Juan Sequeda

Just how old is the idea of a "semantic layer", and using it to "talk to your data"?

At the beginning of this year, I started a book club for Data & Reality by Bill Kent. (you've probably seen me posting about it!)

The 1st edition of Data & Reality was published in 1978, the 2nd edition in 1998.

This is cool Catalog and Cocktails #HonestNoBS Data podcast was just ranked #9 on the list of the top 100 data science podcasts! Thanks to all our listeners and guests!!

www.millionpodcasts.com/data-science...

The Five Laws of Data Enablement: How the father of library science would make his data team indispensable - Locally Optimistic In 1931, S. R. Ranganathan established five laws of librarianship that any modern data leader would be wise to embrace.

And I love the big shoutout @juansequeda.bsky.social gives to Amalia Child, and her article "The Five Laws of Data Enablement"

Read it here: locallyoptimistic.com/post/the-fiv...

@olesenbagneux.bsky.social and @jtalisman.bsky.social gave LIS such amazing representation at Data Day Texas this year in their talks, a trend I am sure will continue.

🚀 How to Start Investing in Semantics and Knowledge: A Practical Guide

Today at @datadaytexas.bsky.social
I shared practical advice based on my experience on a topic I’m deeply passionate about: elevating the need of semantics and knowledge in the enterprise

key takeaways from my talk:

Ringing in the new year with my Rosetta Stone socks, the first semantic layer.

Let’s start working towards making 2025 the year of semantics and knowledge

Observing the same

bsky.app/profile/juan...

The Next Great Leap in AI Is Behind Schedule and Crazy Expensive The startup has run into problem after problem on its new artificial-intelligence project, code-named Orion.

OpenAI is reminding me what Cyc was doing in the 80s and 90s

“OpenAI has worked with experts in subjects like theoretical physics, to explain how they would approach some of the toughest problems in their field. This can also help Orion get smarter. “

www.wsj.com/tech/ai/open...

Spoiler alert: Data engineering is harder than you think— and here's why.

This quote summarizes the issue:

"It is not the domain experts' knowledge that goes to production, it is the assumption of the developers" - Alberto Brandolini (EventStorming Creator)

Cognitive burden is too high!

Thank Big Data for Killing Off the Worst Part of Thanksgiving Meal Prep Butterball started out with a boring, standard software implementation—and ended with the creation of a Thanksgiving turkey that you do not need to thaw.

“At the turkey-processor Butterball, it took advanced analytics systems and some upgraded data plumbing to uncover a hidden but universal truth: people hate thawing their birds.”

No sh********t!!!

And you are proud of the $ spent to come to that conclusion?

www.wsj.com/articles/tha...

Doing the knowledge work. Modeling is one part. Talking to different stakeholders to understand what X means and figuring out where it is in the data. Today that work is done but it’s very ad hoc and ends up being technical focus and lacks the social/people side (talking to the biz)

Our podcast episode with Ethan Mollick on AI was 🔥 A highlight quote:

We are trained to make dumb systems smart. We are not used to making smart systems smarter. We end up making smart systems dumber.

We are focused/concerned so much on limiting LLMs and not leveraging their super powers

With LLMs serving as interfaces & KG handling context and integration, we now have the tools to fully realize this vision. My call to everyone reading this is to revisit the original Semantic Web vision and build on its decades of progress. Agents were always the goal it’s time to bring them to life

A Call for Agents: To those in the AI community, the idea of agents isn’t new. But the Semantic Web vision of autonomous agents, grounded in knowledge representation, planning, and data integration, provides a robust foundation for building the next generation of intelligent systems.

LLMs might suffice. Can LLMs do data integration Technically they are integrating data through training so depending on questions, it may be sufficient. But for rigorous tasks that enterprises require, knowledge graphs remain essential. Let’s not forget that LLMs are focused on generative tasks

LLMs Knowledge Graphs Debate Together w/ Sabrina Kirrane, we moderate a debate on Knowledge Graphs vs LLMs. The conclusion? It’s not an either/or question—it’s about understanding the task context. For “beer reasoning” where you take advice from a friend while drinking beer as Oscar Corcho put it,

Humans understand concepts. Goal is to connect a human concept to a machine weight
LLMs/Semantic Layers: Can LLMs bridge the silos created by diverse semantic layer/modeling tools? Discussions focused on interoperability, schema reuse, potential for LLMs to act as universal data model translators.

Explainability: What makes a “decent” explanation? Context and user expectations are critical. Explanations should balance transparency with usability, whether for bias mitigation, debugging, or trust-building. Explanations are for Human-AI communication. AI understands pixels, weights...

Mappings: From aligning ontologies to evaluating quality and governance, this discussion reinforced the need for human-in-the-loop approaches and the potential role of LLMs in automating parts of the process.

There was a special Dagstuhl style session on Knowledge Graphs and GenAI. I participated in three breakout sessions. The takeaways aren’t surprising, it’s validation that we are aligned on the problems and working on them

They’ve extended GQL and SPARQL to return paths in graph queries in a practical way. I expect to see graph databases implementing this approach. Also check out the Google paper “Relationships are complicated! An analysis of relationships between datasets on the Web” which won the Best Paper Award.

the original semantic web vision is achievable. LLMs can act as user interfaces, knowledge graphs handle data integration, and traditional agent components like planning can complete the picture.

Check out PathFinder: Returning Paths in Graph Queries which won the Best Student Paper Award

Ora reminded us that the original vision of the Semantic Web was centered on autonomous agents—making it inherently about AI. Due to the “AI winter,” the focus shifted away from agents. With LLMs and the scalability of RDF knowledge graphs (e.g., powering Amazon’s supply chain), ...

from multiple domains. Main takeaway is to think about data provenance and lineage through the lens of a supply chain.
Ora Lassila’s Keynote: This was a standout, not just because Ora and I are friends and co-authors, but because it revisited the past, present, and future of the Semantic Web.

direct mapping into graphs, followed by semantic integration and entity resolution. End-users, like journalists, prefer simple search interfaces over complex graph visualizations.
Chaitan Baru’s Keynote: NSF has a grand vision of creating an Open Knowledge Network which integrates knowledge graphs

Thread of my takeaways from the International Semantic Web Conference #iswc2024
@mioana.bsky.social Keynote: Great example of data integration for journalism, highlighting the power of graphs to combine heterogeneous sources like XML, JSON, CSV, and RDF. The approach involves ...

I created a starter pack on KGs! I found others useful as newcomer.

go.bsky.app/EQRCq9R