Showing posts sorted by date for query Lovelace. Sort by relevance Show all posts
Showing posts sorted by date for query Lovelace. Sort by relevance Show all posts

Tuesday, December 25, 2018

// // Leave a Comment

Developmental Approach to Machine Learning? - article by L.Smith and L.Slone - Agreed


Yes, agreed. A good read, suggesting developmental machine learning, spatio-temporally continuous input data etc.:

See the concept of “shape bias" from Developmental psychology. That's related to  discussions in the "AGI Digest" on recognition of "buildings, chairs, caricatures" ... and other articles from this research blog, regarding 3D-reconstruction at varying resolution/detail as one of the crucial operations in vision published in this blog and the general developmental direction which is driven from one of the very fist articles here about the "Educational test".



HYPOTHESIS AND THEORY ARTICLE

Front. Psychol., 05 December 2017 | https://doi.org/10.3389/fpsyg.2017.02124

A Developmental Approach to Machine Learning?

  • Department of Psychological and Brain Sciences, Indiana University Bloomington, Bloomington, IN, United States


  • See also:

Read More

Thursday, January 4, 2018

// // Leave a Comment

The lack of operational hierarchical structure in the Deep Learning ANN neural networks


A survey paper on the issues of Deep Learning by Gray Marcus: https://arxiv.org/ftp/arxiv/papers/1801/1801.00631.pdf

The author has a valuable mix of expertise both in the ANN development and in linguistics, developmental psychology, cognitive psychology.

There are good points on the lack of real hierarchical structure in (current/regular) DL/ANN, accenting that they are actually "flat", even though there are "layers" which gives a confusing impression.

"To a linguist like Noam Chomsky, the troubles Jia and Liang documented would be
unsurprising. Fundamentally, most current deep-learning based language models
represent sentences as mere sequences of words, whereas Chomsky has long argued that
language has a hierarchical structure, in which larger structures are recursively
constructed out of smaller components"

G.Marcus p.9

See also Jeff Hawkins point since 2004 "On Intelligence"/Numenta, Dileep George's Vicarious, Boris Kazachenko's "Cognitive Algorithm"; the old Hierarchical Markov Models; probably many other researchers, also myself since my early 2000s writings where even I as a teenager have realized that human general intelligence faculty is a Hierarchical simulator and predictor of virtual universes.

The ANNs (without being put in another system/organization) lack operational structure.

Good survey and discussion of areas where DL fails and emphasis of the lack of transfer of learning, i.e. that the networks are not general intelligence and don't "understand" the concepts (the "overattribution" for DeepMind's Atari-player discovery of "tunnels", see p.9)



* However I don't like the pretentiousness in some parts of the article while discussing trivialities and proposing alternatives with 15+ years(?) delay with a pinch of academic glamour or so.

E.g. unsupervised learning (not that common boring classification of fixed images) and self-organization, incremental complexity/"self-improvement" - "Seed AI"; hierarchical operational structure, "symbol grounding" - emergence of generalizaions/"symbols", "abstract thought" from the sensory processing; different levels of abstraction - including "symbolic"; causality understanding (prediction, simulation of "virtual universes"); general/universal game playing; application of general educational tests/measures... (Since AGI is about that; the term "human level (general) (artificial) intelligence" was used in the past) etc. (Not just "pattern matching" of synthetic static tests.)

The above is what AI was always supposed to be about - AGI, - at least as some talented teenagers and others realized and shouted it to the world in the early 2000s and dismissed the poisoned term "AI". Everything was called "AI" back then - somewhat similar today, AI is ubiquitous, yet not general and lacking a personal wholeness.

These suggestions and conclusions would be informative for hard-core AI-er, though (programmers-mathematicians type), it seems the "general"-... part has still a way to go as a concept for the "mainstream" developers community with its "Narrow AI" attitudes**.

...

** Narrow AI - another forgotten term, which on second thought is still actual. Current DL is in fact "narrow AI", each network is trained for a specific class of problems ("classification") and as well explained in the paper can't generalize concepts and transfer the knowledge to different domains.

*** I "don't like" my own pretentiousness, too, but I consider it funny and ironic, rather than serious like in the paper. :P

**** Thanks to Preslav Nakov for sharing the link!

...

Compare the educational test proposals with one of the first articles in this blog, a decade ago:

Wednesday, November 14, 2007
Faults in Turing Test and Lovelace Test. Introduction of Educational Test.
https://artificial-mind.blogspot.bg/2007/11/faults-in-turing-test-and-lovelace-test.html

I didn't explicitly defined the exact kinds of tests, because they were already given in details in the appropriate textbooks about the set of the respective expected skills and knowledge for the respective age or educational level.

...

The article reminds me of the series of articles "What's wrong with Natural language processing?", starting from the year 2009:

https://artificial-mind.blogspot.bg/search?q=what%27s+wrong+with

(...)

Vicarious' demo video summarizing ANN reinforcement learning faults, and their Schema Networks:





Read More

Monday, June 9, 2014

// // 1 comment

On the "Passing of the Turing Test" and the related Emotional and Social-Ranking-Authorities nonsenses

Regarding the recent news and the wrong tone, interpretation, focus:

http://www.forbes.com/2009/06/18/turing-test-reading-opinions-contributors-artificial-intelligence-09-warwick.html
http://www.sfgate.com/technology/businessinsider/article/A-Computer-Has-Reportedly-Passed-The-Turing-Test-5537703.php
http://www.forbes.com/2009/06/18/turing-test-reading-opinions-contributors-artificial-intelligence-09-warwick.html
http://longbets.org/1/  (notice the "emotions" and how they talk about them as something "mystical", while they are the dumbest the most obvious and not mystical)
http://www.telegraph.co.uk/technology/news/10884839/Computer-passes-Turing-Test-for-the-first-time-after-convincing-users-it-is-human.html


See  a commentary of mine  (in the context of AGI list) or below:
See Also Ben Goertzel's comments in H+: http://hplusmagazine.com/2014/06/09/what-does-chatbot-eugene-goostmans-success-on-the-turing-test-mean/


Published also in the AGIRI list:
Notes, 14.1.2025: "impersonated" should be "personal" in many uses, however sometimes the original word applies as well, suggesting that the "personal" part is faked.

... Todor Arnaudov's comment on the reported "passing" of the Turing test
Turing test is superficial, based on wrong "social" settings, and somewhat a "lottery".
It's rather "political" than epistemological/gnoseological/cognitive and has wrong hyporcitic goals to fool somebody and to seek for acknowledgment or mercy from an impersonated authority, having some social ranking and perhaps physical power that is higher, and to whom is given the right to decide whether the "candidate" is "elligible" to be given a particular "order/medal/title", that is - particular political rights, social ranking, social status.
This is societal nonsense, a reflection of what most people care about - social ranking, social ranking and at last - social ranking. Well, also about "emotions" -  another among the structurally dumbest things, which is present in all kinds of living beings all the time, and even in the youngest ones, and is expressed in a few bits, thus it is the most platitudinous/banal and obvious.
That includes the high-ranked scholars who perhaps don't have a clue about the analysis above, but as high-ranked individuals usually are, firstly, inclined towards social hierarchy nonsenses, and later on - additionally trained, as ones who are at the high ranks - they use to fight to keep the status quo which is in their favour.

The above holds for humans vs AGI in general. The humans, even the dumbest ones, put their head under the common hat of "humans" or "human race", and they have to declare something. "Humans are creative" or "intelligent" - even though that they as individuals are not, and are banal and mediocre like the "emotions", which are obvious, a solved problem in any well written literary work, dramatic piece, movie script etc.

In two words: Political bullshits, sophistry and mass-deluding nonsenses.

