Re: perfect knowledge in AI

Well, even simply language and communication andthe simplest reasoning
require logic of course,  nobody can dis that.
But
a) logic itself depends on other things being true, such as fact checking,
There was a great un example once on ontolog list, where the first order
logic was being demonstrated (if  fred is a bat and  bats are birds, then
fred is a bird,)  that turned out a wrong answer because the premise was
false (bats  not birds but mammals). It was a good lesson, showing that the
logic is right but the fact is wrong

b) there is a  probably a lot more than logic to achieve perfection, I
posted an example on this list about the dying woman who, as last wish
asked someone to take her urine to the first person who entered the gate in
such and such place, who then saved her life. Such a request may not sound
logical at first, but probably her intuition was greater than anybody s
logic could have been, which became apparent at the end (the first person
who crossed the gate was a doctor who figured out the poison)
Here is the post
https://lists.w3.org/Archives/Public/public-aikr/2021Feb/0012.html

I dont know if that is what DR intended in his reply, but I  would agree
that even logic rests on correct and relevant facts, and possibly other
things

PDM

On Wed, May 11, 2022 at 3:12 AM Adeel <aahmad1811@gmail.com> wrote:

> Logic is testable and does take prior knowledge into account, precisely
> what inferencing process does through reasoning. You can have prior
> knowledge persisted in a KG that is always evolving via inference. The
> representation is machine-readable as well. In statistics you likely to
> have more uncertainties and inconsistencies in rigid models that are
> overfitted to the data where you are essentially dealing in half-truths of
> false-positives and false-negatives. Each time you repeat the same
> iteration you get a different output. Humans are perfectly rational without
> having to rely on statistics. In fact, logic can be formally tested via
> constraints. Statistics at most can be evaluated. On top of which you then
> need to build an explainability model for it.
>
>
> If you think logic is overrated then, how do you think the PC you are
> using came about? How do you think programming languages came about? Even
> your statistical model will rely on logics to process it which all hardware
> devices use and rely on.
>
> On Tue, 10 May 2022 at 11:11, Dave Raggett <dsr@w3.org> wrote:
>
>> With all due respect to Spock, logic is overrated as it fails to consider
>> uncertainties and has problems with inconsistencies, moreover, it doesn’t
>> take the statistics of prior knowledge into account. Logic is replaced by
>> plausible reasoning based upon rational beliefs, and taking care to avoid
>> cognitive biases.
>>
>> On 10 May 2022, at 05:06, Paola Di Maio <paoladimaio10@gmail.com> wrote:
>>
>> Adeel and all
>>
>> It must be logically correct, however logic must be built on a model that
>> is reasonably complete (for some purpose)\
>> In the sense that i aim to create say,  perfect knowledge about something
>> (or anything), unless the model is representative (adequate)
>> the logic (logical model) may not be complete
>> that  actually could be wrong (not adhere to truth)
>>
>> Even a logical correctness must be complete
>> to be adequate
>>
>> You can have a correct logic that is not adequate and far from perfect
>> (many policies are logically correct but do not cover the domain)
>>
>> Perfect knowledge, in the sense of knowledge that mirrors truth
>> without cognitive /subjective ileterin can be neither complete nor
>> correct and still be true.
>>
>> Just sayin.
>>
>> Maybe the notion of perfect knowledge has to be functionally defined to
>> be able to exchange meaningfully about it.
>>
>> On Mon, May 9, 2022 at 6:45 PM Adeel <aahmad1811@gmail.com> wrote:
>>
>>> No, perfection implies logical correctness within the defined
>>> constraints of representation, but under the open-world assumption, it
>>> should not imply completeness.
>>>
>>>
>>> On Mon, 9 May 2022 at 11:30, Paola Di Maio <paoladimaio10@gmail.com>
>>> wrote:
>>>
>>>>
>>>> Does Perfection imply completeness?
>>>> Discuss
>>>>
>>>>
>>>> On Mon, May 9, 2022 at 2:47 PM ProjectParadigm-ICT-Program <
>>>> metadataportals@yahoo.com> wrote:
>>>>
>>>>> If perfect implies complete, we can rule it out because of the proofs
>>>>> by Godel and Turing on incompleteness and undecidability.
>>>>> The concept of perfect only exists in mathematics with the definition
>>>>> of perfect numbers.
>>>>> Unbiased reasoning that leads to results for which truth values can be
>>>>> determined in terms of validity, reproducibility, equivalence and causal
>>>>> relationships are the best way to go for knowledge representation.
>>>>> Knowledge and for that matter consciousness as well are as yet not
>>>>> unequivocally defined, and as such perfection in this context is not
>>>>> attainable.
>>>>>
>>>>>
>>>>> Milton Ponson
>>>>> GSM: +297 747 8280
>>>>> PO Box 1154, Oranjestad
>>>>> Aruba, Dutch Caribbean
>>>>> Project Paradigm: Bringing the ICT tools for sustainable development
>>>>> to all stakeholders worldwide through collaborative research on applied
>>>>> mathematics, advanced modeling, software and standards development
>>>>>
>>>>>
>>>>> On Sunday, May 8, 2022, 04:15:01 AM AST, Paola Di Maio <
>>>>> paola.dimaio@gmail.com> wrote:
>>>>>
>>>>>
>>>>> Dave R's latest post  to the cog ai list reminds us of the ultimate.
>>>>> Perfect knowledge is a thing. Is there any such thing, really? How can it
>>>>> be pursued?
