Re: New SWCLOS community to reactive this awesome project

[Xiaming Chen]

Dear Seiji and Chun Tian,

Appreciate your response a lot. Yet sorry for a late reply. This week is busy with job stuff. My hired company, Alibaba, is going through one of the most important sales festivities, The Double-Eleven, similar to western dark friday. Our team takes responsibility to keep the big data system stable.

Back to our SWCLOS project. Yes, I have recently noticed the release of Allegro CL 11.0 with capabilities of talking to LLM (specific to OpenAI for now). I also went through the neuro-symbolic vision of Allegro Inc. Their infrastructures like Allegro Graph, CL and LLM are fabulous. Before discovering AllegroGraph and SWCLOS, I have been experimenting on semantic web with the Jena framework for a while, because Jena provides a mature and complete implementation of OWL standards. But finally gave up with its verbose Java interfaces and inability to fulfill the `meta programming` requirement of my own project (LambdaCogito) and migrated to Lisp now.

In my opinion about neuro-symbolic AI, a unified representation bridging the neuro-world's vector and symbolic-world's discrete symbols lies at the core. Thus my research focus is on knowledge representation. I want to design a dynamic virtual memory system (with a coined name "Gödel Virtual Memory") based on a unified representation of knowledge, self-driven learning and reasoning capabilities. This idea is not brand-new. Similar concepts are depicted in academic cognitive architectures like Global Workspace Theory, the Conscious Turning Machine, and Lecun's World Model. No matter what kind of architecture, the hardest part is to implement a software or hardware system to bring these theories alive. I am still towards this goal. Lisp is chosen because of its meta programming to enable the GVM to (possibly) modify its behavior dynamically with accumulated knowledge over time. SWCLOS lies at the core of bridging the semantic web technologies and Lisp objects. Maybe in the future, these Lisp objects could interact and compute with vectors generated by neural networks. I still have no idea of how they would map into a unified space, but I have a faith that is possible. Worth mentioning that another team is also working on a similar idea in OpenCog Hyperon (as attached in case you are interested).

As for the development of SWCLOS, I am glad to share some ideas:

• Enhance the software engineering as quicklisp package and porting to CL releases. (boring but solid)
• Involve the kbpedia ontology and make it as the backbone of SWCLOS's knowledge network (KN)
• Create a static compiling tools/methods to compile semantic web representation into SWLOCS form and linked to the backbone KN automatically.
• (maybe hard to fulfill) Create a JIT compiling mechanism to consume and generate compatible knowledge representation.

Your thoughts and feedback are welcome. I look forward to hearing from you!

Best,

Jamin Chen

https://xiaming.site/about

On Wed, 30 Oct 2024 at 22:01, Seiji Koide <seiji...@gmail.com> wrote:

Dear Xiaming Chen and Chun Tian

This is Seiji Koide, the creator of the original SWCLOS. I was very surprised at this email, which also pleased me.

Surely, I have developed SWCLOS on top of Allegro Common Lisp. The reason is that the Franz included  modern lisp in Allegro, in which the downcase alphabetical character of the symbol is preserved. I, a couple of times, intended SWCLOS porting to another Lisp system, for example SBCL. However, I, in the end, abandoned such a plan.  

You, Xiaming Chen and Chun Tian and anyone who wants to modify SWCLOS, may do it anyhow and anything. But I ask you to keep naming the system name SWCLOS.

I am now tackling to develop cl-swp, apart from CLOS, but deeply connected Lisp programming for Semantic Webs.

If you have any questions on SWCLOS, please let me know.

You mentioned ' neuro-symbolic AI', and you look like SWCLOS suggests and lightens the way of  ' neuro-symbolic AI', do you?

Do you have some idea of neuro-symbolic AI on SWCLOS?

Do you know that Franz delivered Allegro 11.0 as the ultimate platform in the age of neuro-symbolic AI?

