AI definition

[Alex Shkotin]

"Bad definition is better than absence of ≝ . As it is a milestone on the way to theory i.e. systematic knowledge" 

"Artificial intelligence (AI), in its broadest sense, is intelligence exhibited by machines, particularly computer systems, as opposed to the natural intelligence of living beings."

https://en.wikipedia.org/wiki/Artificial_intelligence

Alex

[Nadin, Mihai]

Dear and respected Alex Shkotin,

 

There is no such thing as BAD DEFINITION. To define is to express your knowledge about what you define or intend to make happen. There is, however, ignorance expressed in some definitions (“junk-in, junk-out”). You just produced an example. Wikipedia is not a repository of knowledge; it reflects our own ignorance.

 

Mihai Nadin

--
All contributions to this forum are covered by an open-source license.
For information about the wiki, the license, and how to subscribe or
unsubscribe to the forum, see http://ontologforum.org/info/
---
You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/CAFxxROQtfTUP9nF3LgC8dfeaA8qn8ZA5UfKhhy4eTPHkur87yA%40mail.gmail.com.

[John F Sowa]

Alex,

There are two kinds of research in every branch of science, engineering, or endeavors in every field of every kind:  (1) improvements in existing techniques or (2) fundamental research that breaks all the established rules and definitions of the field -- including statements about the value of the field itself.

Type 1 research improves existing methods.  They develop evolutionary changes that work within a known field.  Type 2 research tries to make existing methods obsolete or obsolescent.  Definitions are useful for novices that want to work in a well-established field.  But those definitions may put blinders on students and discourage them from looking for totally knew methods that create revolutions.   

As I wrote in previous notes, I think that current versions of neural networks + LLMs are OK for what they do well.  But they are pathetically weak in comparison to the neuroscience that supports human intelligence.  And they are dangerously bad  in creating errors or hallucinations.  That doesn't mean that they're bad for what they do very well.  But it means that it's important to look for new inspiration elsewhere for what they do moderately well, but not great.

Some people claim that LLM techniques are wonderful because they excel in passing tests, such as bar exams and IQ tests.   But that proves nothing.  The answers to those things are on the internet, and the computer programs found them after an exhaustive search.  Big deal!  They did just as well as any body who had access to the same data.

I would recommend examples to definitions -- especially examples that show novel combinations of technology instead of conventional applications to same-old, same-old kinds of problems.

And very often, it's important to revive even older methods that had been impractical with old technology, but which may be fantastic today.   For example, the decidability gang designed OWL with horrible restrictions because they were afraid of problems that might be undecidable.

But today, a cell phone can test an algorithm that formerly took an hour on computers of the 1960s, but just a second today.  What that means is that if a program gets an undecidable problem that wasted a hour of time on a very expensive computer in the 1960s, it will only waste a second on a cell phone.  That's no big deal.  Just include a timer.  If some proof takes more than a second, switch to something else.

Fundamental principle:  Examples are better than definitions.  A new discovery won't make examples obsolete.  But it can make definitions of recommended procedures obsolete.

John

 

---

From: "Alex Shkotin" <alex.s...@gmail.com>

[Alex Shkotin]

Colleagues,

In yesterday's session, Todd Schneider raised a question to JFS first of all, about the definition of the AI term. It was as usual some kind of infinite discussion may be for 5 or 10 minutes. We may see it in the transcript after some time. 

Request for definition raises a lot of interesting ideas. 

I thought it may be interesting to continue it here.

For as it's fixed on our Summit sessions one of the tasks of ontologists is to get consensus knowledge for community and humanity.

Wikipedia keeps a kind of consensus. 

For definition of term "bad" used, let me refer not to my favorite m-w.com, where we have a lot of definitions but just to the google with a search string "bad definition" and the answer:

So let me reformulate a maxima: 

"poor quality or a low standard definition is better than absence of ≝. As it is a milestone on the way to theory i.e. systematic knowledge"

There are different kinds of definitions and this is a great topic as definitions are a backbone of a theory framework [1].

And we should talk not just about examples [JFS] but about entities, structures our theory is about. As any theory is about something which usually exists.

