Re: Agentic Linked Data -- attempt 2

Hello Everyone,

I’ve been following this thread with great interest and wanted to share
some thoughts and my work, which I believe aligns closely with the idea of
"Agentic Linked Data."

I’m working on a privacy agent that helps web users seamlessly manage their
data and consent preferences. The project was inspired by the challenges of
cookie banners, which are often frustrating, misleading, and opaque in how
they handle user data. These dark patterns can lead to serious privacy
violations.

In my work, I use what I think could be "Agentic Linked Data & AI" as a
bridge between users' innate privacy concerns and actionable data-sharing
preferences. To enable this, I’ve developed:

   1. A schema for cookie banners, providing a standardized way to describe
   their options, addressing inconsistencies across websites.
   2. A schema for "data preferences statements", which translates user
   concerns into structured, interoperable preferences.

These linked data-esque schemas provide the foundation for Agentic AI to
act on behalf of users, both in understanding their cookie and data
preferences and interacting with real-world systems to execute their
preferences. This combination reduces the burden on users while ensuring
their privacy preferences are respected. Here, the "Agentic AI" performs
two key tasks:

   1. User Interaction: Enables users to set their preferences via a
   conversation with an LLM agent, increasing their understanding of what
   exactly they are consenting to.
   2. Execution of Preferences: Enables these preferences to be seamlessly
   executed across websites automatically, by scanning cookie consent banners
   & clicking the buttons which correspond to user preferences.

I've attached a diagram of an architecture I'm using where I think Agents,
Linked Data, and AI converge to deliver value to the web user. I’d love to
explore how these ideas fit into the broader concept of "Agentic Linked
Data" and contribute to the discussion. It’s exciting to see so many
perspectives on this thread, and I look forward to collaborating further!
Thanking you,

Regards,

Ram M. Kripa


Founder, Papaya Privacy Co.

BS in Electrical Engineering and Computer Science @ UC Berkeley '22

MA Candidate in Computational Social Science @ The University of Chicago


On Mon, Dec 30, 2024 at 6:19 PM Joshua Cornejo <josh@marketdata.md> wrote:

