[Fis] Fis Digest, Vol 105, Issue 12 Expressibility of AI models
Eric Werner
eric.werner at oarf.org
Thu Oct 26 11:44:29 CEST 2023
Dear Karl and All,
Interesting link between expressibility of concepts in different formal
and mental representations. To get a deeper coherent understanding of AI
models and how to develop meta-AI (Genius, Artificial Wisdom) models, we
need to come to grips with the very basic notions of how the self
develops in a social setting. The formalism we use as Karl points out
can hinder in the sense of putting inherent combinatorial and
computational limits on what can be represented or can further the
development of higher order AI systems.
What has amazed me is the wide scope of possible formalism's and methods
thought in the different contributors. To come up with a coherent mutual
understanding is a challenge.
Best wishes,
Eric
> Dear All,
>
> Intelligence is a wide subject and there are many opinions on it. Eric
> gives a brilliant overview of areas and aspects of intelligence. We
> should clarify, following Pedro’s question: /is there an inanimate
> form of intelligence?, /which of the many layers of the mental
> construct ‘intelligence’ we talk about, and to what ends.
>
> Eric does not distinguish between mechanical problem solving (finding
> out one specific among many alternatives) and that what comes after
> /cogito ergo sum. /The self-referential application of intelligence on
> the system that does problem-solving intelligently is where a boundary
> between Genius (Eric’s term) and a finite automaton can be drawn.
>
> At our present stage of insight and knowledge, it appears more prudent
> to concentrate on the lower parts of the continuum. This approach
> leads us to the lower boundary, separating ‘regulation, cybernetic
> decision making’ to the ‘ability to remember and to learn’.
>
> Answering Pedro’s question about the inanimate forms of
> interregulation by feedback loops uncovers the fundamental paradigm
> that needs to be changed. We have learnt at Programming 101 to see a
> thermostat as the simplest decision-making unit. *Temp < threshold_low
> : switch on, temp > threshold_up : switch off. *The general idea is to
> trisect a continuum and apply consequences to the perceived position
> on the continuum.
>
> The paradigm question is: where do you get any *thresholds *using the
> Wittgenstein-type rational logic, since it is based solely on *N*? The
> Sumerian concept, which was formalized by Wittgenstein and refined by
> Shannon, does not allow for any irregularities or additional amounts
> for any of its basic elements. Using a counting system with /a = 1 + 1
> + 1+…+1/ where the /number of how many times 1 /is the definition of a
> symbol is not suitable by its grammatical rules to generate
> thresholds. One has to get planar and use trigonometry to establish
> e.g. /sin(x) /to encounter thresholds, albeit these also come up with
> a regularity which gives no rise to any situation of dramatic change
> in whatever properties we talk about. The Sumerian system of symbols
> is incapable of generating explosions, discharges, breaks, collapses
> and the like, because the elements of the symbols set remain
> standardized. There is no bias that could add up, in the Sumerian
> system. The words of the Wittgenstein language suggest that the
> elements are ordered among each other, and no cracks or contradictions
> exist within that part of the picture about which we make sentences
> that are interpersonally understandable. The basic, cultural consensus
> of the presently valid systems of thought is, that the picture of the
> world is built up of symbols that fit seamlessly and are exactly so as
> one has defined them and no discussion about it. There is 1 set of
> definitions and 1 number line of which the unit is 1. Observations
> cannot be otherwise than expectations.
>
> This is what needs to be changed. The paradigm shift shows us the
> world which can be seen by two glasses of a stereoscopic eyepiece. So
> far, we have used he right way of looking at things, where everything
> is built of identical units and the units have no other properties
> than being of the extent 1. In this view, everything is as expected.
>
> Now we learn to look through the left eyeglass.
>
> 1. If you take a collection of symbols which each are made up of 2
> natural numbers, then
> 2. You have cohorts, in dependence of how many variants of /a,b /you
> use, and
> 3. For deeper reasons [1] you play with a cohort of /136 /individuals
> which are each a pair of /(a,b), /namely
> /{(1,1),(1,2),(2,2),(1,3),…,(15,16),(16,16)}/, then you have
> *Cohort 16*.;
> 4. This etalon collection C16 can be described by establishing the
> individuals’ linear place in any of the sorting orders you subject
> the collection to;
> 5. You will find characteristic names for the individuals, which
> names are as good an identification as naming them by their
> identifiers /(a,b), /namely e.g. /{(stands near the middle in
> [//αβ]), (is way off in [//γδ]), (at two-thirds in [//κλ]), etc.}. /
> 6. The identification by association is an alternative to counting so
> many 1-s on *N* that identify the individual in a defined linear
> order. The identification by *N* degenerates into a special case
> of identification by sorting, but without referring to its neighbors.