But necessarily in our world, like centaur ⛲

From my contemporary research, which I hope to share in a near future, let me point out how many nice definitions we have in Geometry where we began with Euclid's definitions and got great progress with Hilbert's one. Like this

"A set of points is an angle iff it consists of two rays intersecting only by their starting points and not lying on the same straight line."

You may look at work in progress for Hilbert's axiomatic theory of Geometry framework [2].

And we know that geometry is a tool for Mechanics 🍀

Alex

[1] https://www.researchgate.net/publication/374265191_Theory_framework_-_knowledge_hub_message_1

Storing the theory of a particular subject area in one place and maintaining it (including formalization) through collective efforts is easily possible with the modern development of technology. The concentration and verification of knowledge achieved in this case should give a powerful ordering of theoretical knowledge, which will facilitate their formalization, i.e. mathematical notation, and therefore algorithmic processing in many cases, up to the semi-automatic proof of various kinds of consequences, for example, theorems. This message describes what the framework of the theory is, intended for unified storage and collective accumulation of its results.

[2] https://drive.google.com/drive/folders/1d3Bmi5sDq5_NoW47hQde3g2bpWkeqcnz 

ср, 3 апр. 2024 г. в 20:18, Alex Shkotin <alex.s...@gmail.com>:

[alex.shkotin]

IN ADDITION have a look at discussion [obo-discuss] Evaluating natural language ontology definitions from a generative large language model 

https://groups.google.com/g/obo-discuss/c/wrMy5fmII4Y/m/DGBv0iCWAgAJ

четверг, 4 апреля 2024 г. в 11:47:54 UTC+3, alex.shkotin:

[alex.shkotin]

I apologize for misprinting. Correct sentence is "But NOT necessarily in our world, like centaur ⛲" of course.

четверг, 4 апреля 2024 г. в 11:47:54 UTC+3, alex.shkotin:

Colleagues,

[Alex Shkotin]

good point about intelligence: 

" AI refers to the concept of computers capable of performing tasks that would otherwise require human intelligence, such as decision making and NLP."

https://www.techrepublic.com/article/what-is-generative-ai/

Alex

ср, 3 апр. 2024 г. в 20:18, Alex Shkotin <alex.s...@gmail.com>:

"Bad definition is better than absence of ≝ . As it is a milestone on the way to theory i.e. systematic knowledge" 

[John F Sowa]

Alex,

The best starting point for defining any term (word or phrase) is the set of all citations (sentences) in which that term is used.  That is the source material for the best unabridged dictionaries,  (Second-rate dictionaries just take definitions from the best dictionaries, select a subset, and adapt it as they wish.)  

Given all that data for any particular term, the editors send the set of citations to a lexicographer who is an expert in the field that uses the term.  That expert subdivides the set of citations into subsets for each variation of meaning.  Finally, the expert for that word writes a definition that most accurately specifies the common meaning of all the citations in that subset.

End result:  A set of definitions for each of the many ways of using that term.  For the term 'artificial intelligence', you would get set of definitions for every way that the term has been used since it was coined at the meeting organized by John McCarthy in 1956.

If you want one general definition that covers all the special cases, the best short definition is very clear and very simple.  It's just two words long:  'artificial' + 'intelligence'.

That short definition plus the collection of all the special cases is clear, precise, and comprehensive.  You can't get better than that.

John

PS:  Of course, lexicographers also check the definitions they select by this process with other dictionaries to see if there are other word senses they may have missed.  Unless they deliberately copy definitions, that is not plagiarism.  When old dictionaries are prior to the copyright protections, free copying is permissible.  But they are required to cite their sources, possibly in a separate volume.

 

---

From: "Alex Shkotin" <alex.s...@gmail.com>

[Alex Shkotin]

John,

Absolutely agree, when we have to deal with common sense terms, i.e. how the term is used in everyday life. And what we have in the dictionary for AI see [1].

In sciences and technologies when we have systematic theoretical knowledge, it's the responsibility of an author or teacher of a theory to give definitions of terms.

As my task is knowledge formalization, I just ask the author to give me every definition separately, not embedded in the text, which is usual even for math.

Consider definition of angle by Hilbert [2]:

And definition of angle in a theory framework [3] for formalization to come: 

"A set of points is an angle iff it consists of two rays intersecting only by their starting points and not lying on the same straight line."

Here definition is a separate entity to keep and study and make it good.