> Hence, forums are amazing for understanding use cases, points of view and
> the right path of adoption.
>
> Comments:
> - Autonomous doesn’t mean ad-hoc, so you must provide the right
> authentication (who can access) and authorisation (what subset of data).
> Any subscription service could be seen as an agent that has to access your
> “pod” (Aad’s comments on his email yesterday about microservices are an
> example).
> - I don’t know the specific use case, but perhaps should also consider the
> requirements for encryption, and anonymisation if this research from
> Nottingham U is about statistics rather than patient care.
> - By default (as far as I understand), a Solid pod doesn’t provide
> immutability, so I would argue that medical history shouldn’t be stored
> there as the source of truth. There is a difference between “how many pills
> you have left in your prescription”(probably OK in a pod?) vs “which
> doctors have given you which prescriptions or even diet (allergies)” ” (if
> this is modified, how do you track for malpraxis!). Similar with “other”
> types of personal information that should be immutable (like the points on
> your driving licence).
>
> From the article posted first thing today:
>
> “The real power of the Semantic Web will be realized when people create
> many programs that collect Web content from diverse sources, process the
> information and exchange the results with other programs. The effectiveness
> of such software agents will increase exponentially as more
> machine-readable Web content and automated services (including other
> agents) become available. "
>
> Regards,
>
> _________________________
> *Joshua Cornejo*
> *marketdata <https://www.marketdata.md/>*
> smart authorisation management for the AI-era
>
> On 30 Dec 2024, at 12:27, Matt Taylor <matt@mattj-taylor.co.uk> wrote:
>
> I don’t see those use cases as sufficiently strong to warrant platform
> level agent access.
>
> They are application use cases and sit in the suggested service layer on
> top of a platform. They would be traditional consumers of data access based
> on set up parameters and user preferences.
> I do not believe there should be any autonomous access to personal data
> without alerts other than the 3 I suggested.
>
> I’m working with researchers at Nottingham University investigating
> benefits to prescribing of understanding sports routines and shopping
> habits etc on top of physiology data and so routine data could be
> considered important medical data as diet and activity affect how quickly
> drugs are metabolised and therefore the required dose to be prescribed.
> This is the primary interest I have regarding Solid.
>
>
> Matt Taylor
> +447768390009
>
> On 30 Dec 2024, at 12:02, Joshua Cornejo <josh@marketdata.md> wrote:
>
> 
> Extending on your use cases, agents could also be personal assistants
> (access to the individual calendar, diary, phone?) / personal trainers
> (access to the exercise routine, step counter, calorie counters, grocery
> shopping?). I could see these coming with different levels of
> “sophistication” for a membership (even a “business/personal” agent that
> sits on top of your entire personal and business diary, just like a human
> PA does today).
>
> IMHO it would be useful to separate the terminology of “personal health”
> (heart rate history, sports routines) from “medical history” (drug
> prescriptions).
>
> My first project (in the very early 90’s) was a multiyear research project
> for a health information system (HIMS), one of the first requirements from
> the doctors was that “all medical history should be immutable and
> traceable”, similar to the black box of a plane.
>
> Agents accessing fine-grained unencrypted medical history should be
> probably amongst the narrowest and in a "special category of interest”.
>
> Cheers,
> _________________________
> *Joshua Cornejo*
> *marketdata <https://www.marketdata.md/>*
> smart authorisation management for the AI-era
>
> On 30 Dec 2024, at 11:04, Matt Taylor <matt@mattj-taylor.co.uk> wrote:
>
> I’m new to the group (Hi!) and I agree that the use cases need to be crisp
> and governance strong.
> I’d like to throw my tuppence in 😊.
>
> Restricting unwanted use of personal data is a goal and not opening it up
> through an Agentic framework must be a priority.
> Keeping a robust and secure platform and then having a [managed] agentic
> service layer on top (as a separate product) is my suggestion.
> I can only think of 3 use cases where agentic access to data would be
> optimal : Personal security (agentic panic button??), Personal health
> (prescribed drug / health info where a person is incapacitated) and data
> access after bereavement.
> All 3 should (and will) be heavily regulated.
>
> I’d love to be included in the discussion / any investigative group that
> spins off from this thread.
>
> Matt
> Consultant in ethical use of technology and data.
>
>
> *From:* Sindhu Raju <sindhu2@me.com>
> *Sent:* 30 December 2024 10:27
> *To:* Joshua Cornejo <josh@marketdata.md>
> *Cc:* public-solid <public-solid@w3.org>; Melvin Carvalho <
> melvincarvalho@gmail.com>; Eric Jahn <eric@alexandriaconsulting.com>; Aad
> Versteden <aad.versteden@gmail.com>
> *Subject:* Re: Agentic Linked Data -- attempt 2
>
> Hello,
>
> I would also be interested in this discussion. I am working on a project
> in regards to personal data management.
>
> I think it would be helpful to delineate where AI agents may be helpful
> and or a hinderance via consumer use cases.
>
> Thanks,
> Sindhu Raju
>
>
>
>
>
> On Dec 30, 2024, at 12:05 AM, Joshua Cornejo <josh@marketdata.md> wrote:
>
> Interesting coincidence that someone posted this on LinkedIn and appeared
> on my timeline.
>
> I like the quote “Agents should be skeptical of assertions that they read
> on the Semantic Web until they have checked the sources of information.”.
>
> Maybe the vision is that authorisation (can the agent read the data)
> coexists with validation (is the data legitimate).
>
> <preview.png>
> Scientific American_ Feature Article_ The Semantic Web_ May 2001
> <https://www-sop.inria.fr/acacia/cours/essi2006/Scientific%20American_%20Feature%20Article_%20The%20Semantic%20Web_%20May%202001.pdf>
> PDF Document · 158 KB
> <https://www-sop.inria.fr/acacia/cours/essi2006/Scientific%20American_%20Feature%20Article_%20The%20Semantic%20Web_%20May%202001.pdf>
>
> _________________________
> *Joshua Cornejo*
> *marketdata <https://www.marketdata.md/>*
> smart authorisation management for the AI-era
>
>
> On 29 Dec 2024, at 22:04, Aad Versteden <aad.versteden@gmail.com> wrote:
>
> Hi all,
>
>
> An exciting and necessary extension indeed.
>
> I would also like to follow this too, perhaps from a short distance
> assuming time doesn't permit.  Agentic Linked Data fits closely to how
> we use Linked Data within semantic.works [1][2] so we can share what
> we've learned over time.  Microservices act as Agents manipulating data
> in semantic.works and we manage to share many microservices across
> applications.  Everything we do is based on SPARQL so some differences
> with Solid do exist.  We integrate Artificial Intelligence [3] in the
> same way as microservices which handle login, registration or sending
> out emails[4].
>
> When we squint our eyes we shouldn't notice too much of a difference in
> an Agent being a if-this-then-that construct, a Deep Neural Network, an
> OWL reasoner, an LLM, or a human.  This should be a natural and
> hopefully straightforward extension to Solid.
>
> A FOAF Agent based approach would also lend itself to support backend
> processes on Solid Pods.  We found this to be one of the bigger
> technical challenges to apply Solid Pods more in our business setting
> and it was one of the key technical challenges for Solid Shop[5].
>
> All in all, this may become a minor guideline or extension opening up
> many possibilities.
>
>
> Exciting work ahead,
>
> Aad
>
>
> [1] https://github.com/mu-semtech
> [2] https://semantic.works/
> [3] https://github.com/semantic-ai/translation-nl-en-service
> [4] https://github.com/lblod/berichtencentrum-deliver-email-service
> [5] https://github.com/redpencilio/app-solid-shop
>
> --
>
> Aad Versteden
> Technical Futurist
> CEO
>
> Proudly building the #WebWeWant at redpencil.io
>
> GitHub: @madnificent
> Mastodon: @madnificent@mastodon.social
> Mail: aad.versteden@redpencil.io
> Web: https://redpencil.io
> Tel: +32476513848
>
>
>
>