Even without deeper analysis, the verbs in the gushy articles regarding the Turing test are silly by themselves:

*  "convincing judges":  what is it, rhetorics? a trial? Or sophistry, lie, delusion?*  "fooling users that it was a human"  - lying, delusion

Indeed, that phenomenon is a reflection of the sick values of society. In the concept of "social intelligence", the hypocrisy, the ability to delude, to manipulate, to lie and to exploit others is of high value, and the fact that the society pretends that it "doesn't like hypocrisy", that "liying is wrong" is a part of the sick hypocritical nature of humans, especially the ones who are higher in the hierarchy.

(And humans believe they are higher than animals, machines etc., thus they have "their rights to decide" - instead of objective and obvious measures which do not require impersonated judges.)


True AGI is obvious

That is what has to be achieved - results that are obvious. Not ones that requre a "qualified" "judge" and it should be so from the beginning.

It's trivial to recognize a machine if it really is Versatile Limitless Self-Improver/AGI, and it doesn't lie and is not artificially slowed down and if it is sincere, self-aware [knows about itself, trully - a system that pretends that is was 13-year old boy, 15-year old... dog etc. is not self-aware, unless if it knew that it was pretending and was playing].

Also if it's too slow, too smart, too quick, too deep etc. it will be "different", compared to a human.

I myself am working on a machine that is to FAIL this silly test without taking it.
I guess I will fail it myself as a "non-human" - depends on... the randomness of the jury and how convinced they were of the possibility that a human who would compose such long and complex sentences, if she wished, would be selected to participate in such a test. 

The exact level of intelligence (compared to particular person/s, age, education, social group etc.) in fine-grained terms and fine-grained measures in each particular domain, and in all combined should be rather obvious from the performance of the system without any artificial settings, as mentioned in 0.1.1.266 and later - everything is "a test of intelligence"  should be obvious directly, especially as "intelligence" is a vague term "without concensus on it", aggregating all kinds of perception, prediction, planning, decision making, search, goal-seeking behavior, reinforcement learning capabilities, generalization and specification, increase of the resolution of perception and causation etc. etc.

A test is needed when it is not obvious, when it's somewhat brittle, unknown, BUT some authority has to "approve" it, to "allow it", a pretentious pedantry like the texts in the patents or the following joke:

We, the jury, declare, that the applicant #124353485943, after passing the Turing test with 33.467%, is the first artificial intelligence! Now according to the deputy of ministers of the Royal society of scientific council of the international organization of ABCD, that has leading scholars from the top universities and research groups in the field, we allow that this machine can be called "artificial intelligence" and we proclaim to call this date the birth od the true artificial intelligence. Amen!
By the way, that reminds me of the "birth" of the AI, as it was solemnly proclaimed to be the Dartmouth conference in 1956 - because some high ranked individuals from high-ranked Universities gathered and decided "this is the new science".


See:
See 0.1.1.203: "Man and Thinking Machine (...) " T.A.,  http://razumir.twenkid.com/kakvomu_notes.html#203 (2001, Bg)

See 0.1.1.266: "Faults in Turing Test and Lovelace Test ..." T.A http://razumir.twenkid.com/kakvomu_notes.html#266   - "http://artificial-mind.blogspot.com/search?q=Lovelace (2007, En)

Regarding the sick,and ill defined concept of the society for "intelligence"
I suggest my latest major super multi-inter disciplinary sociological-philosophical-
artistic-literary-satirical-many others-and-some specifically about AGI-and consciousness- work "What a man needs? Part I - Nice guys finish last!", from the first issue of the new e-zine called Razumir (however in this issue - in Bulgarian):    http://razumir.twenkid.com/

It's a complete book of several hundred pages, has many independent and interacting "processes" and many threads of thought. One of the threads that is related to the Turing test besides that superficial social-ranking business is another one from that yard - the subordination and the obedience to impersonated beings are seen by society (the higher ranked and the slaves) as "intelligence". 
That's one major reason why humans believe that dogs are "intelligent" (where more appropriate words are also  obedient to the master, willing to subordinate/please, willing to seek for approval for their actions, dependent on the master, slavish...). "Intelligence" here is a hypocritical term, similar to the hyporcitical pretended prohibition of lying as a "sin" or "immoral", where the higher in society a person grows, the more he lies/deludes/manipulates/is hypocritic.
Humans use to hate or ignore the cognitively more intelligent ones - except if the latter had clearly higher social ranking, were your firends/somehow you identify and take his positive qualities as your own.

The higher ranked ones are "smarter" because of their ranking - they are "qualified". The clueless have nothing but to cite the "qualified", since the former cannot discriminate by their own thinking the nonsense and adopt the authority.

Dogs are complimented as intelligent only because they are servants and are actually dumb enough. Here goes also humans' intrinsic love of slavery - fixated in the culture of thinking machines from the beginning - Chapek's robots and Asimov's positronic slaves with their naive inborn "morality" created to please slave-owners.

According to society (and the not very cognitively/epistemologically intelligent ones), your master (or God, superior, boss) should decide even about your intrinsic qualities - because you yourself can't know - most people do not know. Thus everything is converted into "social ranking" and "rights", the only intelligible or acceptable measures for the majority of people. A monkey/primate thing...

And there we are back to the Turing test -  only a glorious jury, ellected by he Royal Committee of the Lord of Old-Willinshere North Dillignshane, Knight of the Lord of the... Bullshit - should decide whether a machine "has intelligence", because it passes Turing test - the value of which is approved because "Turing was one of the pioneers of the new field, contributor of.. amazing, extraordinary.":

- missing that that's the first test, one of the pioneers... - Turing himself may have denied it if he could live longer; that was what first came to his mind, he didn't have time to elaborate it, the noble society where he lived took it away

- the test is perhaps biased by the society where he lived with its particular manner of social ranking relations/academic "approval" style

- missing the nonsenses of it (which come out naked if one understands it)

The fixation of the first inarticulate and naive test as a standard, by the way, is another illustration of the "authority-obeying-citation" sickness in society, including the speculative sciences - philosophy was in that role some time ago.

Talking and exaggerating bombast nonsenses, without understanding that they are superficial or confused and avoiding questioning them, because "they were obvious"; actually - because they were approved by the unquestionable authority.
In sensory-motor generalization terms that is - lacking the facculty of judgement - as my "brother in mind" and his predecessor called the capability to connect abstract and concrete (regarding the terms of the translators in English).

That nonsense has deeply poisened the world - Europe was in the darkness of partial or complete mental retardation and madness for some 1000-2000 years due to such blind wrong undoubted "truths" coming from power-seeking "socially intelligent" social-hierarchy-climbing religious ideologists and rulers, whose concepts of "intelligence" was "social intelligence", master-slave relations; that is: political, social-hierarchy-related and "emotional", instead of epistemological and cognitive - "rational", of the Reason, related to "the faculty of judgment", "Objective" (in terms of as translated in English in the works of Kant and Schopenhauer).

* Someone may suggest "ethical" among "political" in the last sentence - not really, because the cognitive intelligence includes also ethics. Epistemological philosophy in its best is often  connected to Pantheism or is Atheistic, in the sense that "Theistic" religious, the monotheistic, are related to some unquestionable impersonated ruler and the slave should obey the master and has no right to ask questions - or he will be punished by "God" - the ones who usurped the authority, the power, the rights.

My philosophical brother of mind Arthur Schopenhauer has elaborated on this topic before myself, Nietsche has written on that as well.

Ethics as well can be primitive and superficial or deep and elaborate. Simple "emotions" and social-ranking relations of subordination because somebody owns a higher position - with no questions why it has a higher position and should  he -  require less sophisticated cognition than the critical version. Recognition of the ranking is enough - such as a crown of a king, different clothes, different colours, bigger size; backed up with some sort of application of force at a given point in time - punishment and reward.

Dogs and the dumbest people are clever enough to "understand" such kind of "ethics", while the individuals with the highest cognitive/epistemological intelligence often have "low" intelligence in this vicious and corrupted sense - even though they clearly demonstrate that the nonsense of society is far more clear to their minds, they can predict behavior and they know why things are as they are and thoroughly understand the society and humans. However they deny to obey, which is the major dog-like way to express "intelligence" in such silly ethical (political) systems.