This is what we are doing to formalize theoretical knowledge: we are looking for axiomatic theory and definitions in it [4].

The main idea is that we keep one theory framework for all around the Globe. And if somebody has another definition we hope that a hypothesis of equivalence is true.

Relationship between formal ontologies, as we have them for example in OBO Foundry, and theory framework and its definitions should be studied separately, I'll see for example this survey.

And the science behind GENO ontology is genomics. Do we have axiomatic theory for genomics? Is it possible to have an axiomatic theory for genomics?

What do you think?

Of course we have textbooks to begin with to create a genomics theory framework,  for example this one.

One framework and many textbooks - this is an ideal for knowledge concentration.

Alex

[1] https://www.merriam-webster.com/dictionary/artificial%20intelligence

[2] https://math.berkeley.edu/~wodzicki/160/Hilbert.pdf

[3] https://docs.google.com/spreadsheets/d/1zhVxfJNT2TZXFGtwrP0SI3mp9SoKVpRRocyJt7rQk90/edit#gid=0

rus	Совокупность точек есть угол еите она состоит из двух лучей пересекающихся только своими начальными точками и не лежащих на одной прямой.
eng

A set of points is an angle iff it consists of two rays intersecting only by their starting points and not lying on the same straight line.

yfl
yf0

[4] https://www.researchgate.net/publication/374265191_Theory_framework_-_knowledge_hub_message_1

пт, 5 апр. 2024 г. в 22:12, John F Sowa <so...@bestweb.net>:

--
All contributions to this forum are covered by an open-source license.
For information about the wiki, the license, and how to subscribe or
unsubscribe to the forum, see http://ontologforum.org/info/
---
You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/3db9d4adc4e941c68e76fbe45e2c2253%40bestweb.net.

[John F Sowa]

Alex,

Your example of a detailed mathematical definition shows what's possible at a very detailed level in a very precise field.  That is only possible in a highly detailed, highly precise, highly specialized subject.

I agree that that level of precision is important for that kind of topic.  But for a very broad subject, such as the totality of all the methods of AI (or any other broad subject), short and broad definitions are useful as summaries.

But if you wanted to define all the detailed methods of AI, you'd need a multi-volume encyclopedia of AI.  Some such things have been published.  But we're not going to do that in a summary of the recent summit.

John

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/CAFxxROTjdHwviXmHbs7SeRMpLAtY22uGJYgowzDav-dn09agLA%40mail.gmail.com.

[Alex Shkotin]

JFS:"That is only possible in a highly detailed, highly precise, highly specialized subject."

The subject is Geometry nevertheless. Geometry is thoroughly used by Mechanics. Mechanics is another highly detailed, highly precise, highly specialized subject.

Which is the base for Structural mechanics and other engineering disciplines.

All of them are highly detailed, highly precise, highly specialized.

I'll report more later, from the way of Mechanics and Structural mechanics theories frameworks creation.

Alex

вс, 7 апр. 2024 г. в 00:29, John F Sowa <so...@bestweb.net>:

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/4e1ff57f5aa141d39a4a4fe184c91f96%40bestweb.net.

[John F Sowa]

Alex,

Every precise specification of any kind can be translated to and from some version of logic.  First-order logic is the simplest (much simpler than OWL).  Common Logic is more expressive, and CL with the IKL extensions is even more expressive.

And by the way, when I say that FOL is simpler than OWL, I mean that for every notation for OWL there is a notation for FOL that is more readable and writeable.  Decidability is a major complication, not a simplification.

The decidability people were very knowledgeable logicians who did not have sufficient practical experience to understand the requirements for complex system design and development.  They hoped to force practitioners to swallow their theories.  In effect, they destroyed Tim B-L's vision for the Semantic Web.

[Chris Mungall]

On Mon, Apr 8, 2024 at 11:46 AM John F Sowa <so...@bestweb.net> wrote:

Alex,

Every precise specification of any kind can be translated to and from some version of logic.  First-order logic is the simplest (much simpler than OWL).  Common Logic is more expressive, and CL with the IKL extensions is even more expressive.

And by the way, when I say that FOL is simpler than OWL, I mean that for every notation for OWL there is a notation for FOL that is more readable and writeable.

