Connective, Collective, Corrective:
Lessons Learned from VRML

Mark Pesce
VRML Architecture Group


Je m'appele Mark Pesce. Je parle Francais tre mal, so I hope you can forgive me as I deliver this talk in English.

I'd like to thank Bob Hopgood and the entire WWW5 organizing committee, who have invited me to deliver this address thinking that perhaps I'd be speaking about VRML. I will, but from a philosophical perspective, for this talk is not about the technical nature of the Web, but the nature of the social forms which emerge from it. This has been a focus of my research for almost five years - long before I ever knew the Web existed, and before Tony Parisi and I began the research that led to the development of VRML.

VRML has confused and confounded many people; not the technology, which is well-grounded in the essence of computer graphics and the Web, but the anarchic process by which the VRML specification is created and maintained. Just last week, at a “Birds of a Feather” meeting at Internet World in San Jose, one individual - from the Defense community - introduced himself to the room and asked, “What is your process?” In return, I could only reply, “You’re standing in it.” This might sound flippant, but it was the most honest answer I could give, for the visible processes of VRML - which has no head or center - are the only processes.

It’s necessary for the Web to grow well beyond even its current boundaries; I insist that it is the initial instance of a new form of communication, probably as important as the birth of human language a million years ago. That may seem hyperbolic, but consider this: what other technological artifact has seduced so many to do so much for so little visible return? The rewards of the Web are not primarily financial, nor do I think they ever will be - although this does not rule out commerce. Rather, the rewards are more esoteric, centered around the longing present in each of us to connect and communicate. The inventory of the effects of that communication has consistently broadened; from the CERN line-mode browser to Mosaic to Netscape with plug-ins for everything from sound to VRML.

To ensure that the Web will continue to grow we must create an ecology where new ideas can flourish - no matter how unrealistic they may seem. The concept of virtual reality on the Web seemed far-flung to most just 24 months ago, but today Netscape and Microsoft are both delivering it as core product offerings. VRML is more than just “an idea whose time has come” - it represents the model for software development in the 21st century. I’ve conducted a detailed analysis of our successes, and from this derived an ethic and logic which appears to be true for the entire Web - and perhaps for all of human communication. Some of these points may appear to be truisms of human behavior, but put together they represent a new form, greater than the sum of any of its parts.

I will now outline a system which can be replicated across the Web to deliver a constellation of new Web technologies in a thoroughly decentralized, “out of control” mechanism. This system has three stages; a period of connection, the emergence of collective qualities, and from this the emergence of corrective behaviors. This is the lesson learned of VRML. As I talk through this I will cast many deeply held opinions about business onto the dust-heap of history; from the moldering ruins of these overworked models we’ll uncover the ferment of a new approach, which accepts the existence of l’intelligence collectif - collective intelligence - as the only concrete artifact of the Web.

But before I discuss the three elements which form the chain of this process, let me say that if you came to hear a talk about VRML, you won’t be disappointed. I’ll talk about VRML 2.0 at the close of this address; but before that, let’s take a critical look at the ways we work.


The Chilean biologist Felix Manturana and ethologist Manuel Varela, in their epochal work “The Tree of Knowledge: the Biological Roots of Human Understanding” examine the way that biological systems reorganize themselves in response to the information flows in their environment. They call this emergent reorganization structural coupling, and identify it as the key quality of a living system. I bring this up because in the three-fold process which brought us VRML - connective, collective, and corrective - we’ll see how the humans involved in the effort have become structurally coupled to the means and goals of the VRML process to produce something greater than themselves.

VRML traces its roots to a “Birds of a Feather” meeting thrown by Tim Berners-Lee and Dave Raggett at the first Web conference in Geneva. This BOF - with a stated goal of discussing virtual reality on the Web - attracted about twenty people, from CERN, from NCSA, WIRED magazine, and others. During this meeting it became clear that there was a real need for three dimensional visualization on the Web. I had submitted a paper which described the first fruits of our own work in that area, but it was clear that much more work needed to be done before it could leave the laboratory and enter day-to-day use on the Web. We presumed that the talent needed to create a robust standard existed in the Internet community, so the BOF led directly to the formation of the VRML mailing list www-vrml, and the creation of the VRML home page at WIRED.

Our first decision was to make room for all who cared to participate, under the suspicion that the connectivity required to engender community went well beyond the boundaries of businesses or nations. While we expected only a few people would be interested in the development of three-dimensional protocols for the Web, we soon found several thousand individuals from all over the planet were deeply interested in the subject. In my role as list moderator, I stressed one tenet above all others - that we should be polite and politic to each other. This seems to be the key to engendering community; if it is possible to create an environment where people feel safe and respected, they will invest themselves in it, intellectually and emotionally. The point of connection between individuals must be respected above all else; any technical meditations should seek to reify this respect. In doing so, they also reinforce the community as a whole.