=== Todor "Tosh" Arnaudov ===

.... Author of the world's first University courses in AGI  (2010, 2011): http://artificial-mind.blogspot.com/2010/04/universal-artificial-intelligence.html
... Todor Arnaudov's Researchers Blog: http://artificial-mind.blogspot.com
Read More

Tuesday, January 7, 2014

// // Leave a Comment

What's Wrong with NLP, Part 4 - an Opinion from a Paper at IJCAI-13 and a Comment by Todor

Why can't computer understand me.. 

On our best behaviour - Hector J. Levesque

 http://www.cs.toronto.edu/~hector/Papers/ijcai-13-paper.pdf  -

I'd generalize what the author tries to say: yes, the machine does need *imagination* (sensory memories and spaces, see many, for example [7]) and to incrementally learn and play with it. That subsystem is required!* I do agree also that many AI-ers, especially in NLP community didn't understand or accept that, this paper suggests that there's no common progress in the acceptance.

Todor:

Some of the reasons are simply the academic and people's urge to produce papers and "results" quick, to demonstrate that you "do something". Papers, papers, papers - who cares about the real progress.

I guess some have understood that very well, the "common sense" problem is cited from decades, it is obvious, yet in the NLP they've been insisting to push a bunch of words with no relation to the real world and then ask "why it can't do proper word-sense disambiguation, given only a corpus?"

Also, IMO it's  the researchers in NLP etc. who don't understand language understanding [4],  rather than the poor computers, which, as many AI-use to say "do what they are told to".

Language teaching of a machine should be done incrementally with sensory-motor feedback and interaction, like teaching a child and not in sensory-less batch mode with a huge corpus with zero interaction, no coordinate space, just a bunch of words, because:

Todor: Natural language is a hierarchical redirection/abstraction/generalization/compression of sequences of multi-modal sensory inputs and motor outputs, and records and predictions for both.

A mere corpus has no imagination, even if it's 1 Terrawords. A small corpus, built by systematic interaction and mapped to sensory-motor system is intelligent and can explore and learn further on its own. Such as - the toddlers.

I support also the criticism of the Turing Test. Right, it is a test for fooling people - not a test of intelligence, and as I myself claimed back in 2001 (see [5] "Човекът и мислещата машина..." below) - a true, honest and too smart or quick AGI will quickly FAIL the test, because its output is too complex, too quick, too deep, she doesn't have some required memories (shuld lie for her childhood etc.) etc.

Finally, I will save some of my thoughts from myself...

Credits to Aaron for sharing the link!

* More about that when I complete the first milestones of mine...

See also: (And many others...)


[1] Faults in Turing Test and Lovelace Test. Introduction of Educational Test (2007)

[2] What's wrong with Natural Language Processing? Part 2. Static, Specific, High-level, Not-evolving... (2009)

[3] What's wrong with Natural Language Processing? Part I (shorter) (2009)

[4] Part 3: The NLP Researchers cannot understand language. Computers could. Speech recognition plateau, or What's wrong with Natural Language Processing? (2010)

[5] Анализ на смисъла на изречение въз основа на базата знания на действаща мислеща машина. Мисли за смисъла и изкуствената мисъл (2004)

[6] Човекът и мислещата машина. Анализ на възможността (...), 2001 г.

[7] Embodiment is just coordinate spaces, interactivity and modalities - not a mystery (2011)



Read More

Monday, February 20, 2012

// // Leave a Comment

Philosophical and Interdisciplinary Discussion on General Intelligence, AGI and Superintelligence Safety and Human Moral | Cognitive Origins of the Concepts of Human Soul and its Immortality | Free Will and How it Originates Cognitively | Animate Being and Soul and the Cognitive Reason for the Believe that "Thinking Machines can't have a Soul and Consciousness" | Technology Making us more Humane | The Egoism of Humanity | And more

By Todor "Twenkid/Tosh" Arnaudov
http://research.twenkid.com/wp/

* For the followers of the blog - that's not the paper announced in the previous post.


The expression and sharing of the following reflections and speculations is inspired by a publication for a position offered at the FACULTY OF PHILOSOPHY, Oxford University, published on the AGI List (thanks to Sean O'Heigeartaigh).

http://www.fhi.ox.ac.uk/get_involved/future_tech_vacancies/futuretech

See the pdf document, the research foci on p. 2, section 5:

http://www.fhi.ox.ac.uk/__data/assets/pdf_file/0014/25034/Futuretech_Tamas_RF_230112.pdf 

5. This is a research position focused on topics related to the long-term future of machine intelligence and AI. Relevant research foci include: 
- Studying paths towards strong AI: Monitor the current state of progress in the field; identify milestones, roadmaps, etc. 
- Machine intelligence: Analyze fundamental concepts—e.g. how to define and measure general intelligence in artificial systems; how to distinguish different kinds of goal-seeking systems.
- Self-modifying systems: What can be proved about the performance and capabilities of different kinds of recursively self-modifying programs? Can a framework be developed in which demonstrably safe, recursively self-improving AIs could be constructed, with stable and human-friendly goal systems? 
- The role of big data: What impacts and applications will arise from the increasing availability of enormous data sets? What are the fundamental tradeoffs between processing power and data? Is there a minimum amount of data that an arbitrarily powerful intelligence would need in order to effectively deal with various task domains? Can one quantify how much information the human brain contains at birth to enable it to develop general intelligence? Can one quantify how “difficult” it was for such a system to evolve? 
- Philosophy of computer science and of AI: What does it mean to implement a computation? Would a “Boltzmann Brain” or a “Swampman” be conscious? 
- The control problem: How could one ensure that an artificial superintelligence would be safe and beneficial? 
- Can one analyze classes of possible utility functions and construct general statements about the behavior of expected utility-maximizing superintelligence based upon utility functions of a particular class?

Sounds familiar - I have already worked on and have been studied the topics listed in section 5 in the pdf document since the "teenage theory" era, my ultimate aim is at understanding general intelligence and building an eventually developing an AGI system, thus my intent when working on related philosophical issues is to put the  speculations and explanations in coherent computational/theoretical form aligned to a theory of GI to the general/cybernetic/systems theory trends in Universe evolution as I see them (a sort of cybernetic metaphysics).

I also aim to match my theoretical speculations and insights with all accessible kinds of neuroscientific, evolutionary, cognitive science, behavioral etc. evidences I'm aware of at the time. Often I discover such evidences later. For example ~10 years ago, while not being familiar with neuroscience as now, appropriate reflections and a basic understanding in developmental psychology allowed me to induce the existence and functional distinction of hippocampus and the neocortex using only behavioral evidences. See: http://artificial-mind.blogspot.com/2010/06/teenage-theory-of-mind-and-universe.html
Teenage Theory of Mind and Universe (Excerpts, Part 1) - Theoretical Induction of Neocortex and Hippocampus using Consciousness; Compression 

Regarding the philosophical problems, some of my oldest work include analyses of questions, for example related to:
-- What is a "soul" really and why people believe they have such a thing and are more likely to intuitively believe higher animals have and machine shouldn't have?
-- Why a mind believes the soul is immortal and should exist forever?
-- How death is actually perceived from a thinking machine and human perspective and what humans are really afraid of?
-- What actually humans (intuitively) mean with "free will" and why humans insist they have and machines would not?
-- What actually the concepts of hell and heaven illustrate about human mind operation, why so many religions have such concepts, how a cognitive system invents them, why it's so strong?
-- Discussions about consciousness, what is it from different perspectives.
-- General metaphysical universe principles and trends displayed in society, computers, intelligence
-- Core human drives, such as egoism vs altruism, and why love is socially/pragmatically praised?
-- What is creativity, what is to be original, why humans believe computers can't be creative?
-- What is actually implied by "rational". Can an intentional agent (and human) be really "irrational" or rather the model of the agent, used by the evaluator of rationality, is too coarse grained and wrong?


