Root, Trunk, Branch, Crown

Growing VRML

Capstone Address
Delivered at VRML '95
San Diego, California
15 December 1995


I'd like to thank you for inviting me here; this represents one of those rare opportunities when I can speak to a group of my peers about matters of importance to me. Many of you have noted that I have no corporate affiliation - that I work for no one, and am answerable to no one. That's a great gift, because it means I can speak my heart without getting anyone else into trouble. I'd like to speak from my heart today, about matters of great importance to everyone within the VRML community.

I come from New England, where there's a long-standing tradition of practical anarchism, as embodied in the concept of the town meeting. These meetings, held every year, provide a forum for the citizens of the community to step forward and decide amongst themselves the important issues of taxation, community services, and quality of life. Quite often, they're loud, contentious affairs, which go on for days, until some rough consensus on the community's goals for the next year has been reached.

In many ways, VRML '95 is our town meeting, but I dare say that we're all much less likely to flame each other on the list, now that we've met face-to-face. The community which formed in the ether has been concretized in a flesh-meet, and now we're really more than just a bunch of lonely hackers, posting messages into the net - we are a united body. Or so I hope.

I think of our community as a single body, which, like any body, has many different parts, with differing goals, capabilities, and potentialities. These differences - this diversity - is a strength, because it accurately reflects something that's visibly true - VRML is not monolithic; it is many things to many people. In this, it's much like the body of a tree, which is an entire ecology, divided into four parts: the root, the trunk, the branch, and the crown.

Root: Digging in the Dirt

Two years ago, Tony Parisi and I dug a small hole in the ground, planted an acorn, covered it with fertile earth, and then did our best to ensure the sunshine and the rainfall. We knew that any successful Internet technology would have to be freely available - a standard, not a product. We knew that it had to work on a wide range of machines, not just high-end workstations, for it to be truly useful. And we knew, even two years ago, that the Web was the correct delivery vehicle for virtual worlds, the appropriate connective medium between these worlds.

We didn't call it VRML, but Labyrinth, preferring a poetic name to a functional one, and wrote Tim Berners-Lee a message, informing him of our development. Tim was delighted to hear that we'd begun work on a project he considered integral to the future of the Web - the creation of a three-dimensional space within it. He gave us encouragement, and invited us to write a paper for the First International Conference on the World Wide Web. Our paper, entitled simply, "Cyberspace", was quickly accepted, and soon I found myself on an airplane to Geneva, to attend the conference.

It's impossible to fully describe the feeling of that first Web conference. Just 350 people, almost all academic researchers, all completely enthralled with the new universe they were creating. No one ever said that the Web would change the world - that was understood. Instead, it felt like training at Starfleet Academy - with all the races, all the cultures, collaborating to create a global infrastructure for human knowledge, a restoration of the library of human knowledge lost when the Romans burned Alexandria, two thousand years ago.

At this conference Dave Raggett introduced the phrase VRML. I winced when I first heard it, for two reasons; because it was so un-poetic, and because, the moment I heard it, I knew it would stick, that we would be stuck with V-R-M-L forever. And, in a cramped meeting room at CERN, Brian Behlendorf and I crossed paths. He was impressed with the ideas, enough so to talk his employers at WIRED - who were setting up a commercial Web site of their own - into giving us enough disk space and computer time to set up a mailing list to discuss the formal specification of a Virtual Reality Markup Language.

You can trace most of VRML's roots back to that conference; I've just flown in from the fourth such conference, held in Boston this week. It is somewhat distressing that the VRML '95 organizers would schedule this event against that - a tree which forgets its own roots will not grow very tall.

However, if this date seems suitable, may I be the first to suggest that we schedule next year's event in the same week as SIGGRAPH.

There is a perception in the VRML community that we're a graphics-oriented community, when in fact that's only part of the truth. I come from a networking background, and the fact that VRML is a protocol for graphics may have something to do with the fact that networks are built from protocols. We owe a lot to the networking community; next year we'll be discussing distributed simulation - in that time, our relationship to the networking community will be seen much more clearly.