Although VRML is thought of as a technical movement, that’s really just the exoteric manifestation of the social infrastructure served by www-vrml. This social infrastructure is VRML.

Furthermore, the social infrastructure is always open and constantly evolving. At the San Jose Internet World, we saw OZ, an Iceland-based VRML company, announce their products. They were greeted by and welcomed into the VRML community - this from their so-called competitors, who are wise enough to know that competition is over. The implicit understanding is this: if OZ can’t find a unique way to make their mark in this field, they won’t last very long within it. So the community is really all about cooperation - even between cut-throat rivals. Companies that refuse to connect and cooperate are ignored.

This connection gives VRML its developmental speed. A few months ago, the ANSI/ISO standards body approached the VRML Architecture Group, seeking to make VRML the ANSI/ISO graphics metafile standard. At first we were rather concerned, believing that the bureaucratic nature of the ISO process would pollute our own anarchism. Imagine our surprise when the committee instead stated their intention to study our methodology, which they found both incredibly responsive and impressively credible. It may seem fast to outsiders, but in fact the VRML community is always in session, connected through our mailing list; we take weeks to debate some issues, and that debate is continual and on-going. Our credibility stems from the simple recognition that we all have something contribute to the process, and from this stone soup we brew a rich broth.


French philosopher and University of Paris professor Pierre Levy has identified a quality which he terms l’intelligence collectif - collective intelligence. It appears to be the inevitable result of intelligent systems which are structurally coupled through electric mediations - that is, as people begin to work together on a large scale and through mediations such as the Internet and the Web, they display properties which are both unpredictable and greater than the sum of their parts. In other words, connecting intelligences breeds collective intelligence, as surely as day follows night.

Collective intelligence has some characteristics which make it distinctly unpalatable to business organizations; it doesn’t care for a “not invented here” syndrome, and is generally characterized by an almost Machiavellian capability to seize every opportunity to its own advantage. That makes it impossible to predict, control, or manage, any more than one neuron could effectively control another - influence yes, but never control. Collective intelligence believes that theft is the sincerest form of flattery, and leaves no stone unturned in its search for an appropriate solution to any problem.

Which brings us to the issue of the “invention” of VRML. A lot of bandwidth has been wasted on “who invented VRML” and a number of marketing people have tried to position their firms as the “source” of all VRML wisdom. It’s commonly thought that perhaps Silicon Graphics invented VRML, or that Tony and I did. Both of these suppositions are false and examples of what Alfred North Whitehead would have termed “misplaced concreteness” - that is, the entire concept represents a fundamental misunderstanding of what’s actually going on. Although Silicon Graphics’ Open Inventor format was selected as the basis of VRML, it really could have been any format, from any company. The decision to use it was never in the hands of Silicon Graphics - that decision made through collective intelligence. In other words, no one invented VRML, or perhaps everyone did.

Yet Silicon Graphics won over the community because they chose to cooperate with it, and came bearing gifts - a license-free standard, a parser, an ready-to-go installed base of hundreds of thousands of machines. This tipped the scales in their favor, and from that we’ve learned another important axiom: the degree to which you cooperate with collective intelligence is directly proportional to your success within it. This is a demonstration of the economics of path dependency, as developed by Santa Fe Institute researcher Dr. Brian Arthur. In his model, it can be demonstrated that giving away a standard yields increasing returns, because the more people use a standard, the more likely other people are to adopt it. We’ve seen this with TCP/IP, the Web, and now VRML. Collective intelligence dictates that its infrastructure be without restriction. It remains to be seen if Java - which doesn’t follow this model completely - will maintain its lead over potential competitors which might adhere more closely to the dictates of this methodology.

The emergence of collective intelligence is an inevitability in any successful Internet-based standards- making process.


Once collective intelligence emerges in a connective environment, the collective begins to monitor and correct its own behavior - a trait analogous to the formation of nervous systems in higher animals. We saw this quite clearly in the birth of the VRML Architecture Group. Once the VRML community had achieved consensus on the VRML 1.0 standard, key members of the community recognized the need for a body to guide development of and conformance with that standard. The VAG presented itself to the VRML community as a self-selected body whose charter was to serve as a focus for the continuing efforts, and - surprisingly - we received little negative feedback for our efforts. We can attribute this to a recognition by the collective intelligence of the necessity of the situation, and the fact that the VAG uses its authority very carefully. Its job is not to govern but to coordinate, and has done so for almost a year, acting as a midwife in the development of the second generation of VRML.