>
> Using both looking glasses in stereo, one will find areas and
> circumstances, where the inner relations among the members of a
> cohesive whole are worthy of consideration. Group sociometry, palace
> intrigues, economic modeling, and apparently also the mechanism
> governing the memory /plusque /genetic, all these methods use symbols
> sets that are doubly indexed: a. according to place in Newton sense
> and b: place according to properties differentiating the elements
> against other elements, individuals among their peers.
>
> In the case that the *three-way interdependence
> /diverse-similar-numbering/*//is at its mathematically possible
> limits, the ideal, maximally cohesive cum diverse arrangement of
> symbols on the etalon collection shows /32, 97 /to be thresholds that
> trisect the continuum, here in the form of the number line. Outside
> the thresholds /32, 97 /the collection can be more similar than
> diverse; inside the range, the collection can be in more diverse
> states than it can be similar within itself. (Matter spills out.)
>
> Intelligence is an activity that is observed on organisms with a
> functioning neural system. The neural system must work in accordance
> with, as a daughter of, the rules governing Physics, Chemistry,
> Physiology. Problem solving must have antecedents in the inanimate
> world. To have the formalized case of problems, one needs to have
> distinction markers on something that can change. No thresholds, no
> problems. Thresholds are existentially necessary to anything out of
> which intelligence and genius can evolve.
>
> Thresholds are impossible to find within the Sumerian system. There
> needs to be an independent, but interacting way of counting which
> allows for thresholds to appear.
>
> When we consider the members of a group, the individuals carry and
> possess their individuating marks (most of the time invisibly to us),
> showing how they relate as neighbors to whom during reorders.
> “Example: X is good as {engineer, cook, cartographer,…} but is risky
> because {no team leader, bad husband, can’t economize,…}.” The fit to
> one’s place is determined by several concurring factors, that have to
> do with one’s likeness to the most average of the elements in that
> comparison.
>
> The conflicting assignments of linear, planar or spatial attributes to
> an element that is a member of a cohesive group creates thresholds,
> levels, units of level change.
>
> AI needs something to ponder about. Why not give it the task of
> tabulating instances of /(a,b) /that can coexist during reorders?
> There are very many details that AM (artificial memory) can remember
> and typify. Creating a generic lexicon organically, starting with
> /(a,b)/, should be no greater a challenge than to write an automaton
> that can play Chess or Go. Bundling cycles of the etalon collection
> that can go together, one could start a work comparable to a
> dictionary of basic fuseki (where the joseki are given by the factual
> bundles of potentially coexisting cycles). Such a Lexicon of All
> Possible Occurrences would be intimately cross-referenced and would
> include all possible words a finite automaton can dream up.
> (Therefore, all sentences the automaton can say.)
>
> In short: use the cycles’ properties to establish a counting system,
> that admittedly works only within groups the members of which are
> consistent with each other (have a liaison running among its members),
> but generates an untold number of alternatives, continuities and
> limits, thresholds and ranges.
>
> It is not a big deal. You ask your student and within a week you have
> a functioning tautomat. Then you know how to shift a paradigm.
>
> All the best:
>
> Karl
>
> /[1] https://urldefense.com/v3/__http://www.oeis.org/A242615__;!!D9dNQwwGXtA!RyX3wbe3vhk-4Ga4WIVvfN5V2-Ye9zZCRGN6PHa6oAokInl7rTYH0Rk8GhKak5eqZhUKeqhECQ3TBFlGDKTZeUk$
> <https://urldefense.com/v3/__http://www.oeis.org/A242615__;!!D9dNQwwGXtA!WbIVI1muA-bUi2sgH-PQNJNVsNmMPNuD5KYiu88Bg8L3Hd7QUTl_gEoOmXRB7r5JyhkrMfVIXVDAv4subhcsC8gXSf4$>/
>
>
> Am Do., 19. Okt. 2023 um 20:52 Uhr schrieb Pedro C. Marijuán
> <pedroc.marijuan at gmail.com>:
>
> Dear All,
>
> As the themes fly so fast, I have mixed some of the ideas
> previously circulated
>
> When Krassimir was asking about definitions of intelligence,
> responded by several parties (Karl, Yixin, Eric, Marcus...), I was
> reminded of a curious fact. Nobody would dare speak about
> intelligence in the "inanimate" world. Undoubtedly intelligence
> appears with life, with the biologic system. Given that life is
> hardly definable, no wonder that intelligence, one of its
> essential characteristics neither is. Of course, we can produce
> many empirical notions approaching it... Nevertheless, where
> exactly can be situated the emergence of intelligence in the
> biologic?
