Skip to content

Twenkid/Artificial-General-Intelligence-AGI

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 

Repository files navigation

Artificial-General-Intelligence-AGI

Theory, lectures, slides, notes, schemes, diagrams, papers etc.

http://artificial-mind.blogspot.com

  • Лекции и други материали от първия курс по Общ изкуствен интелект в Пловдив, 2010, 2011 г.
  • Lectures and other materials from the world's first university course in AGI in Plovdiv, 2010, 2011 (see below)

English:

Matches and parallel comparisons between "prophecies" in AGI, Philosophy of AI, Digital Universe (Universe as a Computer), Transhumanism/Cosmism, Artificial Life etc. written/published between 2001-2004 etc. in the pioneering e-zine "The Sacred Computer" ("Свещеният сметач"), interviews, blog publications in "Artificial Mind: Todor Arnaudov's Researches" (the old title) etc. writings (theory, ideas, proposals, claims, reasoning, generalisations, texts, conclusions, ...)

There is an enormous amount, I've mentioned some in the blog and in the slides for the AGI course (the ones which I knew about and which existed up to 2010-2011), they keep to grow in late years and especially as I've been exploring the amazing exhaustive youtube channel "Machine Learning Street Talk". I can barely keep up reacting or recording the times I react: "Wait a minute, I've discussed and said that decades ago, it is not a new direction or original!".

One example (for now): A critics of the "Turing test" and proposals about what AGI published in issue #13 of "Sacred Computer", December 2001:

https://www.oocities.org/eimworld/eimworld13/izint_13.html (Check some copy in archive.org if you wish https://www.geocities.org/eimworld/eimworld13/izint_13.html ) for Google translate a copy in a new server: http://eim.twenkid.com/old/eimworld13/izint_13.html )

Google Translate:

Turing test is inadequate, AI doesn't have to imitate or lie that it was a human; too antropomorphized; intelligence development should be measured, not (just) the crystalized knowledge; the initial information matters ("priors"), one of the measures of intelligence is the capability to learn from extremely limited amount of input (a prove that it is possible are the blind-deaf)

"...Let's continue with the topic of Alan Turing, more precisely look at the test of the same name to determine the "abilities" of the machine, whether it can think or not. The setup is as follows: a person conducts two casual conversations (asking any questions is allowed) with another person and with a machine by means of a computer (in Turing's time we talked about a text channel, today we can also think of voice communication). The first person does not know the order in which the man and the machine will "stand" on the opposite side and tries to find out exactly which of the two is the machine. If at the end of the interview the first person fails to correctly identify which was a machine and which was a human, then the machine has successfully passed the test and proved that it can think. The Turing test has many weaknesses. It only works for sufficiently "experienced" AIs. But even they could be easily recognized by the person if he asked them questions related to, for example, their parents and childhood. In order to get out of the awkward situation, MM will have to either lie or have the information about her "childhood" "implanted" in advance. The speed of an overly fast MM will have to be artificially slowed down, as instant responses will betray the machine. Slow MMs who need, say, a minute to answer even the most basic question will also be recognized instantly. The "inhuman" complex musings of another MM will again remind the person that they are not communicating with a being of their own kind... The Turing test has overly "humanized" requirements. You don't need to know everything to think. It is not necessary for the machine to "lie" people that it is a person in order to prove that it is a thinking being! Determining machine thinking capacities could test a machine's ability to learn and process information against the amount of its current knowledge, i.e. to measure not only the absolute intelligence but also the potential of the machine. Thinking does not appear suddenly. It is the result of a development that is also noticeable in humans. Do the first steps and words uttered by the one-year-old child, compared to his helplessness a year ago, not show development? This is exactly what we should look for in the machine - development, thinking is progress, not omniscience. A great number of operating principles of a Thinking Machine can probably be created. Some of them will be better than others, either in terms of simplicity, or in terms of requirements for the speed of the programmable system used, in terms of the amount of initial information in the system, etc. If we want to check how "good" the "reasonable algorithm" we created is we can put the machine in a very "severe" information situation, ie. to let it develop with a maximally narrowed flow of information to the AI. The ability to "get smarter" under such extreme conditions defines the potential of AI. The less data an AI needs to become intelligent, the closer it gets to human-like and to "ideal AI", by which I mean the simplest and shortest possible algorithm that has minimal initialization and needs minimal input to evolve into a Thinking Machine. Other types of "ideal AI" can be specified - needing the least hardware, using the most memory, having the highest relative speed, etc. The human brain can be taken as a starting point. Without sight or hearing, even in the absence of both the basic human senses, man (the brain) can become sentient. An example of this is deaf-blind individuals..."