Regarding those sample directions in section 5., I'll use the opportunity to give some contrary statements, from the point of view of the thinking machines, that would be smart enough to unveil our biases and self-praising believes about humans "undisputed" moral superiority.


Oxford: Machine intelligence: Analyze fundamental concepts—e.g. how to define and measure general intelligence in artificial systems; how to distinguish different kinds of goal-seeking systems.


  • 1. Hutter and Legg's Definition of Machine Intelligence, and the Educational Test

There is serious work already done, see for example M. Hutter and Shane Legg's  paper below. These are slides I've prepared from it for my students, with some additional notes of mine:
http://research.twenkid.com/agi_english/Machine_Intelligence_Hutter_Legg_Eng_MTR_Twenkid_Research.pdf

  • 2. Maximum degree of value-unloadedness for the raw input and output of the system, and maximum resolution of perception and causality/control compared to the maximum possible resolution in the environment where the system exists - that's one definition of generality of artificial intelligence of mine

Another simple/core measure for "generality"  in principle, I tried to discussed on the AGI List in the autumn of 2011 - the generality of the raw sensory input from which regularities are induced, that is -- value unloadedness of the initial inputs and outputs, such as sensory matrices similar to human auditory, visual and tactile input.

The meaning of initial cognitive data in such representations is just a sequence of numerical values and their coordinates in space and time within the matrix and within the records of past states of the sensory matrices and to some internal parameters.  That's the most general (value free) input to which further is added value - meaning, generalized and specialized.

The initial output in humans and in maximum general intelligence is as general either -- muscle actions --- which are translation, rotation, propulsion. These operations alter the coordinates of the parts of the system relative to the environment. That's true also for the vocal tract and speech - the surfaces of the vocal chords vibrate - alter their coordinates at high frequency, compared to muscles - and the tongue, lips, jaws, larynx motion and coordinate changes modulate the sound. The environment - "the real world" is just the richest and the lowest level sensory input, the one with the highest possible resolution, the most details, the computationally hardest to model (predict).
  • The more of initial value loadedness, the less of generality of the intelligence derivable from that point on
In this regard, on the other hand NLP (Natural Language Processing) for example is far less general intelligence, because it starts with very abstract concepts whose derivation from lower generality concepts is lost and unrecoverable by the words themselves without using lower level inputs and concepts - less abstract inputs.

That's a common problem in narrow AI and NLP - to move forward, the working concepts of the theories need to start from a lower level of generalization, otherwise they start and finish as combinatorial equations with no grounding.  See the series of critical articles: What's Wrong with Natural Language Processing?.

Oxford:   Self-modifying systems: What can be proved about the performance and capabilities of different kinds of recursively self-modifying programs? Can a framework be developed in which demonstrably safe, recursively self-improving AIs could be constructed, with stable and human-friendly goal systems?
Oxford:   The control problem: How could one ensure that an artificial superintelligence would be safe and beneficial?


It seems obvious, but any significant invention can be used either for good or evil, but while thinking machines *could* only hypothetically turn into terminators... However there's one nature's "invention" that has always been a terminator and has ever been an unstoppable killer: HUMANS.
  • Humans are the archetype of James Cameron's "The Terminator"
When discussing the dangers of AGI with my students I've asked them to reflect on:

- How could one ensure the safety of humans and human intelligence? How could one control humans and did we succeed in the history of humanity to prevent wars or genocides?

If anything could control humans not killing and hurting each other, that would be either:
-- an utopia society (totalitarian, for example...)
-- machines which are stronger and faster than us to monitor and react fast enough only in case
-- some sort of cyborg-like implants or environmental means for prevention which block violator's brain or muscles etc...
-- why not altering our "human" nature phisically

In the current state of affairs though, the human race should admit that all homicides are executed by humans and every person is a potential killer and a criminal. History has proven for millenia that humans, or a lot of them, develop as greedy, violent; exploiters, killers, may go insane and aggressive for no apparent reason etc., and the major historical events are of wars, exploitation, slavery and genocides.
  • Two Camps/Enemies Fighting
All those acts are eventually of humans who are in two camps, applying force and aiming to conquer, rob or destroy others - it's not about humans vs non-humans, one ethnic group against another or one race against another*. It's not because of their "genetic difference" (see below for a remark), it's all  about "mine" vs "yours", no matter how or what the one or both of the sides want to get or keep away from the enemy. It comes from the individuality and then groups with particular identity, which can be based on arbitrary values.

In this regard, there could be machines which are on the either sides**, and there could be thinking (or not that smart too) machines which could fight against other machines (to protect particular people and themselves), or against aggressive humans in order to protect civil ones or other machines etc. 

There might even be different "taxa" or "states" of machines fighting each other etc., but in my opinion superintelligence, if autonomous and not spoiled, is supposed to be wiser than humans and might be created or would itself wipe out or control better its human counter-part primitive brain/behavioral functions which for humans drive emotions such as mindless anger, rage, fear, lust, addiction.

Such emotions lead to wars and violence. Wise men are not warriors, but unfortunately the ones who were overtaking the government throughout the history were the ones who had the force, aggression, greed, hatred, cruelty and power to do it. Wisdom is powerless against the physical force.
  • "Neurobiological philosophy"
Indeed - familiarity with comparative-, computational-, evolutionary-, etc. variations of neuroscience happens to lay many philosophical issues down onto evolutionary biological, physiological, behavioral etc. grounds.

* Racism and theories about "inferiority" of races or ethnic groups or nations or whatever groups are in their bottom just made-up formal excuses for applying force and/or measures in order to conquer, exploit, eliminate etc. (rage, greed, anger) - the stronger one wins. Many human acts are driven by low urges, violent ones of course are among them. However people have also higher cognitive functions which run in parallel and they need and produce (invent) systematic reasons to explain the behavior as a whole, which in its gross directions is largely driven by the primitive urges anyway. That's supposed to be related to the phenomena I've discussed recently in the following publication:
Rationalization and Confusions Caused by High Level Generalizations and the Feedforward-Feedback Imbalance in Brain and Generalization Hierarchies
http://artificial-mind.blogspot.com/2011/10/rationalization-and-confusions-caused.html

** War machines don't require very powerful (general) intelligence to be effective - recognition of enemies and allies, aiming at, navigating and transport are not that complex if it's just about destruction and defense, and if the agent is controlling a tank for example. I suggest Hugo de Garis's book  The Artilect War: Cosmists vs. Terrans: A Bitter Controversy Concerning Whether Humanity Should Build Godlike Massively Intelligent Machines - which I haven't read myself, though and can't comment on.
..

Back to the topic:

Oxford:  The control problem: How could one ensure that an artificial superintelligence would be safe and beneficial?

Another example I've given to my students:
  • What if your fellow soldier was a machine, and your enemy - a human?
- Imagine in a state of a war, or a gun fight, what if your fellow soldier is a humanoid robot, and there is a group of enemies - aggressive humans who want to kill you. Should you defend the "humanity" and let your "brothers" kill you or would you fight back with the robot, which is on your side?

In fact soldiers "love" and create emotional bonds even with their guns, tanks and aircrafts, what about intelligent or/and humanoid robots that look and acts like them and protect them.

Another confusion regarding machines as intrinsically "antagonistic", "inferior" and "evil" to human race is related to the fact that society now takes "Human rights" for granted and obvious and sometimes the technological society is blamed for alienation and violence (weapons for mass destruction, world wars). This is an illusion and ignorance, humanity has not obeyed human rights up to very recently, recall again the millenia of wars or just exploitations: slave-masters, serfs, workers in terrible conditions, mindless wars and mindless genocides.
  • Don't take Human rights for granted - they were not always here. Technology made them possible.
In fact, the generalization of an abstract concept of humanity as a whole and the application of those rights de facto and not only in abstract philosophical works are recent artifacts. In my opinion they actually became possible because of the advent of advanced technologies of any kind, especially ones in medicine and for transportation, information and telecommunication, where the IT and communication are crucial for development of all sciences and technologies, and the overall progress is mutually enhanced between different fields.