>
> A very recent contribution letter in Nature was claiming with good
> and brief arguments that sentience and cognition are absolutely
> related to the living cell, even the simplest ones (our occasional
> FIS colleague Bill Miller was one of the signatories--in Vol 620,
> p. 37, 3 August 2023). In a similar direction I also very recently
> have published a contribution in BioSystems journal (with Jorge
> Navarro --233, 2023, 105039) attempting the intellective link from
> cells to nervous systems and the human case, connecting with AI.
> In human intelligence, the extraordinary role played by social
> emotions, we argue, should be put in a new light (remember
> Kahneman about S1 and S2 utterly different human responses to
> novelty), at least if we want to contribute somehow to a better
> understanding of today's mounting techno-troubles. Let me state
> that referring to human intentions, purposes, values, etc., they
> do not quite make a cogent sense except properly connected with
> the reality of our life courses or "cycles"... In fact, these
> connections are frequently established in a biased and tricky way
> by most commercial AI systems. But there are positive hints there
> (see the field of "sentiment analysis"), for the hope is that AI
> might open new windows to the rather limited understanding of our
> whole intellection (intelligence/emotions), and even evolve
> towards a new understanding of AI itself, more properly
> intertwined with the extended realms of, say, natural intelligence.
>
> It is in the above sense that I welcome the call to a new
> paradigm, etc., as a possibility to provoke new discussions.
> Although I disagree with the scientific-philosophical validity of
> the term, and with some of the gross simplifications about the
> characterization of physical paradigm. (Some of the most
> magnificent syntheses of human history precisely were there: what
> was the Newtonian theory but a fantastic synthesis of the
> celestial motions and all the diverse motions on Earth? An
> amazing, epochal integration). I also fail to make sense of
> "wisdom"-- does it abide in common folks, in social networks, in
> "experts", in committees, in governments, in entire societies or
> cultures, in our civilization? Is it just a vague idealization out
> from common sense? In any event, wisdom seems to be the most
> scarce, depleted public resource today. In particular, I think a
> well-arranged AI system for medical diagnosis could be far more
> reliable and wiser than a facultative of primary care or a
> specialist (I mean, becoming a great helping hand for the troubled
> practitioners of our overwhelmed public health systems). And
> thinking more in general, these days I was trying to compile a
> list of our common intellectual limitations (maybe I will send
> them to the list for advise later on), in this respect the mirror
> that AI could offer on us could be scary...
>
> To conclude, it looks as if a good rhetoric instrumentation has
> been deployed by the presenters (thanks!), which is important and
> interesting at the time being to promote a general debate on the
> AI complex and somehow risky enterprise, but in my opinion with
> some gaps yet. I will try to advance more precisions in later
> exchanges.
>
> Best regards,
> --Pedro
>
> El 19/10/2023 a las 11:55, Eric Werner escribió:
>>
>> Dear Yixin,
>>
>> Can you be more specific what you mean by "change the paradigm
>> used in AI". It might help to give a specific example.
>>
>> *At present AI systems certainly behave as if they are goal
>> directed.
>>
>> *AI systems appear to have wisdom in that they can propose wise
>> courses of action
>>
>> * What do you mean by "pure formalism"? It seems one of the
>> powers of formalism is to understand AI and human intelligence.
>>
>> * It seems AI systems exhibit human-like wisdom when they offer
>> advice or guide the actions of a virtual assistant or self
>> driving car. The react based on the circumstances and goals of
>> the other, at leas to an extent.
>>
>> * Why can't a machine understand human goals and purposes if it
>> gains a model of those from human data?
>>
>> * Why can't an AI system have intentions?
>>
>> My overall problem is understanding your specific criticism of
>> the present AI paradigm? This notion seems to me to need clearer
>> definition.
