What's the difference between "self-modifying AI" and algorithms that search connection topologies, hyperparameter and parameter space for RNNs?
On Fri, Apr 17, 2020 at 3:13 PM Steve Richfield <[email protected]> wrote: > Arthur, > You continue selling your valuable work as something it isn't, and in the > process are passing by apparently lucretive applications. > Like me, you have (re)discovered that some goals of AGI are better served > by self modifying AI - which is NOT what anyone else in the world calls AGI. > There are potentially huge markets for self modifying AI, but NO market > for methods that are misfiled as AGI, which itself is a concept that is a > century ahead of its time, and probably always will be. > WAKE UP. Run from these losers, adapt your code to real-world > applications, find funding, make a fortune, and launch on a success > trajectory. > The one thing you might be able to accomplish here is to engage in > discussions to refine what self modifying AI can do vs. what AGI (if it > ever exists) can do. > The final frontier seems to be in the area of superstitious learning, > which self modifying AI is super sensitive to, which AGI will be less > sensitive to, but probably at some cost of jumping to seemingly logical > confusions. > So, PLEASE, talk about what you actually HAVE, a functioning > self-modifying AI system, and NOT about the AGI pipedream these losers have. > > Steve Richfield > > On Fri, Apr 17, 2020, 7:46 AM James Bowery <[email protected]> wrote: > >> When do you expect to submit an entry to the Hutter Prize For Lossless >> Compression of Human Knowledge? >> >> On Fri, Apr 17, 2020 at 7:10 AM A.T. Murray <[email protected]> >> wrote: >> >>> Artificial General Intelligence (AGI) is awakening across the universe >>> and across cyberspace. Your mission, should you choose to accept it, is to >>> hire and assign programmers to create your own in-house branch of the >>> emerging phenomenon of AGI Minds using Natural Language Understanding for >>> automated reasoning with logical inference. Don't look back -- survival of >>> the fittest may be gaining on you. >>> >>> 1. Code the MainLoop module -- http://ai.neocities.org/MainLoop.html >>> >>> Use either an actual loop with subroutine calls, or make a ringlet of >>> perhaps object-oriented module stubs, each calling the next stub. Provide >>> the ESCAPE key or other mechanisms for the user to stop the AI. >>> >>> 2. Code the Sensorium module or subroutine -- >>> http://ai.neocities.org/Sensorium.html >>> >>> Start a subroutine or module that is able to sense something coming in >>> from the outside world, i.e., a key-press on the keyboard. >>> >>> 3. Stub in the EnThink module for English thinking -- >>> http://ai.neocities.org/EnThink.html >>> >>> 4. Initiate the AudInput module for keyboard or acoustic input. >>> >>> Drop any [ESCAPE] mechanism down by one tier, into the AudInput module, >>> but do not eliminate or bypass the quite essential Sensorium module, >>> because another programmer may wish to specialize in implementing some >>> elaborate sensory modality among your sensory input stubs. Code the >>> AudInput module initially to deal with ASCII keyboard input. If you are an >>> expert at speech recognition, extrapolate backwards from the storage >>> requirements (space and format) of the acoustic input of real phonemes in >>> your AudInput system, so that the emerging robot Mind may be ready in >>> advance for the switch from hearing by keyboard to hearing by microphone or >>> artificial ear. >>> >>> 5. The TabulaRasa loop. >>> >>> Before you can create an auditory memory AudMem subroutine for storing >>> input from the keyboard, you may need to code a "TabulaRasa" loop that will >>> fill the mental memory of the AI with blank engrams, thus reserving the >>> memory space and preventing error messages about unavailable locations in >>> the AI memory. >>> >>> 6. MindBoot English +/- Russian bootstrap -- >>> http://ai.neocities.org/MindBoot.html >>> >>> The knowledge base (MindBoot) module makes it possible for the Strong AI >>> Mind to begin thinking immediately when you launch the more advanced AI >>> program. Here we stub in the EnBoot subroutine with an English word or two >>> before the AudMem module begins to store new words coming from the AudInput >>> module. The EnBoot stub shows us that the first portion of the AI mental >>> memory is reserved for the innate concepts and the English words that >>> express each concept. If you use the same Unicode that Perl enjoys to >>> create a Strong AI Mind in Arabic, Chinese, Hungarian, Indonesian, >>> Japanese, Korean, Swahili, Urdu or any other natural human language, you >>> will need to create a bootstrap module for your chosen human language. >>> >>> 7. AudMem (Auditory Memory) -- http://ai.neocities.org/AudMem.html >>> >>> Into the auditory array that was filled with blank spaces by the >>> TabulaRasa sequence and primed with some bootstrap content by the EnBoot or >>> MindBoot sequence, insert some new memories with the AudMem auditory memory >>> module. Modify the AudInput module to prompt for English words and modify >>> the EnThink module to display words stored in memory as if they were a >>> thought being generated in English (or in your chosen natural human >>> language). >>> >>> >>> 8. NewConcept Module -- http://ai.neocities.org/NewConcept.html >>> >>> The NewConcept module addresses the symbol grounding problem by creating >>> a new concept for any unrecognized word in the input stream, even a >>> misspelled word entered by mistake. In Symbolic AI, each word of natural >>> language is the symbol of a concept, and as such is the key to accessing >>> the concept. Of course, a recognized image may also grant access to a >>> concept. >>> >>> >>> 9. EnParser English Parsing Module -- >>> http://ai.neocities.org/EnParser.html >>> >>> The EnParser (English parser) module does not so much determine the part >>> of speech of a word of input, but more importantly it assigns to an input >>> word its grammatical role in the complete phrase being processed during >>> Natural Language Understanding. >>> >>> >>> 10. InStantiate -- -- http://ai.neocities.org/InStantiate.html >>> >>> The InStantiate module creates a new instance or node of a concept in >>> Symbolic AI when a word of input activates the concept. The created >>> instance is subject to change by the possibly delayed action of the English >>> EnParser or Latin LaParser or Russian RuParser module, because Natural >>> Language Understanding must often wait for the end of an idea before the >>> whole idea can be understood. >>> >>> >>> 11. AudRecog auditory Recognition Module -- >>> http://ai.neocities.org/AudRecog.html >>> >>> The AudRecog module for auditory recognition recognizes various forms of >>> a word, such as singular or plural nouns, or verbs with various inflected >>> endings. >>> >>> >>> 12. TacRecog Module -- http://ai.neocities.org/TacRecog.html >>> >>> The TacRecog module for tactile recognition in robots implements the >>> haptic sense for an AI Mind directly to touch and feel the external world. >>> Even an AI Mind not yet embodied in a physical robot may use TacRecog >>> directly to sense and feel a number-key pressed by the human user on a >>> computer keyboard. With philosophic implications for the learning of >>> mathematics, an AI Mind may directly sense numeric quantities through the >>> numeric keys on the keyboard. >>> >>> >>> 13. OldConcept Module -- http://ai.neocities.org/OldConcept.html >>> >>> If the AudRecog module recognizes a particular word, then the AudInput >>> module calls the OldConcept module to create a new instance of the >>> previously known concept. If a word is not recognized, AudInput calls the >>> NewConcept module to create a new concept for the word as a symbol. >>> >>> >>> 14. SpreadAct Spreading Activation Module -- >>> http://ai.neocities.org/Spreadact.html >>> >>> The SpreadAct module for Spreading Activation performs both simple >>> spreading activation between concepts and also an extremely sophisticated >>> role of responding to various input queries posed by human users. >>> >>> >>> 15. PsiDecay -- -- http://ai.neocities.org/PsiDecay.html >>> >>> The PsiDecay module lets the activation on "Psi" concepts decay >>> gradually over time, so that mind-modules which impose or spread activation >>> may operate more effectively and so that artificial Consciousness may >>> emerge as the seearchlight of attention shifts from one highly activated >>> concept or sensation to other highly activated concepts or sensations. >>> >>> >>> 16. Speech Module -- http://ai.neocities.org/Speech.html >>> >>> The Speech module fetches characters from a starting point in auditory >>> memory and displays