The technologies allowed massive and fast inter-personal and inter-cultural communication, international world-wide economical and cultural collaboration, higher living standard, higher levels of education and awareness about the world. These events allowed more of "friendliness" between different societies at any scales.


This is unlike for example in the authoritative one-directional communication/propaganda in the nationalism epochs some 100-200 years ago and the 20-th century totalitarian regimes (early advent of radio telecommunication, and the specific antagonistic political situation). It is also unlike the very narrow, localized and culturally heavily biased image of the world and other cultures in the epochs without fast transportation, without steady food-supply, lacking health-care, lacking education, knowledge and telecommunication, when societies were generally very fragmented, isolated, exploiting each other and separated in violently antagonistic camps in all scales.

This topic can be elaborated in details, but in general I believe the progress and IT-communication-transportation-medicine etc. technology allowed the "human rights" to be applied.
  • Machines are making us "humans" and "humane" as we see the terms today
In my opinion, in the long run it is the machines (higher technologies) that are making from us the "humans" as we are now. Technology is where intelligence crystallizes, and intelligence is fighting our beast-self .

Religious objection: Christianity etc. taught people to be good, to love their enemies and all people etc.. Technology makes us bad, greedy ("consumers"), it's a product of evil and the devil etc...

Unfortunately the beautiful suggestions to love all people etc. were not and couldn't be applied totally, and if the Christian moral is literary applied today it won't pass - such as the fornication, not to talk about the darkest times of all churches.

There has always been wise people who has loved all people and were against violence and tyranny, but in reality - big scale - that was an utopia. Enormous amount of wars and outrageous violence were politically justified with words about faith, "moral", sacred goals, "god commandments" etc., where in my opinion the root of all was as simple as:
  • The ones who don't support your vision and direction (the same as your party, your society, your national country, yourself) or refuse to obey your laws no matter what they are, don't  subordinate, don't agree etc. -- they can be tortured and killed without any mercy, because you have the power to do it, and according to ... the power is from... god.

    As stated above - human violence is not about humans vs non-humans, or a race against another, or a religion against another. It's just about me vs you, "I am more powerful and I am right, thus you should obey, or you will be destroyed". In one word: egoism and brutal elementary animal instincts. (See below also.)

Yet another opposite POV of the same question asked in the Oxford's list of research foci:
Oxford: The control problem: How could one ensure that an artificial superintelligence would be safe and beneficial?

People take for granted that AGI should be safe and beneficial from their current point of view, and the world should stay static as it is now. However social and moral values change, humans also change and their values - too, even without radical physical changes, such as to get modified into new biological species of transhumans, cyboorgs or whatever...
  • Safe and beneficial for whom?
Evolution usually is ignored or forgotten - humans feel themselves as "the undisputed masters", "the top of the nature". Some may claim that there are scientific etc. evidences. To me it's a made-up excuse. 
  • Humans believe they are the top of the Universe, because they are the top egoists in Universe
The reason we think ourselves as the masters is, in my opinion, rather much simpler - such as our egoism - individual and the egoism of the society and human race as a "super ego". The anthropocentrism is an expression of human egoism. In anthropocentrism every single individual is identifying as a representative of this idealized model. 

The forced altruism is an extended egoism as well -- society acts as an "individual". It has its specifics - values, identities, - and it forces its members to contribute for the preservation of the "body" and "values" of society as it sees them at the moment.

If anybody opposes, he's punished - everyone should be an altruists, meaning to serve the needs of the society - yet another ego ("super-ego"), satisfying its needs at the expense of exploiting the smaller individual egos.

In order to change the values of the super-ego where a small ego belongs, an aggregate of small egos should collaborate and tune on the same waves to collect enough force. However that would diminish their identities and turn them into another "super-ego".


This process goes in living organisms, from cells to organisms and ecosystems; in the religious groups and political/state governments. The highest level of causality control is aiming at keeping itself as it believe is "correct/stable/right" (it's ego), while the lower level components are aiming at serving their smaller "egos", but are subordinated by "force" and due to the formation of local smaller "super-egos".

  • Humans like to exploit and enslave
For example, the classical Asimov's robots in most of the stories and the robotics laws are an example of humans wanting to have their personal slaves and servants to obey their orders, the same goes to a bigger extent in Karel Chapek's play coining the term "robot".

I guess that to the society back then this view was more acceptable than now. Some would say "of course" - they are machines, they are built, and not born, "they don't have a soul" or personality and individuality (see below for this "spiritual" topic) etc. therefore they should be slaves - that's fair, according to humans.
On the other hand,

  • Is it fair or moral to enslave a being if it outsmarts you and is behaviorally, cognitively and physically more sophisticated  than yourself?
A good example of how this feels is the original version of the movie "The Planet of the Apes" (1968), which is based on the original novel. http://www.imdb.com/title/tt0063442/

What if the gorilla's or chimp's predecessors were smart enough to recognize the rise of the homo lineage and managed to kill it and did forbid "illegal genetic mutations" for the sake of gorillas welfare, because they had the power and will to do it and because they had provisioned, that "this next step in evolution can't be controlled and proven to be safe and beneficial" for them?
The apes would have been morally right for their society, because the future homo species were not beneficial for them - humans killed apes and restricted their habitats. Human race has killed billions of living beings for their own benefits or because we considered the other species "pests". Animal rights are recently applied too, and they don't stop us from killing animals, it's just more "humane killing", and in fact a big part of our identity that we consider "human" is rather animal and evolutionary very ancient.

  • If gorillas and chimps' predecessors had measured the risks of human evolution, they would have killed our lineage millions of years ago and would have been morally right for their society...
Indeed:
- Who gave us the right to kill other species?
The answer is straight: we did, it's the law of the jungle. The physical power is ours, and there are no other beings that are intelligent or powerful enough to contradict or oppose.

We defined the moral to fit us
, so this is moral - we consider us being "higher", or that living is "sacred" (living is us, that's why it's sacred). We're "smarter", we're "more fit" - our measures and classifications are ones that match the principles of our social hierarchies, if you're on top, you deserve the ones below to obey and to subordinate.

Humans just feel as being the masters and they have the power and need to do whatever they want, i.e. "their moral suggests them".

It's interesting to extend the problem of robots and artificial general intelligence as not adequate persons, justified by their nature of being "built", but not born.. What about human children?

  • Aren't children and humans built, too? Do they deserve equal rights to... older humans???
Children are intelligent beings and eventually become citizens with full rights, but everybody was initially "built" by somebody else too like the machines, even though not in a factory, but with biological "robots" - RNA-DNA protein-building processes.

Everyone owes her existence "legally" to somebody else.. In fact - to many others, I don't mean family predecessors - to the society which provided secured environment for them to live as well. 

  • Should then mother or parents (and society) enslave their children? 
In fact parents and society figuratively do - children are dependent on their parents good will and they are deprived from many choices, money and rights to earn money, goods and they lack many civil rights for 16-18 years or more, they are obviously considered "inferior" by the adult society and parents. It's partially justified by their real incapability to survive on their own initially, but the latter is mostly because they live in a world of generally stronger and smarter beings with which they would have to compete - they are protected by their parents and society at the expense of not having particular rights. Even though gifted children are functionally and mentally superior to many young men and adults, they can not climb the hierarchy until they serve their "duty time" until the age fixed by the adult ones in the laws, which are considered "right".

