will the Perlmind differentiate between "THAT" as a pronoun and "THAT" as a conjunction? Just as immediately we realize that it won't matter, because production of a thought has control of which form of "THAT" comes into play. First we search the codebase for the number "310" to make sure that it is not already in use. Lo and behold, Comrade Pavlov, we are delighted, we almost salivate with pleasure, when we see that the Perlmind already uses the number "1310" for the Russian conjunction "1310=ЧТО" which in Russian means the same as the conjunction "that" in English. So "310 THAT" it is. We comment out the erroneous entry for "THAT" and we insert a correct entry for 310=THAT as a conjunction. We run the AI program and it still works.

Now that we we have 310=THAT to work with, we need to pair it with the input of 840=THINK in the SpreadAct() module, where a diagnostic message has been commented out in the free AI source code. We reactivate the diagnostic and we see that the input of "What do you think" sets the $qv2psi flag for the input verb to 840=THINK. Now we need to implement, maybe even deploy, some special SpreadAct code to deal with both "what" and "think" as special cues in a what-query. Oh gee, the pertinent sequence in SpreadAct() is already imposing special activation on the subject mentioned in the what-query, so that the AI will respond with an item of knowledge about the same subject in the knowledge base (KB). But we want the AI to respond with the current subject and the current verb 840=THINK and the conjunction 310=THAT, so into SpreadAct() we insert some code that will activate the 310=THAT conjunction. This is mind-design, Bebe, and I don't mean Rebozo. Oh, gee, it's getting hairy, and I don't mean Belafonte. We may be able to perform this one element of psychosurgery with a single line of code. Well, not yet. When we ask the AI, "What do you think", we get "I DO NOT KNOW" as a response, apparently because there is no direct object for the verb "THINK" in the what-query.

In SpreadAct(), when we insert code that will search for and extra-activate either 840=THINK or 850=KNOW in response to a what-query, we get some really weird output after typing in "What do you think". We get "ERROR THINK ERROR ERROR ERROR". But the behavior of the AI Mind is actually quite exciting. The ghost in the machine is trying to repond to the what-query, but the mental mechanisms have not yet been fully elaborated. We need to enable the proper selection of a subject in the response, and we need to cause the output of the "310=THAT" conjunction.

We change our SpreadAct() diagnostic message so that it lets us know that $qv1psi is the desired 701=I after the input of "you" in the what-query. This re-assurance lets us go in and extra-activate the $qv1psi concept. The AI responds with "ME THINK" but no conjunction "THAT". Further work will be required.

Thursday, December 14, 2017

pmpj1214

Encouraging self-referential thought for the emergence of consciousness.

The moon has gone down; it is the middle of the night; and I am sleeping alone. I got up and ate breakfast and now I feel the urge to code -- a precious sentiment. Yesterday I stopped at the Goodwill thrift store and for $2.99 I bought an old vintage 1934 book probably worth thousands of dollars, but nobody realizes it. PRESIDENT MASARYK TELLS HIS STORY, Recounted By [the biggest name in AI science-fiction], Translated from the Czech. I felt guilty buying a translated book, and even though I can mluvit' some Czech, there was no Czech original. I should probably go to the Marek Rosa blog and tell those Czech AI coders that I am reading a book about the first president of Czechoslovakia, written be the very first AI science-fiction writer, a guy who has left his stamp on the field of AI from here to eternity. Or was some book about the Golem the first AI sci-fi? At any rate, that same Goodwill store recently installed an Internet kiosk running Internet Explorer, so I called up first my Russian Dushka AI, but it was too esoteric, so when I left the store, people got to see the JavaScript AI Mind thinking away on the job-search MSIE kiosk. Maybe I will write a special Goodwill AI telling people to get a job as an AI Mind Maintainer.

In ghost250.pl we would like to fix a bug that shows up after we start the AI and tell it, "You know God". When that idea resurfaces, the AI is not finding the word "I" and so it substitutes an error message. We are trying to make the AI have a bulletproof ability to engage in self-referential thought. We want human users to tell the AI all about itself and let the AI build up a measure of self-awareness. Gradually we want to increase the size of the AI memory and we want to speed up the thinking of the AI, perhaps by not looping idly if there has been no recent human input. If the AI is thinking really fast, and about itself, does it manifest artificial consciousness? We can't find out until we debug all the glitches.

