Re: The future of KR in retrospective

Le jeu. 27 juin 2019 à 09:35, Paola Di Maio <paola.dimaio@gmail.com> a
écrit :

> Any comments on how the predictions in this article performed?
>
> The Future of Knowledge Re resent at ion Extended Abstract Ronald J.
> Brachman
>
> https://pdfs.semanticscholar.org/1f94/b9ed37cdac08d8f1732e9a49c2f735bb28fb.pdf
>
>
> PDM
>

I continue my (shallow) analysis on the predictions.

Regarding the predictions at some point Ronald J. Brachman writes that
versioned database will be thing I don't think it happened. Datomic is a
thing and has some commercial success but it is not a KB/KR software per se.

Regarding KB in general, at least as source available there is little one
can use:

- Concept Net
- WordNet
- WikiData, even if I did not manage to find useful ways to use it
- DBpedia, same as WikiData, difficult to use imo

Maybe that is enough?

Large KB are a thing like wikidata or DBpedia, biggest KB are closed source.

Another thing that Brachman did predict is that it would be possible to
interop several KB like RDF federation queries. But to some extent it
doesn't work really well (imo again).

> KR systems will increasingly find their way into the hand of
"non-professionals"?

That is true and false. There is more people using KR. I would not say they
are not professional but more like new kind of professionals that are
specifically trained to work on specific KR, instead of KR theories. There
is also the personal assistant thing which more or less allow to access a
KB and the Internet in general using NL.

> No HAL.

I can not confirm, I don't know HAL.

> Can we have realistically useful KB's that are designed in the absence of
specific intended applications?

It is not clear to me what are the "intended applications" while NL to KB
query is thing in personal assistant. They do not project the overall
intent of the user. One thing where it is particularly striking is search
engines: all searches seems independent, but in reality it is not true.
Search engines, fail to link the long-running intent of the user. They do
not capture a user's project or activity.


I think the predictions are mostly accurate 8/10


Something that strikes me is that the KR community, at least through this
text doesn't link KR or KB with the Memex
<https://en.wikipedia.org/wiki/Memex>.

Another thing that I can not find in the paper while just scanning
is the poly-structure and poly-type nature of KR systems. Some part of the
KB rely on space, time and truth values as numbers. Others are symbolic.
Maybe it is too basic to be stated. But as far as I am concerned it has
been difficult to find a system that allows to represent everything (I
think) I need. I mean the tools were not here until recently.


I would like to read more. I am especially interested in non-monotonic
logic.


-- 
Amirouche ~ amz3 ~ https://hyper.dev