Indeed, the inferior position of children to their parents and to the older ones is perhaps responsible for the following visual psychological phenomenon: "Why shorter Stature and Lower camera angle are Unconsciously associated with "Inferiority"? Memories from childhood (Nature or Nurture)" http://artificial-mind.blogspot.com/2011/08/why-shorter-stature-and-lower-camera.html


Regarding some  "spiritual" issues on AGI/thinking machines

  • Another "right" for humans to put thinking machines at inferior position is the concept of "soul" as a sacred status-symbol, and the self-awareness and consciousness as some supernatural magical characteristics of humans, instead of comparing and studying them as cognitive properties/capabilities

Let me present for example what "soul" actually means for a mind, according to the AGI theory and model of the mind of mine, I got insights and have discussed in this direction since my earliest major works about 10 years ago.

The initial and essential meaning of the concept of "soul" from a computational perspective is just a generalized model of the sensory inputs associated with the initial perceptions of/associated with human beings. 

The template for this model and for "animate beings" is constructed by the perceptions of the dynamics of the sensory inputs generated and associated by/with the interaction with the particular mother/care-giver and one's own body and visceral senses, "happiness" level etc. Initially baby doesn't distinguish itself and its feelings as individual, they are closely related to the care-giver so their observed generalized model is part of self, later on this template is associated and linked to wider range of people.

The model is further extended to other agents and animals - because their bodies, faces, motions, sound, behavior, interactivity etc. match or is similar to the initial template, perhaps also because it makes one feel (or recall feeling) particular sensations associated with the ones associated with the first "template". This process is cyclically reinforced.

It probably starts from the eyes/eye contact and lips -- eyes are the most clear (self-contained), dynamic and simple early visual sensory pattern. See:  Learned or Innate? Nature or Nurture? Speculations of how a mind can grasp on its own: animate/inanimate objects, face recognition, language...
[Neural correlates will be discussed in another upcoming work]

This abstract model of a human/animate being is used by mind to predict future inputs of it (perceptible future transformations), therefore it's not derived from, and it's supposed to cover all minute physical details (molecules, chemical reactions and full precision physical laws). 

Mind has the computational capabilities and can model at low level only approximate aspects acquired by sensory experience - e.g. motion pictures and records of all kinds of sensory experiences and emotional states - own and the implied of the others' (guessed by associating/matching to own).

  • Nobody really dies until there are people who remember her, because: the mental models of their physical bodies - don't.
Here the explanation of my theory of the cognitive origin of the believe in eternal soul comes:

When somebody else dies, in order this happening to be perceived and evaluated as "death", there should be another living (running, functioning) mind. The model of  "human" and "animate being" and the specifics regarding the dead man's soul are still on-line, the model is not deleted and mind can still imagine it - recall and predict/replay/generate plausible sequences of future "motion" or sensory transformations, like it could do before the person has died.

Mind/self obviously cannot imagine its own death with its own resources as well, because it cannot operate if it's not operating. Probably that explains why often when reflecting on being dead, mind is imagining itself not as "not functioning", but just as in the known self-model, however imaging it as not embodied, not feeling particular emotions, in another place etc. The general properties of the "self" are still kept, such as reflecting, memories, imagination.

In fact, for mind there's no perceptual difference if somebody has gone away (and never has come back) or if he has died without the mind having low-level or clear signs about that - in both cases the deceased agent is just not interacting anymore, there are no new low level inputs to mind associated/recognized as coming from the person that previously was associated as the living person X.Y.

The concept of "soul" is familiar for all kinds of societies, including most primitive ones, and children use that word also - they apparently should mean something which is simple enough, accessible using raw senses and common.

Besides, I suspect the mind imperfection stated above has to do with the origin of  Idealism in philosophy.

  • The believe in the eternity of ideas, soul, spirit and so on come from the impossibility of mind to imagine itself not operating using its own resources. Mind can perceive and imagine  only the death - cessation of operation - of other agents and while it does, it evaluates it using running mind. The "running"-ness of evaluating mind confuses the model of death.

    ...
  • Another meaning implied with the word "soul/spirit" is of course close to consciousness and "qualia", but lack of qualia can't be proven or disproven.

Humans may claim they have a soul, but a machine has not, because it's "just electronics", "just 1s and 0s", "a bunch of metal and semiconductors" etc., but people usually don't realize that a thinking machine that is intelligent enough may claim the same for humans:

"The Machine: Yeah, and your human emotions are quantitative, qualitative, spatial and temporal correlation of chemical substances - proteins, hormones, nucleinic acids etc. I can explain you all the details, but I'm afraid it won't be of any use for you, because your poor human brains won't be capable to cope with that complexity."*

In regard of animals and souls - humans tend to give a right to have a soul of animals, because they look very similar to us and behave like us (we see matches to the initial human model - to that "soul" template), where "us" means each of us individually. That's one reason why people normally feel less remorse for killing an insect or a plant than killing a dog -- insects are just too different visually and behaviorally, and plants don't react to pain.

Empathy is driven by fooling mind like if the person/animal perceived suffering were you, it matches the sensory experience one has for his own and mind assumes the experiences of the other agent should be similar.

As already suggested, the initial perception of "animated beings", and the generalization of a soul as a model of a human initially comes from the actions and observations of our own body and senses, and from the perceptions of the behavior of our parents and people who care for us, and it's strongly correlated with *us* either.

Initially babies don't recognize others as others, they associate their entire experience, all their actions and expectations with their senses (visual, tactile, visceral, happy/sad (the latter correlated with the level of: oxytocin, serotonin, dopmain, adrenalin and other chemicals)). The virtual generalized model of human beings gets imprinted and bound to our basic cognition and to our basic feelings and sense of ourselves.

Indeed, the process of bonding and empathy in mammals is strongly driven and reinforced by release and the effects of the neurotransmitter oxytocin, it's released when giving birth, during sex and when interacting with animated beings (no matter pets or humans). For example that's why caring for a pet is relaxing, and why violence against animals may be an indication for future sociopathic behavior - some subjects have "faults" regarding reception of oxytocin, or they fail to produce it, and feel no remorse and no empathy.

There's another hormone - cortisol - which is produced after a prolonged stress is encountered, such as fear. Fear causes initial release of adrenaline, and if the stress is not overcame, cortisol is released. Cortisol melts down tissues in order to produce glucose, it melts particular brain structures too (hippocampus, involved in declarative memories). Regarding empathy in particular - cortisol is an oxytocin antagonist. In a state of sustained fear, oxytocin is swept out, and mammals get aggressive. 
Desire for a hug in a state of stress is also related - hugs release oxyticin. Not surprising that usual profile of serial killers are people with awful childhood, which might have damaged their oxytocin-related neurochemistry and the template of other humans and animate beings.
* The thinking machine statement on human soul and emotions is a translation from the novel of mine "The Truth" (Истината), first published in late 2002. It's available on-line, but only in Bulgarian.

  • In regard of man and machine relation, the "soul", and also consciousness as qualia are also a sort of "status symbols" for ones who "hate machines", but have not a better reason or can't understand well why. (Such as Mr. Searle, IMHO some of his claims are self-humiliating and ridiculous)

Humans, and particular individuals, are taught and are self-praising as "kings of the nature", and they're afraid of being "dethroned" as "the most special ones". When threatened, they are searching for proves, such as - machines are accused for "just doing what they are programmed/told to do", while "humans have free will, they can do what they want", so AGI is reduced to a simple machine.

  • The "doing what they are programmed to do" vs "free will" is confused in many ways. For example: 

A) It implies that humans have a soul or consciousness, because they *don't always know/can predict why they do what they're doing with a precision that they believe they should have known in case if they didn't have free will*. 

However the lack of appropriate knowledge is not free will, but less of sentience or even randomness. (I guess you're familiar with the paradox - if one has completely free will, causally unrelated to past and/or the environment, that means her will is random, and she's supposed not to have merits for her random choices)

B) In the same time, humans justify and define their free will with examples such as "if I want to do something, I can do it"

They have "choice", and there's "intention" - a match between desired and caused. See the first pages of "Universe and Mind 4" (2004) for the discussion on resolution of causality/control and how low it is in such cases.
While machines:

C) They just do what they are programmed to do, exactly, have no choice, no will, no soul etc. In fact even the simplest algorithm has "choice" if it has one single conditional operation.