See https://franz.com/products/allegrocl/

Let us keep contact loosely

Sincerely yours

Seiji Koide

2024年10月30日(水) 12:47 Xiaming Chen <chen...@gmail.com>:

Hi Seiji Koide, Chun Tian,

Hope this email finds you well. Recently I am working on a neuro-symbolic AI project and came up with an idea to combine Semantic Web and the powerful LISP language. Luckily the SWCLOS project shows up after a quick googling. Thanks for Seiji's original ideas and efforts and Chun's porting to more CL implementations and combination of OpenCyc. 

In my opinion, this idea and project would become one of the fundamental infras of future explainable AI (XAI), far beyond current transformer-based large language models. The Lisp language also has the advantage of programmable nature on the fly, whose CLOS-based interface would bring about more imagination on the self-evolution of AI systems. These instincts may encourage us to renew the SWCLOS project to a new level, with more capabilities to integrate with modern AI technologies/systems.

You guys both made great contributions to SWCLOS and I found current development is independently progressing in separate projects. Here the community of SWCLOS is proposed to merge your fantastic prior work into this community and make it active. We should keep a consistent license with previous decorations.

New project repo: https://github.com/SWCLOS/SWCLOS

This is a fork from Chun's latest commit with a shallow clone (to eliminate pull volume and give it a refresh start). This repo still lacks two Seiji's recent fixes on Namespaces. I have sent invitation messages on Github to you both (Chun already in) and make you the owner of this community. Hope we become a strong team to evolve this project further.

Cheers,

Jamin Chen

https://xiaming.site/about

[Seiji Koide]

Hi, Jsmin Chen

Thank you for your sophisticated response for my naive question and introducing a paper of OpenCog.

I see that your concern and feeling is very close to me.  "Gödel Virtual Memory", what sense do you indicate in it?  Anyway, very interesting!!

If you have not read my paper " Inquiry into RDF and OWL Semantics", please read what I attached to this email.

I expect that you can evaluate and appreciate the content of it.

The consideration has come up from my study of modern set theory and Goedel's incomplete theorem by Lisp implementation.

These series are on my  Blog "Semantic Web Diary", unfortunately, these articles  are in Japanese.

http://blog.livedoor.jp/s-koide/archives/1918908.html and http://blog.livedoor.jp/s-koide/archives/1904519.html

You know self-referencing or self-mentioning is troublesome on set theory and type theory and AI.

You may agree with the opinion that self-mentioning is a key of AGI.

Best.

Seii Koide

2024年11月9日(土) 0:18 Xiaming Chen <chen...@gmail.com>:

[Xiaming Chen]

Dear Seiji and Chun Tian,

Thanks for Seiji's sharing of sharp insight on self-reference and demonstration on the relation between set theory and lisp. I went through your blogs with the help of ChatGPT. I agree with you that self-reference is vital for future AGI, if we would get there someday. It basically determines if we can get a consistent and normal knowledge base. Meanwhile, Godel's incompleteness theorem and set theory paradoxes both highlight the constraints of formal logic on a general description of knowledge. Maybe some probabilistic element should be involved and to be compatible with the self-reference paradoxes. This may be one of the obstacles towards a real powerful and useful AI system.

The name of GVM comes from the Godel machine, which was a proposed self-evolving machine based on meta-learning. In a primitive version, GVM would sit on the basis of set and category theory, by integrating existing semantic web knowledge. But it is still under design and I will share more details in the near future when I get some progress. 

I am also interested in your new project cl-swp on semantic web. Can you share more about your vision or progress? 

As for SWCLOS, I am planning to continue Chun's work to finish the porting to SBCL and quicklisp first while training myself as an expert of Lisp. I have abundant experience in object-oriented programming. Lisp is still a new frontier for me. Hope you can help me during the progress.

Best,

Jamin

Seiji Koide <seiji...@gmail.com> 于 2024年11月9日周六 15:21写道：