>>
>> How would you overcome the present AI paradigm and what
>> specifically is different when you want to "change the paradigm
>> used in AI"???
>>
>> This is not a criticism it is a real question in trying to
>> understand you. At present I just don't see the difference
>> between the present AI paradigm and your new AI paradigm.
>>
>> Best wishes,
>>
>> Eric
>>
>>
>>
>> On 10/19/23 8:48 AM, 钟义信 wrote:
>>> Dear Krassimir, Dear Eric, and Dear Colleagues,
>>>
>>> The discussion is going on well thanks to all your efforts.
>>>
>>> Here is a few points I would like to mention (or re-mention).
>>>
>>> (1) The purpose of the "declaration on Paradigm Change in AI" is
>>> to make an appeal for _change the paradigm used in AI._
>>>
>>> (2) There may have different understanding on the concept of
>>> paradigm. However, _the concept of paradigm for a scientific
>>> discipline has been re-defined as the scientific world view and
>>> the associated methodology_ because the scientific worldview and
>>> its methodology as a whole is the only factor that can determine
>>> whether a scientific discipline needs a "revolution" (Kuhn's
>>> language).
>>>
>>> (3) The major result of "paradigm change in AI" is _to change
>>> the methodology used in AI, including the principles of "pure
>>> formalism" and "divide and conquer"_. This is because of the
>>> fact that _the former principle leads to the ignoring the
>>> meaning and value and thus leads to the loss of understanding
>>> ability and explaining ability_ while _the latter one leads to
>>> the loss of the general theory for AI_. Note that "no explaining
>>> ability" and "no general theory" are the most typical and also
>>> most concerned problems for current AI.
>>>
>>> (4) There is _difference between human intelligence and human
>>> wisdom_. One of the functions of human wisdom is to find the
>>> to-be-solved problem which must be meaningful for human purpose
>>> of improving the living and developing. Yet, the function of
>>> human intelligence is to solve the problem defined by human wisdom.
>>>
>>> (5) Human intelligence can be simulated by machine. But human
>>> wisdom cannot be simulated by machine because machine is
>>> non-living beings that has no its own purpose and cannot
>>> understand human purpose. No purpose means no wisdom.
>>>
>>> I wonder if you agree or not. Comments are welcome!
>>>
>>> Best regards,
>>>
>>>
>>>
>>> Prof. Yixin ZHONG
>>>
>>> AI School, BUPT
>>> Beijing 100876, China
>>>
>>>
>>>
>>> ------------------ Original ------------------
>>> *From: * "Krassimir Markov"<itheaiss at gmail.com>
>>> <mailto:itheaiss at gmail.com>;
>>> *Date: * Thu, Oct 19, 2023 03:32 AM
>>> *To: * "fis"<fis at listas.unizar.es> <mailto:fis at listas.unizar.es>;
>>> *Subject: * Re: [Fis] Fis Digest, Vol 105, Issue 12
>>> Dear Yixin, Eric and FIS colleagues,
>>> Let me present some thoughts about
>>>
>>> *The “Intelligence” Paradigm*
>>>
>>> For those who are not familiar with the concepts of "paradigm"
>>> and "paradigm shift", I would recommend texts from Wikipedia
>>> that explain it clearly enough.
>>>
>>> I myself maintain a neutral position in the dispute between
>>> Popper and Kuhn regarding the development of science. Both
>>> theses have their grounds, but at different levels and stages.
>>> In fact, in this case, the law of quantitative accumulation,
>>> which leads to qualitative changes, applies. Obviously, in a
>>> number of cases the paradigm shift happens in leaps and bounds,
>>> while in others it happens smoothly and barely perceptibly.
>>>
>>> For example, the accumulation of sufficient observations and
>>> evidences regarding the shape of the earth required a shift to a
>>> new paradigm: from the "Earth is flat" paradigm to the "Earth is
>>> not flat" paradigm.
>>>
>>> Sometimes opposing paradigms can coexist, not negating each
>>> other, but complementing each other. For example, this is the
>>> case with Euclid's fifth postulate (the parallel postulate).
>>>
>>> The postulate has long been considered self-evident or
>>> inevitable, but no evidence has been found. Eventually, it was
>>> discovered that reversing the postulate gave valid, albeit
>>> different, geometries. A geometry where the parallelism
>>> postulate does not hold is known as non-Euclidean geometry.
