A couple of folks on LinkedIn have commented recently that they don’t believe the version of contrarian AI that I described in my last post is actually contrarian. One of these people talked about using an RDF database and associated tooling. He didn’t feel it was contrarian to have done so.

Webby contrarianism: Still relevant

In response, I’d say this. Consider what the bigger picture is for a moment.  We’ve had the World Wide Web for 32 years now.  In spite of the web’s ubiquitousness,  much enterprise architecture is still in essence pre web. it’s still heavily centralized, and many of the web standards that encourage decentralization have not been adopted..

Enterprises have had a chance for 32 years now to adopt a web architecture focused on data decentralization, contextualization and sharing. Instead, most have chosen mainly to perpetuate the status quo: Data siloing, duplication and code sprawl. To my mind, an application-centric mentality still encourages companies to ignore, de-emphasize or cover over evident problems at the data layer. Such mindless conformity obviously doesn’t benefit business or society as a whole.

Enterprises have also been as a rule quite passive. These companies have also thus allowed a data oligarchy class of hyperscaling social media companies/data farms to emerge. These one-way contracts we’ve grudgingly agreed to on social media platforms like LinkedIn often don’t benefit the public at large.

That’s what I mean when I say those who commit to true semantic layers and the standard semantic/knowledge graph stack including shared ontologies, decentralized storage (Social Media Linked Data or SOLID), for example) uniform resource identifies (URIs) and decentralized IDs (DIDs) are contrarians.

I put together a table to compare and contrast what conformists do and how they’re perpetuating problematic centralization, accumulation and technical debt in the process. 

Being a contrarian in this context–a follower of Head Contrarian Tim Berners-Lee– is a good thing.

Conformist versus Contrarian Data Quality for AI
Conformist	Conformist’s Rationale	Contrarian	Contrarian’s Rationale	Contrarian’s Full Argument
SQL and RDBMSes only for structured data integration	Path of least resistance; methods work with what people already know.	Add semantic layer via RDF, SPARQL, and semantic graph DBMSes for heterogeneous integration at scale	Rationalize and combat technical debt; best to eliminate duplication avoid sprawl	Software Wasteland, Data-Centric Revolution and The Future of Accounting
Vector-only RAG	100 percent statistical approach; like horseshoes and hand grenades, vectors are close enough	Add graph DBMS and federated querying for search and retrieval that supports semantic standards, common English entity linking and community-level disambiguation	Data model of the business needs tiers of logically connected, open standard abstractions for accuracy	“Common English Entity Linking”
Lakehouses or data warehouses	Best to keep analytics separate from operations	Zero-copy integration	Reduce time data professionals spend on integration	“Zero Copy Integration and Radical Simplification”
Labels	Best to keep things tactical and local	URIs or IRIs	Globally unique, unambiguous, machine-readable and reusable method	“A URI is Worth a Thousand Tags”
BI-centric semantic layer	Local semantic layer; favor the same BI stack others are using	Global, true semantic layer	Single, unified method for data, content and knowledge management	“The characteristics of a true semantic layer”
Conventional data management	Established data wrangling methods, platforms and training	Adopt FAIR principles and federated, boundary-crossing querying; blend data, content and knowledge management	Follow the approach big pharma is taking; build and adopt industry-specific standards to avoid reinventing the wheel	“Implementation and relevance of FAIR data principles in biopharmaceutical R&D”
Application- centric IT culture	The right app will solve the problem.	Data-centric IT culture	Focus on a graph-integrated knowledge+data foundation first, using disambiguated, machine-readable models and knowledge	Why Data Centricity Is Key To Digital Transformation