25 Years of Hybrid Collective Intelligence

Robin Berjon

Why the Web?

The World Wide Web (…) is an information space. Web agents [are] people or software acting on this information space.

Architecture of the World Wide Web (AWWW), https://www.w3.org/TR/webarch/

Massively collective, planet-scale endeavour
Agents so much more than browsers (Siri, OK Google…)

The Web is essentially a massive experiment in ecosystem design.

The Web is kinda crap…

…but it works!

What is the Web?

First page of 'Information Management: A Proposal'

For Everyone
http://www.foreveryone.net/

A Project Still in Infancy

Planetary Scale: billions of people still to join
Deep Historical Scope: numerous issues remain
Ubiquitous Irrigation: devices, abilities not all there yet

Goal: a universal information space and computing platform, for everyone (including AI).

Defining the Web

Not just a Cat Transference System
Do not enter this discussion…

To count as being part of the web, your app or page must:

Be linkable, and

Allow any client to access it.

Dieter Bohn

Linking

Lots of linked computers — (CC-BY) believekevin / flickr

URL

Unified Identity & Location

A resource’s name is effectively its address in the network.

Delegated authority: https://berjon.com/the-confusion/
URL structure and semantics can be locally meaningful
Opaque by design, but can be made usable to humans

Arbitrary Network

No root, arbitrary traversal, arbitrary attachment &s; inclusion
Reflects human usage
Preferential attachment is not intrinsic to the Web
Implicit structure that can be analysed

Follow Your Nose

        https://berjon.com/about/#robin
            @type http://schema.org/Person
            http://schema.org/givenName = "Robin"
            http://schema.org/familyName = "Berjon"
            http://schema.org/url = https://berjon.com/
            http://schema.org/birthDate:
                @type   http://www.w3.org/2001/XMLSchema#date
                @value  "1977-03-15"

A Web of Agents

Hieronymus Bosch: A Violent Forcing Of The Frog

A Simple Document

Robot Bartender Never Tires of Your Stories

“Come with me if you want olives!” The faux-metallic voice of the new bartender at Tin Can, a new watering hole in downtown Brooklyn might at first seem not so much hip as hip replacement. But Arnold — the star robotic bartender who runs the place — soon charms you over with his winsome mix of joviality and wry puns.

As Sarah C., a regular, downs her Souther Comfort, she barely pauses in…

The Same Document, as an Image

But: Mobile

But: Mobile Again

But: Machines

        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,
        black pixel, black pixel, black pixel, black pixel, black pixel, black pixel, black pixel,
        black pixel, black pixel, black pixel, black pixel, black pixel, black pixel, black pixel,
        white pixel, white pixel, white pixel, white pixel, white pixel, white pixel, white pixel,

But: Humans (blind)

HTML: Mobile

Robot Bartender Never Tires of Your Stories

“Come with me if you want olives!” The faux-metallic voice of the new bartender at Tin Can, a new watering hole in downtown Brooklyn might at first seem not so much hip as hip replacement. But Arnold — the star robotic bartender who runs the place — soon charms you over with his winsome mix of joviality and wry puns.

As Sarah C., a regular, downs her Souther Comfort, she barely pauses in…

HTML: Machines

        <h1>Robot Bartender Never Tires of Your Stories</h1>
        <div role="contentinfo">
          Karen Page, New York Bulletin
        </div>
        <p>
          “<q>Come with me if you want olives!</q>” The…
        </p>

HTML: Human (blind)

        Emphasis voice: Robot Bartender Never Tires of Your Stories
        [short pause]
        Article information: Karen Page, New York Bulletin
        [short pause]
        Voice 2: Come with me if you want olives!
        Voice 1: The faux-metallic…

Arbitrary Styling

Robot Bartender Never Tires of Your Stories

“Come with me if you want olives!” The faux-metallic voice of the new bartender at Tin Can, a new watering hole in downtown Brooklyn might at first seem not so much hip as hip replacement. But Arnold — the star robotic bartender who runs the place — soon charms you over with his winsome mix of joviality and wry puns.

As Sarah C., a regular, downs her Souther Comfort, she barely pauses in…

Really Arbitrary…

Robot Bartender Never Tires of Your Stories

“Come with me if you want olives!” The faux-metallic voice of the new bartender at Tin Can, a new watering hole in downtown Brooklyn might at first seem not so much hip as hip replacement. But Arnold — the star robotic bartender who runs the place — soon charms you over with his winsome mix of joviality and wry puns.

As Sarah C., a regular, downs her Souther Comfort, she barely pauses in…

Fundamental Separation

Separate: content, presentation, behaviour
The backbone is the raw semantics
Styling is an indirection to semantics, but a poor one

Target All Humans

Modality-independence helps software agents
In turn, this helps the full variety of humans using those agents
Not just accessibility: internationalisation, Unicode

Principle of Least Power

HTML is dumb: that’s good
«Powerful languages inhibit information reuse.» (TAG Finding)
Use the least powerful language that will adequately capture your semantics

Tree Decoration

        <div role="contentinfo" property="schema:author"
                typeof="schema:Person">
          <span property="schema:givenName">Karen</span>
          <span property="schema:familyName">Page,
          <span property="schema:worksFor">New York Bulletin
        </div>

Tree Decoration

HTML trees can be arbitrarily enriched
Open-ended ontologies
Single-source of truth: don’t duplicate content
Specifically designed for simultaneous human/machine consumption
What You See Is What It Means
«Web-First Data Citations»
https://research.science.ai/web-first-data-citations

Fuzzier is Less Noisy

Cater to your primary data producers, not consumers
Increasing ontological precision beyond the training of data producers (humans) just leads to noise
Massive success, very loose: schema.org
Decentralised capture (eg. schema:PublicToilet)

Collective Robustness

Elephant engraving — A Young Elephant / The Met

Postel’s Law

Be conservative in what you send, be liberal in what you accept.

—Jon Postel, Transmission Control Protocol (TCP), RFC 761.

Implementation Drift in a Postel World

Humans are not conservative (at scale).

Authors check content against browser (not tech spec)
Dominant browser has bug (or is just being liberal)
Content relies on bug
Competing browsers have to emulate bug to be usable
Go to 1

Web technology started to change on its own as a collective phenomenon.

Issues

People are lazy, computers are buggy
Policing and decentralisation do not coexist at scale
Content outlives software (by a lot)
At any serious scale, you cannot go offline for upgrades

Assume Errors

Early hypertext systems tried to prevent errors
Best effort scales better, but implies embracing errors
Errors should be exceptional (404) and undefined behaviour obliterated

Predictability

No behaviour should be undefined in a format
Every input maps to an output
Paired with testing, this leaves nowhere to drift
HTML: pretty much any junk will produce a tree

Upcoming Challenges

Implementation cost
Privacy
Centralisation

There is more…

Thank you!

https://berjon.com/ § @robinberjon § robin@berjon.com