>>>
>>> With regard to the paradigm of "intelligence" we have a similar
>>> situation. We have at least two opposing paradigms based on two
>>> opposing postulates.
>>>
>>> The first, let's call it the "flat intelligence postulate", was
>>> well articulated by Yixin in his post:
>>>
>>> "Intelligence is the ability to solve problems, but not the
>>> ability to detect and define problems, the latter of which is
>>> one of the faculties of wisdom."
>>>
>>> The second, let's call it the "non-flat intelligence postulate",
>>> will sound unifying: "Intelligence is both the ability to solve
>>> problems and the ability to detect and define problems" (Eric),
>>> but in different directions in the hierarchy of intelligences
>>> (KM)". This is how we arrive at the idea of cybernetic systems,
>>> where there is a controller and a controlled, but the controller
>>> is connected to the environment from which it receives
>>> controlling influences and is, in practice, both "controller"
>>> and "controlled", but in different aspects of the system.
>>>
>>> image.png
>>>
>>>
>>> To be continued ...
>>>
>>>
>>>
>> ----------
>
>
>
> <https://urldefense.com/v3/__https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient__;!!D9dNQwwGXtA!XRs1aHCeukEbh8_ycn8OWx-LVxoN0CAv7jFRZZ_WStbmS11a36Z1dHCDJH-xi4P9iqcrHp3wBiFnmw5An8mmT3xuij-z$>
> Libre de virus.https://urldefense.com/v3/__http://www.avast.com__;!!D9dNQwwGXtA!RyX3wbe3vhk-4Ga4WIVvfN5V2-Ye9zZCRGN6PHa6oAokInl7rTYH0Rk8GhKak5eqZhUKeqhECQ3TBFlGq4cryys$
> <https://urldefense.com/v3/__https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient__;!!D9dNQwwGXtA!XRs1aHCeukEbh8_ycn8OWx-LVxoN0CAv7jFRZZ_WStbmS11a36Z1dHCDJH-xi4P9iqcrHp3wBiFnmw5An8mmT3xuij-z$>
>
>
> <#m_-9188411091358084838_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
> _______________________________________________
> Fis mailing list
> Fis at listas.unizar.es
> http://listas.unizar.es/cgi-bin/mailman/listinfo/fis
> ----------
> INFORMACIÓN SOBRE PROTECCIÓN DE DATOS DE CARÁCTER PERSONAL
>
> Ud. recibe este correo por pertenecer a una lista de correo
> gestionada por la Universidad de Zaragoza.
> Puede encontrar toda la información sobre como tratamos sus datos
> en el siguiente enlace:
> https://sicuz.unizar.es/informacion-sobre-proteccion-de-datos-de-caracter-personal-en-listas
> Recuerde que si está suscrito a una lista voluntaria Ud. puede
> darse de baja desde la propia aplicación en el momento en que lo
> desee.
> http://listas.unizar.es
> ----------
>
>
> _______________________________________________
> Fis mailing list
> Fis at listas.unizar.es
> http://listas.unizar.es/cgi-bin/mailman/listinfo/fis
> ----------
> INFORMACIÓN SOBRE PROTECCIÓN DE DATOS DE CARÁCTER PERSONAL
>
> Ud. recibe este correo por pertenecer a una lista de correo gestionada por la Universidad de Zaragoza.
> Puede encontrar toda la información sobre como tratamos sus datos en el siguiente enlace:https://sicuz.unizar.es/informacion-sobre-proteccion-de-datos-de-caracter-personal-en-listas
> Recuerde que si está suscrito a una lista voluntaria Ud. puede darse de baja desde la propia aplicación en el momento en que lo desee.
> http://listas.unizar.es
> ----------
--
/Dr. Eric Werner
Oxford Advanced Research Foundation
https://urldefense.com/v3/__https://oarf.org__;!!D9dNQwwGXtA!RyX3wbe3vhk-4Ga4WIVvfN5V2-Ye9zZCRGN6PHa6oAokInl7rTYH0Rk8GhKak5eqZhUKeqhECQ3TBFlGMGB2u7I$
/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listas.unizar.es/pipermail/fis/attachments/20231026/47a43a58/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 5C75AE29 at 656B1E7E.55D1306500000000.png
Type: image/png
Size: 31269 bytes
Desc: not available
URL: <http://listas.unizar.es/pipermail/fis/attachments/20231026/47a43a58/attachment-0001.png>
More information about the Fis
mailing list