... Notice for example:

  1. "The Turing test has overly "humanized" requirements." - "Antropomorphizing"
  2. Compare the second part of the text with Chollet's suggestions.

etc.

Original (Bulgarian) " Нека продължим с темата за Алън Тюринг, по-точно да разгледаме едноименният тест за определяне на "способностите" на машината, може ли да мисли или не. Постановката е следната: един човек провежда два непринудени разговора (позволено е задаването на всякакви въпроси) с друг човек и с машина посредством компютър (по времето на Тюринг се е говорило за текстов канал, днес можем да помислим и за гласово общуване). Първият човек не знае реда, в който на отсрещната страна ще "застанат" човека и машината и се стреми да разбере именно кой от двамата е машината. Ако след края на събеседването първият човек не успее да определи правилно кой е бил машина и кой човек, то машината е издържала успешно изпита и е доказала, че може да мисли. Тестът на Тюринг има много слаби места. Той работи само за достатъчно "опитни" изкуствени интелекти. Но дори и те биха могли да бъдат лесно разпознати от човека, ако той им зададе въпроси свързани, например, с техните родители и детство. За да се измъкне от неудобното положение, ММ ще трябва или да излъже, или информацията за "детството" й да бъде "внушена" предварително. Скоростта на прекалено бърза ММ ще трябва изкуствено да се забавя, тъй като мигновените отговори ще издадат машината. Бавните ММ, които се нуждаят например от минута, за да отговорят и на най-елементарния въпрос също ще бъдат разпознати на момента. "Нечовешки" сложните размисли на друга ММ пак ще подсетят човека, че не общува със същество от своя вид... Тюринговият тест има прекалено "очовечени" изисквания. Не е необходимо да знаеш всичко, за да мислиш. Не е необходимо машината да "излъже" хората, че е човек, за да докаже,че е мислещо същество ! При определяне на машинните мисловни капацитети биха могли да се проверяват уменията на машината да учи и да обработва информация, спрямо количеството на нейните текущи знания, т.е. да се измерва не само абсолютната интелигентност, но и потенциала на машината. Мисленето не се появява изведнъж. То е резултат на развитие, което се забелязва и при хората. Нима първите крачки и думи изречени от едногодишното дете, сравнени с безпомощността му преди година, не показват развитие? Именно това трябва да търсим у машината - развитие, мисленето е прогрес, а не всезнайство. Вероятно могат да се сътворят огромен брой принципи на действие на Мислеща Машина. Някои от тях ще бъдат по-добри от останалите било по простота, било по изисквания към бързодействието на използваната програмируема система, по обема на началната информация в системата и пр. Ако искаме да проверим колко е "добър" създаденият от нас "разумен алгоритъм" можем да поставим машината в много "тежко" информационно положение, т.е. да я оставим да се развива при максимално стеснен приток на информация към ИИ. Възможността да "поумнее" при такива изключителни условия определят потенциала на ИИ. Колкото по-малко данни са необходими на един ИИ, за да стане разумен, толкова повече той се доближава до човешкия и до "идеалния ИИ", под което разбирам възможно най-прост и кратък алгоритъм, който притежава минимална начална и се нуждае от минимална входна информация, за да се развие до Мислеща Машина. Могат да бъдат посочени и други видове "идеални ИИ" - нуждаещи се от най-малко апаратни средства, използващи най-пълно паметта, имащи най-високо относително бързодействие и т.н. Човешкият мозък може да бъде взет като отправна точка. Без зрение или слух, дори при липса и на двете основни човешки сетива, човекът (мозъкът) може да стане разумен. Пример за това са сляпо-глухите индивиди."

Compression, hierarchical compositional patterns, general learning algorithm, computer vision is solvable etc.