VertebrateHand SubClassOf has_part min 3 Digit, has_part max 8 Digit

 

 Decidability is a major complication, not a simplification.

On this one, I am 100% with you. 

See slides 25-32 from my DeclMed talk

https://docs.google.com/presentation/d/16WgsN5AmKw65mpMX_Lwcjc4_ZbcKx2Y3WB4iEC9lY4g/edit#slide=id.g27cd596a4d0_0_218

OWL should have been syntactic sugar for a subset of CL. I tried to propose this in 2009, but people reacted in horror that such a system could be undecidable. 15 years later, we are unable to express the things we really need to and we all end up using EL++ reasoners like Elk for mid to large size ontologies anyway

 

The decidability people were very knowledgeable logicians who did not have sufficient practical experience to understand the requirements for complex system design and development.  They hoped to force practitioners to swallow their theories.  In effect, they destroyed Tim B-L's vision for the Semantic Web.

John

 

---

From: "Alex Shkotin" <alex.s...@gmail.com>

JFS:"That is only possible in a highly detailed, highly precise, highly specialized subject."

The subject is Geometry nevertheless. Geometry is thoroughly used by Mechanics. Mechanics is another highly detailed, highly precise, highly specialized subject.

Which is the base for Structural mechanics and other engineering disciplines.

All of them are highly detailed, highly precise, highly specialized.

I'll report more later, from the way of Mechanics and Structural mechanics theories frameworks creation.

Alex

вс, 7 апр. 2024 г. в 00:29, John F Sowa <so...@bestweb.net>:

Alex,

Your example of a detailed mathematical definition shows what's possible at a very detailed level in a very precise field.  That is only possible in a highly detailed, highly precise, highly specialized subject.

I agree that that level of precision is important for that kind of topic.  But for a very broad subject, such as the totality of all the methods of AI (or any other broad subject), short and broad definitions are useful as summaries.

But if you wanted to define all the detailed methods of AI, you'd need a multi-volume encyclopedia of AI.  Some such things have been published.  But we're not going to do that in a summary of the recent summit.

John

--
All contributions to this forum are covered by an open-source license.
For information about the wiki, the license, and how to subscribe or
unsubscribe to the forum, see http://ontologforum.org/info/
---
You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/36d870c8a6464dc0b67042fe73476d73%40bestweb.net.

[Alex Shkotin]

John,

Exactly! We discussed three years ago the advantages of formal definitions in our Ontology Summit session.

Keeping each definition as a separate block, in various languages, including formal ones, is one of the tasks of the theory framework. But not just one or more definitions, but in their interrelation within the framework of the theory.

Here is a page I made specifically to show how a definition block can be embedded on a website in the text where the definition needs to be shown.

https://sites.google.com/view/open-science-2018/sciencies/ugt [1]

Give me the same definition on CL and I'll add it as a line in this block of Undirected graph theory framework with my pleasure.

And I am personally for HOL, as English is a HOL,  you know. 

And the first task is to study what kind of theoretical knowledge we have for one or another subject. And then during formalization we can look at what to do if formal theory is inevitably undecidable.

Well, in this case we ought to make proofs manually, as usual.

Let me point out that in hets.eu or DOL projects there are a lot of formal languages collected. For me there is a lack of theoretical knowledge which is ready to be formalized. Most of this knowledge is recollected in formal ontologies, but we need formal theories.

Alex

[1]

пн, 8 апр. 2024 г. в 21:46, John F Sowa <so...@bestweb.net>:

Alex,

Every precise specification of any kind can be translated to and from some version of logic.  First-order logic is the simplest (much simpler than OWL).  Common Logic is more expressive, and CL with the IKL extensions is even more expressive.

And by the way, when I say that FOL is simpler than OWL, I mean that for every notation for OWL there is a notation for FOL that is more readable and writeable.  Decidability is a major complication, not a simplification.

The decidability people were very knowledgeable logicians who did not have sufficient practical experience to understand the requirements for complex system design and development.  They hoped to force practitioners to swallow their theories.  In effect, they destroyed Tim B-L's vision for the Semantic Web.

John

 

---

From: "Alex Shkotin" <alex.s...@gmail.com>

JFS:"That is only possible in a highly detailed, highly precise, highly specialized subject."