>>>>> Can we distinguish
>>>>> perfect knowledge rom its perfect representation
>>>>>
>>>>>
>>>>> Much there is to say about it. In other schools, we start by clearing
>>>>> the obscurations in our own mind  . That is a lifetime pursuit.
>>>>> While we get there, I take the opportunity to reflect on the perfect
>>>>> knowledge literature in AI, a worthy topic to remember
>>>>>
>>>>> I someone would like to access the article below, email me, I can
>>>>> share my copy
>>>>>
>>>>>
>>>>> ARTIFICIAL INTELLIGENCE 111
>>>>> Perfect Knowledge Revisited*
>>>>> S.T. Dekker, H.J. van den Herik and
>>>>> l.S. Herschberg
>>>>> Delft University of Technology, Department of Mathematics
>>>>> and Informatics, 2628 BL Delft, Netherlands
>>>>> ABSTRACT
>>>>> Database research slowly arrives at the stage where perfect knowledge
>>>>> allows us to grasp simple
>>>>> endgames which, in most instances, show pathologies never thought o f
>>>>> by Grandmasters' intuition.
>>>>> For some endgames, the maximin exceeds FIDE's 50-move rule, thus
>>>>> precipitating a discussion
>>>>> about altering the rule. However, even though it is now possible to
>>>>> determine exactly the path lengths
>>>>> o f many 5-men endgames (or o f fewer men), it is felt there is an
>>>>> essential flaw if each endgame
>>>>> should have its own limit to the number o f moves. This paper focuses
>>>>> on the consequences o f a
>>>>> k-move rule which, whatever the value o f k, may change a naive
>>>>> optimal strategy into a k-optimal
>>>>> strategy which may well be radically different.
>>>>> 1. Introduction
>>>>> Full knowledge of some endgames involving 3 or 4 men has first been
>>>>> made
>>>>> available by Str6hlein [12]. However, his work did not immediately
>>>>> receive the
>>>>> recognition it deserved. This resulted in several reinventions of the
>>>>> retrograde
>>>>> enumeration technique around 1975, e.g., by Clarke, Thompson and by
>>>>> Komissarchik and Futer. Berliner [2] reported in the same vein at an
>>>>> early date
>>>>> as did Newborn [11]. It is only recent advances in computers that
>>>>> allowed
>>>>> comfortably tackling endgames of 5 men, though undaunted previous
>>>>> efforts
>>>>> are on record (Komissarchik and Futer [8], Arlazarov and Futer [1]).
>>>>> Over the
>>>>> past four years, Ken Thompson has been a conspicuous labourer in this
>>>>> particular field (Herschberg and van den Herik [6], Thompson [13]).
>>>>> As of this writing, three 3-men endgames, five 4-men endgames, twelve
>>>>> 5-men endgames without pawns and three 5-men endgames with a pawn [4]
>>>>> can
>>>>> be said to have been solved under the convention that White is the
>>>>> stronger
>>>>> *The research reported in this contribution has been made possible by
>>>>> the Netherlands
>>>>> Organization for Advancement of Pure Research (ZWO), dossier number 39
>>>>> SC 68-129, notably
>>>>> by their donation of computer time on the Amsterdam Cyber 205. The
>>>>> opinions expressed are
>>>>> those of the authors and do not necessarily represent those of the
>>>>> Organization.
>>>>> Artificial Intelligence 43 (1990) 111-123
>>>>> 0004-3702/90/$3.50 © 1990, Elsevier Science Publishers B.V.
>>>>> (North-Holland)
>>>>> 112 S.T. D E K K E R ET AL.
>>>>> side and Black provides optimal resistance, which is to say that Black
>>>>> will delay
>>>>> as long as possible either mate or an inevitable conversion into
>>>>> another lost
>>>>> endgame. Conversion is taken in its larger sense. It may consist in
>>>>> converting
>>>>> to an endgame of different pieces, e.g., by promoting a pawn; equally,
>>>>> it may
>>>>> involve the loss of a piece and, finally and most subtly, it may
>>>>> involve a pawn
>>>>> move which turns an endgame into an essentially different endgame: a
>>>>> case in
>>>>> point is the pawn move in the KQP(a6)KQ endgame converting it into
>>>>> KQP(a7)KQ (for notation, see Appendix A).
>>>>> The database, when constructed, defines an entry for every legal
>>>>> configura-
>>>>> tion; from this, for each position, a sequence of moves known to be
>>>>> optimal
>>>>> can be derived. The retrograde analysis is performed by a full-width
>>>>> backward-
>>>>> chaining procedure, starting from definitive positions (mate or
>>>>> conversion), as
>>>>> described in detail by van den Herik and Herschberg [18]; this yields
>>>>> a
>>>>> database. The maximum length of all optimal paths is called the
>>>>> maximin (von
>>>>> Neumann and Morgenstern [16]), i.e., the number of moves necessary and
>>>>> sufficient to reach a definitive position from an arbitrary given
>>>>> position with
>>>>> White to move (WTM) and assuming optimal defence
>>>>> CONCLUSION
>>>>> It has now become clear that the notion of optimal play has been
>>>>> rather naively
>>>>> defined so far. At the very least, the notion of optimality requires a
>>>>> specific
>>>>> value of k for k-optimality and hence a careful bookkeeping of all
>>>>> relevant
>>>>> anteriorities. These additional requirements form but one instance of
>>>>> aiming to
>>>>> achieve optimal play under constraints; of such constraints a k-move
>>>>> rule is
>>>>> merely one instance. In essence, it is not our opinion that a k-move
>>>>> rule spoils
>>>>> the game of chess; on the contrary, like any other constraint, it may
>>>>> be said to
>>>>> enrich it, even though at present it appears to puzzle database
>>>>> constructors,
>>>>> chess theoreticians and Grandmasters alike.
>>>>>
>>>>>
>>>>>
>> Dave Raggett <dsr@w3.org> http://www.w3.org/People/Raggett
>> W3C Data Activity Lead & W3C champion for the Web of things
>>
>>
>>
>>
>>