The skeptics - the mathematician H.Dreyfus and pattern recognition

Dreyfuss dismisses the possibility that artificial intelligence will ever be created: "the simplest organisms are significantly better at pattern recognition than the most complex machines. The delay in the development of machine pattern recognition is not a simple stumbling block, but a brake, a petrification, an upper limit".

This statement is quite old (at least twenty years) and not current. I'm looking at it because I think it's easier to find the general solution to the MM problem, and then look for the "answers" to the particular cases: Some time ago I watched a few short films about a robot in the form of movable eyes-cameras, which can "look" at any moving objects without having any prior information about them. This robot was constructed a few years ago and is controlled by the power of only one personal computer, which simultaneously moves the three-dimensional face on the EIM screen in time with the pointing of the cameras. In terms of skill, this system can rival cats, and that's no small feat. The creation of MM, (which "by design" may not have any capabilities for processing graphical images) should qualitatively change pattern recognition, as well as all other operations that we define as human and difficult to automate. In humans, most objects (of any kind) are not remembered "photographically", but are "retold" in the brain, the most characteristic features of the information objects are recorded, the input information is compressed. In "photographic", "phonographic", "textographic", "stereographic" (spatial) etc. memory, probably only the basic concepts are stored. Human memory is not particularly strong in accurate memorization, and it also takes up a lot of space (in the supposedly "limitless" capacity of the human brain). It is tricky to explain the newly arrived information object with the available information objects. It is simply given a label, and its essence is described by the known concepts using their labels - links to their meaning, containing only an "address". We humans call such remembering "understanding" and "making sense." If the machine thinks, therefore it understands and makes sense, from which it follows that it can itself create a system for "narrating" the images, for recording the most characteristic elements of the information, even if it was not designed for this. A peculiarity of thinking is the solving of tasks, the method of solution of which is not programmed in advance. Therefore, recognizing any images should not present any difficulty to MM, just as we have no difficulty distinguishing people's faces and voices without being specifically trained to do so. MM programming consists of creating the basic learning and self-programming algorithms. The specialized modules for accelerated information processing. objects - sound, visual image, etc. are useful for MM, but are not mandatory for her work. If we make an analogy with EIM, the central processor can perform all operations on graphic objects performed by the graphics accelerator. It can generate sound without auxiliary sound processors, it can replace the modem, it can manage disk and tape drives by itself. Of course, all this will lead to a decrease in overall performance, but nevertheless, the CPU can do all this on its own. We conclude that: The peripheral processors, e.g. for pattern recognition, are useless without a central (Thinking Machine) presence.

Скептиците - математикът Х.Драйфус и разпознаването на образи

Драйфус отхвърля възможността някога да бъде създаден изкуствен интелект: "най-простите организми разпознават образите значително по-добре, отколкото най-сложните машини. Забавянето в развитието на машинното разпознаване на образи не е обикновена спънка, а спирачка, вкаменяване, горна граница".

Това изказване е доста остаряло (поне двайсетина години) и не е актуално. Разглеждам го, защото смятам, че е по-лесно да намерим общото решение на задачата за ММ, и след това да търсим "отговорите" на частните случаи: Преди известно време гледах няколко кратки филмчета за робот във формата на подвижни очи-камери, които могат да следят с "поглед" всякакви движещи се обекти без да имат предварителна информация за тях. Този робот е конструиран преди няколко години и се управлява от мощта на само един персонален компютър, който едновременно с насочването на камерите, в такт движи и триизмерно лице на екрана на ЕИМ-а. По умелост тази система може да се съревновава с котките, а това не е никак малко постижение. Създаването на ММ, (която "по проект" може да няма никакви възможности за обработка на графични изображения) би трябвало качествено да промени разпознаването на образи, както и всички останали операции, които определяме като човешки и трудно се поддават на автоматизация. При човека, повечето обекти (от всякакъв вид) не се запомнят "фотографски", а се "преразказват" в мозъка, записват се най-характерните особености на информационните обекти, входната информация се компресира. Във "фотографска", "фонографска", "текстографска","стереографска" (пространствена) и пр. памет вероятно се съхраняват само основните понятия. Човешката памет не е особено силна в точното запомняне, пък и то заема много място (в уж "безграничния" капацитет на човешкия мозък). Хитро е новопостъпилият информационен обект да се обясни с наличните информационни обекти. Просто му се дава етикет, а същността му се описва с известните понятия, като се използват техните етикети - връзки към значението им, съдържащи само "адрес". Ние хората наричаме такова запомняне "разбиране" и "осмисляне". Ако машината мисли, следователно разбира и осмисля, от което пък следва, че тя може сама да си създаде система за "разказване" на образите, за записване на най-характерните елементи на информацията, дори и да не е проектирана за това. Особеност на мисленето е решаването на задачи, чиито начин на решение не е програмиран предварително. Следователно разпознаването на каквито и да било образи не би трябвало да представлява никаква трудност за ММ, както на нас не ни е трудно да различаваме лицата и гласовете на хората, без да сме специално обучавани за това. Програмирането на ММ се състои в създаване на основните алгоритми за обучение и самопрограмиране. Специализираните модули за ускорена обработка на информац. обекти - звук, зрителен образ и пр. са полезни за ММ, но не са задължителни за нейната работа. Ако направим аналогия с ЕИМ, централният процесор може да извърши всички операции над графични обекти, изпълнявани от графичния ускорител. Той може да поражда звук без помощни звукови процесори, може да замести модема, може да управлява сам дискови и лентови ЗУ. Разбира се, всичко това ще доведе до понижена обща производителност, но въпреки това ЦП може да прави всичко това сам. Стигаме до заключението, че: Периферните процесори, напр. за разпознаване на образи, са безполезни без наличие на централен (Мислеща Машина).