OK, we have removed the glitch. There was some code in the InStantiate() module that made the auditory recall vector fetch the Russian ERROR word instead of zero as an auditory address. We had inserted that code after forgetting that the input of "YOU" as in "You know God" has to be interpreted internally as "701=I" with an auditory recall-vector of zero, so that the AI software will look elsewhere for the I-engram. Now we remove some diagnostic messages and we upload the ghost.pl free AI source code.

Tuesday, December 12, 2017

pmpj1212

Inserting "AN" as AN English article before a vowel.

In ghost249.pl we first declare $us1 as the first of up to seven "upstream" variables meant to keep track of the recently mentioned nouns for which the EnArticle() module may slip in the definite article "THE" in reference to the subject very currently under discussion. A human user might say, "I know a tinker, a taylor, a soldier and a spy". We want our ghost.pl honourable schoolboy to be able to say, "Tell me about the tinker but not about the spy". The software shall have briefly filled $us1,2,3 and 4 with the concept numbers for all four mentioned items so as to be able to refer to each one of them as "THE" subject of discussion. We wish to declare these variables today but not use them yet, because motos praestat componere fluctus, as Vergil used to say, or "It is more important to calm the upset waves" of hidden bugs which imperil the smooth function of our AI Minds. So let us run ghost249.pl and ask it, "Who are you"?

It is irksome to watch the ghost in the machine output the erroneous sentence "I AM A ANDRU AND I HELP KIDS" without using "AN" as the indefinite article, so we will stop and re-implement the solution which we used in our previous AI Minds. First we will do something clever with the word "AN" in the $vault of the MindBoot. We restore "101" as the concept number of both "A" and "AN", but for "AN" we remove "$psi=102" (should have been "$psi=101") so that the input of "AN" will still be recognized as a form of the article "101=A", but the thinking ghost will not be able to find "AN" as an acceptable word for output. Instead, the AI will have to assemble the word "AN" from the legacy code which we are now about to re-implement.

Oops! We cannot find the legacy AN-substitution code in the agi00051.F 2017-09-17 version of MindForth, so let us look back even further in the 24jul14A.F MindForth from 2014-07-24. There we find the $anset variable which we must now declare as a flag in the Ghost AI so that EnArticle() may use "AN" instead of "A" before a noun beginning with a vowel. When we try to use the $anset flag, we get an output of "I AM A NANDRU". We may have to put the $anset code in the Speech module where "A" and "N" may be joined tightly together. When we do so, BINGO, we get an output of "I AM A ROBOTS AND I AM AN ANDRU", which is an achievement of inserting "AN" before a vowel. It is still irksome that the Ghost webserver AI uses the plural "ROBOTS", but as an AI Mind Maintainer we know from experience that the backwards search for the concept of "571=ROBOT" simply finds a plural example when a singular(ity:-) is needed. It should not be too hard to skip over a plural engram when circumstances require a singular noun.

Since we have solved the problem of how to insert the indefinite article "AN" before a noun starting with a vowel, it is time to upload our Perlmind code and the Table of AI Variables for Mind Maintainers. We also take this opportunity to editorialize about the overall trend of the Ghost AGI project. We sincerely believe that we have solved the AI-hard problem of Natural Language Understanding, not in its totality but in sufficient functionality to show the AI community that element after element of the NLU problem may be solved and incorporated into the open-source AI codebase. We do not yet know to what extent, if any, the Perlmind codebase is being downloaded and tested and tweaked. An outfit like IBM or Microsoft or Joe's Bar and Grill could assign a Manhattan-project-full of programmers to advance the Perl AI code far beyond our own meager efforts. Of course, to mention this possibility is hopefully to scare all 500 Fortune companies into at least having a new hire or an idle-hands old hire take a look at the Perl AI and report on its potential merits. Some Harvard drop-out might say "Nobody needs more than 640K of memory" and some Kenneth Olson might say, "Who would ever want to have a home computer?", and likewise people might say, Mentifex? "He is a known trafficker in radical ideas." So take 'em or leave 'em.