The subject is Geometry nevertheless. Geometry is thoroughly used by Mechanics. Mechanics is another highly detailed, highly precise, highly specialized subject.

Which is the base for Structural mechanics and other engineering disciplines.

All of them are highly detailed, highly precise, highly specialized.

I'll report more later, from the way of Mechanics and Structural mechanics theories frameworks creation.

Alex

вс, 7 апр. 2024 г. в 00:29, John F Sowa <so...@bestweb.net>:

Alex,

Your example of a detailed mathematical definition shows what's possible at a very detailed level in a very precise field.  That is only possible in a highly detailed, highly precise, highly specialized subject.

I agree that that level of precision is important for that kind of topic.  But for a very broad subject, such as the totality of all the methods of AI (or any other broad subject), short and broad definitions are useful as summaries.

But if you wanted to define all the detailed methods of AI, you'd need a multi-volume encyclopedia of AI.  Some such things have been published.  But we're not going to do that in a summary of the recent summit.

John

--
All contributions to this forum are covered by an open-source license.
For information about the wiki, the license, and how to subscribe or
unsubscribe to the forum, see http://ontologforum.org/info/
---
You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/36d870c8a6464dc0b67042fe73476d73%40bestweb.net.

[Alex Shkotin]

Chris,

The main crucial idea is to begin formalizing theories. 

By the way, why don't formal ontologists use Isabelle, Coq, HOL4, etc.? Yes, we may have to prove manually - is that a problem?

Well, if your favorite language is Obolog, give me a definition of the degree of a vertex in a graph in it and I will place it in the framework 🎣

It is also important how the structures are set and what is calculated on them.

Accordingly, we need not only a framework of theory but also a framework of tasks. I will soon publish a description of the framework of specific problems solved using graph theory. English is not ready in full.

When we build a theory it may turn out to be undecidable. Is this possible for the natural sciences? But first we need a theory and then its properties. After all, for example, Tarski proved that planimetry is decidable, but this did not find practical application.

First we need to write out the theory on natural language, and then see how to formalize it.

And theory begins with a system of precise definitions kept in the theory framework.

Alex

пн, 8 апр. 2024 г. в 22:15, Chris Mungall <cjmu...@lbl.gov>:

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/CAN9AifsHwSPsEExZ09WtUhxG_jMo-hLmbzTw29yrt3jjb_NQnQ%40mail.gmail.com.

[John F Sowa]

Alex,

Common Logic supports a kind of second-order logic for quantifying over functions and predicates.  And it does so with a theorem-proving methodology that does not go beyond the complexity of first-order logic.

Furthermore, CL is specified in an abstract syntax that may be represented in any kind of notation (graphic or linear) that has a precise mapping to and from the abstract syntax.  You can, if you wish, specify an English-like or a Russian-like syntax.  If you have a precise mapping to and from the abstract syntax, you can call it a dialect of Common Logic.

John 

 

---

From: "Alex Shkotin" <alex.s...@gmail.com>
Sent: 4/9/24 6:24 AM
To: ontolo...@googlegroups.com
Subject: Re: [ontolog-forum] Re: AI definition

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/CAFxxRORQUSBU40T%2B-hVgS44krGvg0_aBNqoEnKP-9RDCA4uecA%40mail.gmail.com.

[Alex Shkotin]

John,

Wonderful! This is what the framework of the theory is intended for: to keep in one block for each definition all its possible formalizations and all its equivalent definitions in different natural languages.

To add the definition of the degree of a vertex in a graph to the framework of the theory (this is just one line), we need an enthusiast who knows CL.

On the other hand, there are more people who know OWL2. And maybe there will be someone who will add a line in OWL2.

My favorite language is YAFOLL. And there is a line there with code "yfl".

I myself am probably going to add a formalization on Isabelle somewhere in the summer where they have a formalization of a large piece of the theory of undirected graphs.

But now I need to make a framework for the axiomatic theory of Mechanics, since this is an important transition to axiomatization in physics.

The theory of undirected graphs is one of the simplest to axiomatize and formalize, but with the widest application.

Alex

вт, 9 апр. 2024 г. в 20:56, John F Sowa <so...@bestweb.net>:

To view this discussion on the web visit https://groups.google.com/d/msgid/ontolog-forum/0eef69de3d814cde8696b91520379da3%40bestweb.net.