Yet one of our deepest roots reaches into the heart of the graphics community, into the "Visual Magic" division at Silicon Graphics. Gavin Bell, an engineer & systems architect working on Open Inventor, a rapid prototyping tool for graphics applications, joined our work, and became a major www-vrml contributor. Truth be told, Clay Graham was the first to recommend Open Inventor as a VRML candidate - long before he started working for Silicon Graphics. That lead us to Kevin Goldsmith, and then to Gavin, whose incredible understanding of graphics issues - both in terms of portability and expressabiility - became indispensable as we discussed the foundations of VRML.

Brian and I went to Silicon Graphics to pay a courtesy call a few weeks before SIGGRAPH in 1994; meeting with Gavin and Rikk, we entered into the first discussions concerning the use of Open Inventor as the base file format for VRML; this would take some fancy footwork on Rikk's part - but, as he pointed out, it'd probably be a year before anyone at SGI had realized what had happened.

Well, we know that wasn't correct.

At SIGGRAPH '94, I'd begged Coco Conn to make us part of SIGKIDS; Tony and I spent weeks developing a Windows version of our VRML browser, and, with our artist, Scott Young, created the first compelling VRML site - a single-room walkthrough of the U.S. Holocaust Memorial Museum, in Washington D.C. I was amazed by the reaction we received from the kids - who loved it - and the press, who went bananas. I came to understand that, unlike much of VR, we'd created compelling content that didn't focus on testosterone and thumb-twitch, and attempted to articulate the profound experience of the Warsaw Ghetto.

That week, I ran into Rikk Carey in a crowded hallway at the Orlando Convention Center - we agreed that when we got home we'd close this deal, and make an Inventor-based VRML happen. That, for you history buffs, is the only time I've been flamed on www-vrml, because I threw my support behind Gavin's proposals. My reasoning has been borne out in the intervening year - getting commercial buy-in for a technical proposal goes a long way to ensuring its success. When a legion of Open Inventor programmers realized they could create VRML content from the moment VRML was announced, our community got an unbelievable boost.

As part of "the deal", the VRML community received two key gifts from SGI; first, the guarantee that the Open Inventor file format would be made freely available to the Web community, second, that a VRML parser, known now as QvLib, would be made available, as a reference implementation, again for the use of the community. The value of these two gifts can not be underestimated; Paul Strauss and Gavin Bell created the VRML industry with QvLib, and Silicon Graphics created a standard a community could adopt.

So, five short months later, Gavin and Tony delivered a paper at the Second International Conference on the World Wide Web, which specified the features of the first revision of the Virtual Reality Modeling Language. It was - and remains - an impressive, well thought-out effort, and the plethora of VRML sites and applications testify to its functionality. It was impressive enough to convince Robert Weideman at Template Graphics Software that a VRML browser could be a useful thing, maybe even as big a success as the Netscape browser; his firm joined with SGI to create WebSpace, the canonical VRML browser. Tony founded Intervista, and began work on WorldView, and, all through the winter, Silicon Graphics' PR machinery kept grinding away, generating support for the emerging 3D Web standard.

By the way, if you ever have a PR job, just enlist the support of a $3 billion company. It makes the job a lot easier.

This spring was a busy time; most of you heard about VRML on "Day Zero", the third of April. I was asked to give a keynote address at the Third International Conference on the World Wide Web, informing developers of the details of our work as a community. It was the beginning of a publicity blitz unlike any we'd seen in the history of the Web; only one week after WebSpace was released for the Silicon Graphics platforms, it garnered a two-page spread in Newsweek. Companies jockeyed for position inside of a still-embryonic community, and created great products that didn't really interoperate. By Midsummer, this hype had become a froth, and by SIGGRAPH, the froth had become a frenzy; nothing will ever surprise me quite as much as IBM's booth at SIGGRAPH, completely given over to VRML and VRML tutorials. My IBM/VRML T-shirt is one of my prized possessions, as one of the first signs that we were winning the mind-share battle in the graphics community.

By SIGGRAPH it had become clear that the ad-hocratic environment of the community had to have a focus, a point where proposal became specification; this led directly to the formation of the VRML Architecture Group, as a collection of individuals (and companies) dedicated to the advancement of the VRML specification. The VAG doesn't act on its own; instead, it works to develop the deliberations and concerns of the community, and generates concrete proposals and specifications. When the VAG first met, we agreed to a set of criteria, developed by Rikk Carey, by which we could gauge our success. Foremost among these criteria was the wide-spread acceptance of VRML as a standard for Web-based networked simulation.