The corrective is also the protective, for it attempts to maintain the integrity of the collective intelligence through a relentless culling of the weak or superfluous. The corrective has no prejudices and willingly eats its own children, or the children of others. We can see this in the VAG, which must see itself die, only to be replaced by the VRML Consortium.

But the most singular example of this is the case of a certain very large software firm, which sought to derail the VRML 2.0 specification process by taking a research project, renaming it, and positioning it as the best contender for VRML 2.0. It was through this “attack” - and you will now hear me be unusually frank - that the VRML community became acutely conscious of the integrity of its own process - in other words, the existence of an “other” engendered the consciousness of collective intelligence.

Although somewhat blind-sided, key members of the community took a wait-and-see approach, listening to the merits of the proposal. Eventually, the proposal was rejected, for reasons that have less to do with the technical base of the work - which is reasonably solid - than with a misapprehension of the essence of collective intelligence.

This unnamed large software company connected to the community only after they announced their product plans, and only with respect to them. Other community members have always sought to participate in the lively debate around all issues VRML, and through this they have become well-known. Corrective intelligence operates as a meritocracy at the individual level; if your ideas have merit you’ll be heard, otherwise you’ll likely be ignored. Corrective intelligence is conservative insofar as it seeks to explore and verify before it amends its own direction, and this unnamed large software company gave the community no time to adapt or adopt its methodology, but simply began to argue that theirs was better approach than those in development elsewhere. When community members raised issues concerning distributed behaviors or integration with other languages - which are collective goals of the community - these individuals were informed that their concerns were immaterial. This did nothing to instill a sense of common purpose - rather, it was like being dictated to by a schoolmaster intent on imparting a lesson. Collective intelligence is multilateral and continual - qualities demonstrated by Silicon Graphics, Paper Software, Black Sun or Intervista, but never evidenced by this unnamed large software company.

And so the social fabric of the community rejected that unnamed large software company; the collective intelligence seemed to be saying that any organization which sought to monopolize “the truth” would further attempt to dictate the expression of the collective mind. The corrective and conservative function of the collective intelligence closed the door - effectively shutting this unnamed large software company out of the development process - a reaction to their own attitudes. It will remain this way as long as that unnamed large software company confronts VRML as a battle to be won, and not as a community to seduce. Corrective intelligence does not willingly submit to the dominion of others.

Place that approach against the winning VRML 2.0 proposal - “Moving Worlds”. Drawn from work at Silicon Graphics, Sony, the San Diego Supercomputer Center and others, it represented a pastiche approach to design - but with none of the problems normally associated with design by committee. The encroachment of the aforesaid unnamed large software company engendered a siege mentality among the various VRML 2.0 collaborators, and forced them to work together to create a unified response. The community did not realize the degree of its own cohesion until it had been threatened; this is a characteristic quality of corrective intelligence - it is never evidenced until it acts. But because of it we have a very solid VRML 2.0 draft specification, reflecting the diverse needs of the community. Perhaps we should thank that unnamed large software company; without their interference, its doubtful that we would have developed anything so substantial.


We’re exiting the age of software and hardware engineering and entering an era of social engineering - that’s what the Web is about. Building an infrastructure for connection means that you must expect the emergence of collective intelligence, which will then execute the dictates of correction. To be successful, none of this can be under the control of any individual or organization. This message has broad application beyond VRML - indeed, it may represent a new political form - or one that is perhaps quite ancient. To ignore these principles would seem to ordain failure, while the application of them appears to ensure success - with the caveat that the form of this success can never be completely guaranteed. It’s an electronic Taoism - the reed that bends in the wind will never be snapped in the storm.

Our reed - VRML - has been tossed and turned in the last six months, and has emerged from the storm upright and strong. The VRML 2.0 specification - debated and decided through a conscious process of collective intelligence - is in its first draft release. In just a week, Silicon Graphics will be releasing VRML 2.0 browsers for Windows 95 (an example of a response to collective intelligence if I ever saw one) - and you’ll all be able to create your own fully interactive VRML worlds, with spatialized audio, support for foreign languages, and integration - at the API level - to almost any computer language. With these new features - the result of the corrective nature of the VRML process - we’ll see VRML blossom beyond anything we’d previously dreamed possible. Along with that will come a new influx of “shock troops” - people intent on creating cyberspace. It’s up to all of us to welcome them, and accept them as the latest addition to our collective intelligence. Some of them will become our bright shining stars, who will carry the torch forward into VRML 3.0.

Thank you!

Paris - 8 May 1996