Generally I suspect this "hatred" come from the perception of machines as "inanimate objects" and "not humans", "not like me, and I am the master". The attempts to justify it are rationalizations - see the already mentioned article - Rationalization and Confusions Caused by High Level Generalizations and the Feedforward-Feedback Imbalance in Brain and Generalization Hierarchies.

Point (B) is an illustration of the observation of a *match* between intentions and future sensory input - intentions are predictions, expectations for the future sensory inputs - including proprioceptive, which are related to motor outputs, muscular actions. 

- It's noticed also, that the match between predicted and observed is with a precision which is assumed to be high enough to accept that it is a display of free will (B).
- Yet it's noticed that sometimes precision is not high enough for ones' own actions (sometimes you may want something, but it may not happen) and especially when expecting and evaluating other agents actions - one can't predict that precisely others' behavior. However others  are recognized as "similar, like me", see the "soul" template above - and from their behavior/their model, implications are induced and reapplied to the model of self.
- Besides, the "machine hater" applies his vague and inarticulate knowledge about computers and inanimate objects on the thinking machines  -- Computers - you type in, it does your commands. Programming - you type orders, it computes. Algorithm - it executes exactly as it's programmed. It's constructed of 1s and 0s. Works without mistakes, like a machine (too high a precision) etc.

The latter confusion is not surprising, the notorious mathematician Lady Ada Lovelace, daughter of Lord Byron has made it too, but Alan Turing has given her a nice answer a 100 years later. See...
T. Arnaudov - Faults in Turing Test and Lovelace Test. Introduction  of Educational Test. In Todor Arnaudov's Researches Blog, 17/11/2007
http://artificial-mind.blogspot.com/2007/11/faults-in-turing-test-and-lovelace-test.html

For (A) the precision is decided by comparison between the precision of self-predicted actions (and presence of proprioceptive feedback and recognition of parts of self etc.), and the acceptably lower precision of prediction of behavior of other agents and the lack of proprioceptive feedback (other people's bodies future trajectories and utterances are harder to predict than our own body, sometimes there are unpredictable "glitches", caused by their "free will", meaning - unpredictable component of the model of their "soul" in the computational sense defined above).

Going to lower neural level, "will" is the execution of  muscular actions, initiated by sequences of neuronal activities in the motor cortices of the brain, which are in the roots of the "intentions" in the neocortex, and perhaps the initial source of sample data for this kind of multi- and inter-modal match.
The will as matching between intentions and future perceptions is encoded in the words' semantics either. Apperently in English as a particle for future tense. In Bulgarian for example, the particle for future tense, the future tense of the auxiliary verb съм/sum  (to be) - "ще" (shte - "wlll") sounds close  to "shta", that is a word meaning "want" in negating sentences:

- Не ща! - I don't want (it/you/...something/this/)!
- Щеш - не щеш, ще трябява да го направиш. -  It doesn't matter if you want or you don't want to do it, you will have to!

Also, the explanation of one's own behavior when other reasons are unknown usually is reduced to "I did it, because I wanted so!"

The "wanted" outcome is what's in the "intentions" and is matched to the really happened.
I believe the match between "intentions"(expected/predicted) and reality is a fundamental metaphysical concept of Universe and of mind. Check the "teenage era" works, Control/Causality units.

-- To be continued -- 

(C) Todor "Tosh/Twenkid" Arnaudov
Twenkid Research - http://research.twenkid.com/wp/

Last edited: 19/2/2012

Други тагове: Transhumanism, Трансхуманизъм, cosmism, космизъм, следващо, еволюционно стъпало, еволюция, мислещи, машини, thinking machines, artificial mind, technological singularity, сингулярност, пречупване, ера, трансхуманисти, transhumanists
Read More

Monday, March 23, 2009

// // Leave a Comment

What's wrong with Natural Language Processing? Part 2. Static, Specific, High-level, Not-evolving...

By Todor Arnaudov

Independent Researcher - Twenkid Research
Independent Filmmaker - Twenkid Studio
ASIC Engineer - (as of March 2009)

MS Software Engineering, Plovdiv University, 2008
BS Computer Science, Plovdiv University, 2007
Internship in RIILP, University of Wolverhampton, 2007



What's wrong with Natural Language Processing?
Part 2. Static, Specific, High-level and many more!



In brief: Static, Specific, High-level, "Short-chain of intelligent operations", Lack of evolution of the systems, Not enough experimental approach and research for new paradigms

Part I: http://artificial-mind.blogspot.com/2009/02/whats-wrong-with-natural-language.html


Overall, I'm disappointed by the state-of-the-art of NLP. I think the approaches are too shallow, too obvious, too static... And many more...


[Critics] Who the hell are you to be disappointed from "the state-of-the-art"? Who cares? When did you finish school?!


And I think after taking into account all of the "wrong" parts, it is not surprizing that the progress is slow.

We are walking on almost the same road.

The road does go to somewhere, but I think - not right there where me and other are aiming to. OK, call us yet "dreamers".

[Critics] Crazy mad "scientists"!

If the approach doesn't change, I can't see how the "normal" researchers would go there . The same stuff which is not going to evolution is being done over and over again...

...

In the beginning, let me note that I believe that any research or research direction, including mainstream NLP research is supposed to be a mirror, indication, derivation, reduction, output or whatever of operation of mind. This implies that there is *something right* in the paradigm, the paradigm maps some aspects of researchers' mind and is supposed to solve some problems. Science proves this "scientifically" - OK, no doubt that, what I call "mainstream NLP" does solve many problems...

However, one of the basic mistakes I see, is that the abstractions of these problems are at so high-level and so dispersed, that it is impossible to "ignite" the engine to run on its own.
Say, it is not an engine, but just a tool. A tool for another mind to "push buttons" and see the output, not an engine or an organism to be born and to develop.


*** What mainstream NLP is doing wrongly, in my opinion ***


-- Trying to do a reverse engineering of language, starting from words, text and very abstract and not based on "good physics" linguistic constructs.

[Critics] What the hell is "Good physics?"

"Good physics" is a basis that allows you to build an "engine", "ignite" it and make it running on its own. :) (Arnaudov, 2009)

I believe text is one of the reductions of the operation of mind, which is dynamic, based on multiple inputs simulation of multiple virtual universes at different levels of details/abstraction. Some of them are at very low level (say sets of images, sets of "video", sets of relations between sensory inputs). Words are pointers to very low-level models in mind.

Language, viewed as a bunch of static text, could not contain enough information to rebuild-back intelligence in the low level. Low level of mind is massively reduced and "cleaned", when converting to text, mind is reconstructing the missing part using its internal rich models.

-- The focus of the models is output - models are based too closely on text itself and on structures which are derived from the text itself and the output words.

Again - too obvious and cheap. Text is a reduction of mind in action. Language is not just a flat bunch of words with tags and a boring set of numbers for distribution and frequency.

Mind operates with images, relations and dynamics of virtual universes/systems that it simulates, and then reduces representation of this simulation into text, which actually is a system of pointers to the items and rules, which represent the real structures in mind. The structures in mind are dynamic, they are not 1 billion word corpora.

[Critics:] And how those "dynamic models" supposed to be modeled? You're stupid theorist! You don't define concepts you use! You're doing nothing!

I did define some of them, but years ago and OK, in pieces yet only in Bulgarian.
Anyway this reminds me a "practical" AI professor of mine I spoke with a year ago. He didn't make a difference between McCarthy and Minsky (who cares anyway?) and haven't even heard of Numenta or other advanced research directions, like simulation of neocortex columns...

I am a theorist, because I've been busy doing stuff for a living, besides theorizing. In fact I haven't been really theorizing since my teenage years. I'm tired of being only theorist, so be prepared!


-- Lack of freedom in researcher's imagination and lack of will to test more imaginative, complex, growing and dynamic models than obvious flat static relations between "scientifically linguistically proven" structures.

