Re: the intersection between AIKR and COGAI

Hello,

Vectors and features to a ML model are not knowledge they are just data.
The ML model processes data and outputs data. The most you can get from ML
model is to go from data to either data or some form of information that a
human interprets. If you want knowledge the data would have to contain some
representation that is machine-readable, which the machine can infer on new
knowledge, often to aid in the semantic context, and going further in
generalizability and transfer learning of the said ML model. The former set
of steps then become a Hybrid AI of combing KR + ML. Otherwise, without the
representation the ML model has no clue as to what semantic meaning is held
in the data without further recourse in training.

Thanks,

Adeel



On Mon, 31 Oct 2022 at 09:23, Dave Raggett <dsr@w3.org> wrote:

> I don’t understand what you are trying to say.  Knowledge can be expressed
> symbolically as with RDF and Predicate Calculus, or in a distributed way
> when using vectors in noisy high dimensional spaces, as is the case with
> neural networks, both artificial and biological. Stable Diffusion knows the
> human faces have two eyes, two ears and one nose, as well as the variations
> in their shapes.  It isn’t using symbols, though, despite knowing the
> differences between humans, dogs and cats. Whilst it recognises the word
> “dog”, it does so by mapping it to a vector space for a latent
> representation of meaning, avoiding symbols.  Reasoning can likewise be
> implemented in vector spaces without resorting to symbols. Human languages
> uses words, which can be thought of as symbols, but the closer you look at
> them, the fuzzier they are, as the meaning is context dependent and hard to
> pin down.
>
> On 30 Oct 2022, at 20:41, ProjectParadigm-ICT-Program <
> metadataportals@yahoo.com> wrote:
>
> I beg to differ on KR is by definition symbolic. Is is slightly more
> complicated. Its is a question of signs, symbols, concepts and how to
> encode (assigned) meaning. And consequently the concept of languages is
> also slightly more complicated.
>
> Chomsky, Saussure and Peirce basically define our current scope on
> linguistics, and semiotics therein, but we use artificial languages with
> symbols in mathematics, logic, physical sciences, computational
> linguistics, computer science and NLP.
>
> The discussion here is more of a philosophical nature, but is essential.
> Because we intend AI to be open, inclusive and explainable, the KR must
> reflect this as well.I don’t
>
>
> Dave Raggett <dsr@w3.org>
>
>
>
>