https://research.twenkid.com/agi/2010/

https://research.twenkid.com/agi/2010/Todor_Arnaudov_Theory_of_Hierarchical_Universal_Simulators_of_universes_Eng_MTR_3.pdf

Repo of the book in progress: "The Prophets of the Thinking Machines: AGI & Transhumanism: History, Theory & Pioneers" https://github.com/Twenkid/izkustven-razum-i-razvitie-na-choveka-kniga/

Български:

http://research.twenkid.com/agi/2010/AGI_What_is_Universal_AI_MTR.pdf

http://research.twenkid.com/agi/2010/Artificial-Mind-Blog-Links.doc

http://research.twenkid.com/agi/2010/Artificial-Mind-Blog-Links.htm

http://research.twenkid.com/agi/2010/Boris_Kazachenko_Articles.htm

http://research.twenkid.com/agi/2010/Brain_Architecture_22_4_2010.pdf

http://research.twenkid.com/agi/2010/Cognitive_Algorithm_Scalable_Intelligence_Topics_MTR.pdf

http://research.twenkid.com/agi/2010/Cognitive_Focus_Specialist_vs_Generalist_MTR.pdf

http://research.twenkid.com/agi/2010/Complexity_Probability_Chaos_10-2010_MTR.pdf

http://research.twenkid.com/agi/2010/Developmental_Psychology_30_4_2010.pdf

http://research.twenkid.com/agi/2010/Evolution_Meta-Evolution-Intro_2011_11_Feb_MTR.pdf

http://research.twenkid.com/agi/2010/Intelligence_by_Marcus_Hutter_Agent_14_5_2010.pdf

http://research.twenkid.com/agi/2010/Jeff_Hawkins_memory-prediction_HTM_MTR.pdf

http://research.twenkid.com/agi/2010/Machine_Learning_basics_4-Feb-2011-MTR.pdf

http://research.twenkid.com/agi/2010/matematicheska_teoria_na_razuma_7.doc

http://research.twenkid.com/agi/2010/Narrow_AI_Review_Why_Failed_MTR.pdf

http://research.twenkid.com/agi/2010/Razum_8_cut.pdf

http://research.twenkid.com/agi/2010/Razum_principles_Caffe_Sci_2009_MTR.pdf

http://research.twenkid.com/agi/2010/Reinforcement_Learning_Anatomy_of_human_behaviour_22_4_2010.pdf

http://research.twenkid.com/agi/2010/Todor_Arnaudov_Theory_of_Hierarchical_Universal_Simulators_of_universes_MTR.pdf

http://research.twenkid.com/agi/2010/Whats_wrong_with_NLP_links.htm

И др.

...

This address: https://github.com/Twenkid/Artificial-General-Intelligence-AGI

About

Theory, lectures, slides, notes, schemes, diagrams, papers etc.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published