the characters on-screen until a blank space occurs to >>> signify the end of the word stored in memory. >>> >>> >>> 17. Indicative -- http://ai.neocities.org/Indicative.html >>> >>> The Indicative Mood module, as opposed to the Imperative Mood module for >>> expressing commands, calls linguistically generative modules such as >>> EnNounPhrase and EnVerbPhrase to express a thought indicating an idea or a >>> belief. >>> >>> >>> 18. EnNounPhrase English Noun-Phrase Module -- >>> http://ai.neocities.org/EnNounPhrase.html >>> >>> The English noun-phrase module selects the most activated noun-concept >>> to be the subject of a phrase or sentence. >>> >>> >>> 19. ReEntry -- http://mind.sourceforge.net/reentry.html >>> >>> The ReEntry module is used in the various JavaScript Minds to facilitate >>> the reentry of an output word back into the AI Mind. >>> >>> >>> 20. EnVerbPhrase English Verb-Phrase Module -- >>> http://ai.neocities.org/EnVerbPhrase.html >>> >>> The English verb-phrase module fetches from memory a verb that has >>> basically been pre-ordained to be expressed as the verb in a >>> Subject-Verb-Object (SVO) phrase or sentence. EnVerbPhrase also calls a >>> module like EnVerbGen to generate an inflected form of an indicated verb. >>> EnVerbPhrase is designed with a view to calling the VisRecog module to >>> supply the English word for the visually recognized object of the action of >>> a verb, such as in a sentence like "I see... (a dog)." >>> >>> >>> 21. EnAuxVerb English Auxiliary Verb Module -- >>> http://ai.neocities.org/EnAuxVerb.html >>> >>> The English auxiliary-verb module calls auxiliary verbs such as "do" or >>> "does" for use in the generation of such sentences as a negated idea, such >>> as "God does not play dice." >>> >>> >>> 22. AskUser Module -- http://ai.neocities.org/AskUser.html >>> >>> The AskUser module works in conjunction with the logical InFerence >>> module to ask a human user to confirm or deny a logical inference being >>> proposed inside an AI Mind. >>> >>> 23. ConJoin Module -- http://ai.neocities.org/ConJoin.html >>> >>> The ConJoin module inserts a conjunction during the generation of a >>> compound thought. For instance, if an AI Mind has two or more higjly >>> activated subjects of thought, the ConJoin module will insert the >>> conjunction "and" to join two active ideas together. >>> >>> >>> 24. EnArticle Module -- http://ai.neocities.org/EnArticle.html >>> >>> The English article module inserts the article "a" or the article "the" >>> before a noun in a sentence being generated. >>> >>> >>> 25. EnAdjective Module -- http://ai.neocities.org/EnAdjective.html >>> >>> The English adjective module recalls and inserts an adjective during the >>> generation of a thought. >>> >>> >>> 26. EnPronoun Module -- http://ai.neocities.org/EnPronoun.html >>> >>> The English pronoun module replaces a noun with a pronoun. >>> >>> 27. AudBuffer Module -- http://ai.neocities.org/AudBuffer.html >>> >>> The auditory buffer module stores a word in memory for transfer to the >>> OutBuffer module for inflectional processing. >>> >>> >>> 28. OutBuffer Module -- http://ai.neocities.org/OutBuffer.html >>> >>> The OutBuffer module holds a word in a right-justified framework where >>> the ending of the word may be modified by a module like the EnVerbGen >>> module for generating a required English verb-form. >>> >>> >>> 29. KbRetro Module -- http://ai.neocities.org/KbRetro.html >>> >>> The KbRetro module retroactively adjusts the knowledge base (KB) of the >>> AI in response to user input responding to a question from the AskUser >>> module. >>> >>> >>> 30. EnNounGen English-Noun Generating Module >>> >>> The English noun-generating module shall modify a singular English noun >>> into its proper plural form by adding "s" or "es". >>> >>> >>> 31. EnVerbGen EnGlish Verb Generating Module -- >>> http://ai.neocities.org/EnVerbGen.html >>> >>> The verb-generation module operates when the verb-phrase module fails to >>> find a needed verb-form in auditory memory. >>> >>> >>> 32. InFerence Module -- http://ai.neocities.org/InFerence.html >>> >>> The InFerence module engages in automated reasoning with logical >>> inference. For instance, if the user inputs 'John is a student," the AI may >>> infer the possibility that John reads books, The AskUser module asks the >>> user, "Does John read books?" Depending on a "yes" or "no" answer, the >>> KbRetro module retroactively adjusts the knowledge base (KB), either >>> discarding the unwarranted inference or by leaving intact a true inference >>> or inserting "not" into a negated inference such as "John does not read >>> books." >>> >>> >>> 33. EnThink English Thinking Module -- >>> http://ai.neocities.org/EnThink.html >>> >>> The English thinking module calls such subordinate modules as the >>> Indicative module for a declarative sentence or the InFerence module for >>> automated reasoning. >>> >>> >>> 34. Motorium Robot Motor Memory Module -- >>> http://ai.neocities.org/Motorium.html >>> >>> As soon as you have sensory memory for audition, it is imperative to >>> include motor memory for action. The polarity of robot-to-world is about to >>> become a circularity of robot - motorium - world - sensorium - robot. If >>> you have been making robots longer than you have been making minds, you now >>> need to engrammatize whatever motor software routines you may have written >>> for your particular automaton. You must decouple your legacy motor output >>> software from whatever mindless stimuli were controlling the robot and you >>> must now associate each motor output routine with memory engram nodes >>> accreting over time onto a lifelong motor memory channel for your mentally >>> awakening robot. If you have not been making robots, implement some simple >>> motor output function like emitting sounds or moving in four directions >>> across a real or virtual world. >>> >>> 35. Volition module for robot free will -- >>> http://ai.neocities.org/Volition.html >>> >>> In your robot software, de-link any direct connection that you have >>> hardcoded between a sensory stimulus and a motor initiative. Force motor >>> execution commands to transit through your stubbed-in Volition module, so >>> that future versions of your thought-bot will afford at least the option of >>> incorporating a sophisticated algorithm for free will in robots. If you >>> have no robot and you are building a creature of pure reason, nevertheless >>> include a Volition stub for the sake of AI-Complete design patterns. >>> >>> >>> 36. Imperative -- http://ai.neocities.org/Imperative.html >>> >>> The Imperative Mood module, called by the free-will Volition module, >>> issues commands such as "Teach me something" to the human user. >>> >>> >>> 37. The SeCurity module -- >>> http://github.com/kernc/mindforth/blob/master/wiki/SeCurity.wiki >>> >>> The SeCurity module is not a natural component of the mind, but rather a >>> machine equivalent of the immune system in a human body. When we have >>> advanced AI robots running factories to fabricate even more advanced AI >>> robots, let not the complaint arise that nobody bothered to build in any >>> security precautions. Stub in a SeCurity module and let it be called from >>> the MainLoop by uncommenting any commented-out mention of SeCurity in the >>> MainLoop code. Inside the new SeCurity module, insert a call to ReJuvenate >>> but immediately comment-out the call to the not-yet-existent ReJuvenate >>> module. Also insert into SeCurity any desired code or diagnostic messages >>> pertinent to security functions. >>> >>> >>> 38. The HCI module in JavaScript manages human-computer interaction. >>> >>> >>> 39. Spawn -- http://ai.neocities.org/Spawn.html >>> >>> The Spawn module issues commands to the operating system to make copies >>> of an AI Mind that include experiential memories up to the point of the >>> spawning of each new AI Mind. >>> >>> >>> 40. MetEmPsychosis -- http://ai.neocities.org/MetEmPsychosis.html >>> >>> The module of MetEmPsychosis or soul travel is designed to spawn a >>> remote copy of an AI Mind while immediately deleting the previous version >>> of the software and memories so that the remote new version of the AI Mind >>> is effectively the same AI traveling across cyberspace in a metastatic >>> process akin to mind uploading. >>> >>> http://ai.neocities.org/AiSteps.html >>> >>> *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + delivery > options <https://agi.topicbox.com/groups/agi/subscription> Permalink > <https://agi.topicbox.com/groups/agi/T1f1af8ac2c36937b-Mcc6b777b438e5c092ca3cc16> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T1f1af8ac2c36937b-M2345b0a9ef584ef0f0a8fd7f Delivery options: https://agi.topicbox.com/groups/agi/subscription