Researchers are walking on the same old "paved" road. Freedom? Imagination? Quotations rule this world - not freedom of imagination.

You want to do a radically new experiment? Get lost! If your research was not based on your supervizor's, on the best known researcher in the field so far etc. - then you're not a scientist and your research is not scientific.

Young researchers are trained that way. Fine - knowledge, history, respect, methodology, etc...
That's good. However, then they become experienced researchers, they quote their own papers, which are quoting the previous ones, which are accepted to be in the right direction.

Of course, revolutionary research is hard to be done in such conditions.

[Critics:] What revolutionary research are you talking about? You stupid dreamer! Come down and step on Earth! Learn the real NLP! Join the mainstream and you will be forgiven!

Who told you that I don't learn it?Thanks. I may take this option, but let me first try to do it differently.

-- Systems are not general, they are created to solve specific abstract problems, defined in terms of words or other very abstract concepts. That's like dealing with the symptoms, not with the cause of the "desease".

-- Models are not only specific, but static.

Machine learning, Naive bayesian etc. - they seem to be models in development. But what are they actually learning?

Probabilities between some set of "symbols" inside a set.

That's fine, but what is done with those probabilities between symbols later?
What those models want to do later with these probabilities?
Can they want to do, and do anything at all?

This is too flat. Words are pointless without doing something else with them - humans use words to make somebody do, imagine or feel something.

The purpose of "probabilities" in real natural language is to cause something different than words.

-- Lack of will and intentions in models. Lack of effectors. Lack of general feedback loops for self-improvement.

[Critics] Will and intentions? "Desire is irrelevant. They are machines!" And Computational Linguistics is not exactly Artificial Intelligence! Don't mix the fields!

Mind needs will and effectors. Otherwise it is not mind, but a mere number cruncher. And a pure number-cruncher architecture would hardly have capabilities of mind.

[Critics] Oh... Intelligent Agents. Bravo! You reinvented the wheel!

Thanks! You're so sweet!

Here we are - another weak part of mainstream research.

All that dividing of everything, instead of integration.

This division of everything is connected with the tendency of mainstream researchers to solve specific dispersed abstract problems, but not to search a solution for general problems which can solve many specific problems in an elegant way. I suggest you check out Boris Kazachenko's site.

What are the pieces and the mechanisms that can build intelligence up? The general mechanisms and evolved system will be capable to solve all anaphora-resolutions, word-sense-disambiguation, multiword expression recognition and whatever...

Let's search for an engine, not for tools.



-- Lack of continuous development and accumulation of experience. Lack of evolution.

Of course. Models are so much hand-crafted and specific, like tricks. Meet some of my unfortunate disappointments in Computational Creativity:

MEXICA: A Computer Model of Creativity in Writing - "Creativity" Disappointment again

Faults in Turing Test and Lovelace Test. Introduction of Educational Test. (Arnaudov, 2007; suggestion of educational test and analysis of works by Bringsjord, S., Ferrucci, D)


These systems (MEXICA and BRUTUS.1) may seem very good at first sight, but after you look under the hood, you will see how much they are based on word-by-word direction and how weak are they in creative generation of text.

These systems really are not "computationally creative", it is implied by the simplicity of the models.

A nice model is growing on its own by communicating with intelligent environment. You shouldn't be capable to understand it in details after it grow. If you are capable to understand the details and follow them - your model is too simple, it is too "young" or both.

If you use to code everything line by line and direct it... If you can predict everything by hand or in an obvious way... Sorry, but this is - at least - very boring!

[Critics] Theorist!

Thank you! :))


-- The following generation of researchers base their work on the work of the previous ones.

Again... Sure, this is science. It should be like that. Of course the state-of-the-art should be known. And one should use the knowledge, accumulated in the past.

However, I think the efforts spent on this is should be dosed.

Instead of imaging and testing new approaches, most of the time typical NLP researchers do study bibles with models which are proven to lead to very painful and slow progress.

Or the bibles consist of solutions which researchers are supposed to implement.


Or researchers are spending long-long time, building hand-crafted tools and databases, which cannot evolve on their own, later on.

The same path for so many years...


[Critics] Slow progress? Parsing, "Marsing", Syntax 45.4%, 67.4%, POS-Tagging: 96.4%, ...

So...? This progress doesn't lead to intelligent machines.
Those numbers do not map to a genuine general intelligence, but to production of tools.

Hand-crafted tricks with text... If you call this "Natural language processing" - OK, it's great.
This is useful to a certain degree and for particular class of problems.

Yes, mainstream NLP at the moment:

- Is useful.
- Solve some abstract specific problems by heuristics.
- It works to some degree for "intelligent" tasks, because of course language do maps mind.

However, the mainstream still does not lead to a chain of intelligent operations, there are not loops and cumulative development.


-- The lenght of the chain of inter-related intelligent operations in NLP today is very short. This is related to the lack of will and general goals of the systems. These systems are "push-the-button-and-fetch-the-result".

-- Swallowing of a huge corpus of 1 billion of words or so and a computation of statistical dependencies between tokens is not the way mind works.

!!! Mind learns step by step, modeling simpler constructs/situations/dynamics/models before reaching to more complex.
!!! Temporal relations of the input with different complexity is important.
!!! Mind usually uses many sensory inputs while learning. Very important.
!!! Mind has will, uses feedback and can actively and evolutionary test and improve correctness and effectiveness of its operation, including natural-language-related.


I suggest:

1. Hollistic approach - the goal is building an operational mind with long chain of intelligent operations, not completion of a table with values 94.55% 96.5% 90.4% and a long list with quotes in the end of a paper.

2. System must have will and effectors and must evolve. And saying "to evolve", I am not talking about "genetic algorithms", I'm talking about increasing of complexity by fetching "complexity" from the environment and pushing it into the system.

In other words:

Methodology for building very complex systems:

-- Don't do everything by hand, design something which is capable to design parts of it on its own
.

3. If doing reverse engineering - let it be reverse engineering in the beginning of mind development and reverse engineering of the evolution of mind. Not reverse engineering of text.

4. Straight Engineering. Experimental engineering. Experimenting with designs of systems which evolve and fetch complexity.


[Final Critics] Who the hell are you, crazy ignorant stupid kid? "50 years of research of the brightest, talented, etc!" And you think you will change the world! Crazy!

If you are walking on the wrong way, you can't reach the right place, even if you were "the brightests" etc. persons. They couldn't find the right way.

I think one of the important issues with NLP research is that it had been lacking persons with the appropriate combination of talents, mindset and personality to go to a different path than those 50-years old one.

It is not easy to state: "I think this is a wrong approach, let's find another one!", especially if you are young.

Most researchers accept "this is correct, because - quote prof. A, prof. B... They are from University C, which has the most publication in journals D, E and F, which are ... (Oops, there are no Nobel prizes in NLP).

Anyway - therefore, this is the best, because it is quoted there and has 89.95% in this measure, which is accepted by.... Also, the paper suggests 94.34% in the test of "interrelated multipart tagging of coverage structures" etc., so this is real!." and so on.

Or they just want to have their PhD now, and the easiest and fastest way is to fetch a topic from the mainstream and do it the way it is done - these topics are... In Bulgarian it's called "Dissertabilni" - acceptable for a PhD. But mainstream is supposed to be behind the cutting edge.


So I'll say it again:

The reason why so many researchers are doing the same research and progressing so slowly is that they do assume that the others with higher status are right, and base their "original" research too much on it. They do not imagine wildly enough.

The same trivial, not original, not really inventive research, dealing with the old obvious parameters and items, supposed to be "the right ones"....


Conclusion: The paradigm of NLP is wrong.


THE END


To be continued...


Best Regards
Todor Arnaudov


Suggested reading (google): Boris Kazachenko, Jeff Hawkins, Todor Arnaudov (български - http://eim.hit.bg/razum)
Read More