Once again, our success has exceeded our wildest expectations; last week I saw SUN and SGI holding hands - with Marc Andreesen in the middle - proclaiming VRML (and Java) as core technology for twenty-first century computing. And late last week, we saw Microsoft take the wraps off of several years of development in a time-based language that they've named ActiveVRML.

Yesterday, I delivered another Developer's Day talk at the Fourth International Conference on the World Wide Web, where I attempted to outline the future direction of VRML for the Web community. Unfortunately, this task has been complicated, in part because there's now billions of dollars riding on precisely what VRML is. This points up a problem which is as great as any of our strengths; our tree has many roots, but none of them are strong enough to bear it up through a billion-dollar battle. I'll come back to this, presently, but keep this in mind: the garden we've planted for ourselves has suddenly become the cash crop of cyberspace; we need to move carefully, judiciously, politically - we've just been thrust into the big, bad, world, a world whose goals are not necessarily our own.

Trunk: Bearing the Load

The inspiration for all of my own work came almost five years ago; I was reading Mondo 2000, issue #2. Jaron Lanier said he envisioned VR not as the television of the future, but rather, as the telephone. When I read that sentence, my entire life changed - I knew what to do with my life.

My background is in networking, specifically in great big internets. Working for Shiva Corporation - the company which pioneered dial-up networking, I was intimately familiar with the scaling and latency issues which plague large-scale internets. AppleTalk, a highly adaptive and dynamic protocol works well in small-scale internets, especially in an environment where machines are constantly entering and leaving the network. Conversely, Internet protocols are static, inflexible, but scale to millions of nodes - as we've seen. As an engineer at Shiva I spent time working within both environments, and began to comprehend that a hybrid protocol - which leveraged the strength of both - could form the foundation of cyberspace.

The problem I began with is simple and canonical - how do you know what's around you in cyberspace? Simply put, when you open your eyes, how do you know what's there? In the real world we can rely on Maxwell's Equations, but there's no analog in cyberspace - because what's around you in the virtual world may not be coincident on the network; in other words, your spatial topology is unconnected to the network topology. Yet, you, as a user of cyberspace, want to be able to "place" items within that space, and then retrieve them based upon where you've placed them within that space. This represents a total reversal of the computer-centered metaphor of the URL, which specifies an unambiguous path from client to server; in a human-centered computing paradigm - VR at its essence - the computer interprets human gestures - things like put and get - and converts them to a parsable statement.

In the case of cyberspace this means there must be some binding between a volume of occupied space; that is, a space that has been populated with VRML objects, and some set of services which are bound to these objects. That binding is the heart of cyberspace, and the three-dimensional equivalent of the Domain Name Service.

The solution to this problem is rather more complicated than DNS, for two reasons; first, space is larger than the name space of DNS, and second, these objects are highly dynamical; that is, they may be coming and going freely, especially if they're representative of avatars. The first problem prevents monolithic database representation; eventually there will be billions of individual objects in cyberspace, and that's just too much there there for every user, because either everyone is dependent upon a humungous database which is centrally located - a bad policy in networking - or the database must be very small.

Which is where things sit today. We have VRML worlds that are tiny in the absolute sense; we struggle to make meaning in a thousand or more polygons, constrained first by rendering architectures, and next by database size. The rendering architectures will grow to encompass the proximal rendering and proximal caching techniques which have been explored in distributed simulation. In these systems, the virtual egocenter - what we call the avatar - is used as input for the rendering culling algorithms; in addition, proximal caching insures that an portion of a scene graph will have been loaded before you encounter it, by anticipating the movement of the egocenter, and requesting the appropriate objects.

All of this implies that the scene graph as a metaphor is overworked; the monolithic nature of a world with a single root node doesn't necessarily scale to a planet-full of data; the Virtual San Francisco project that I and others are working on has been estimated to be at least ten billion bytes, with several million polygons, and many thousands of texture maps. LOD is a deterministic method for maintaining a database of this size and complexity, but it's limiting in that it requires the scene builder to know everything about the scene at the time of creation, and it builds a very large local database very quickly. It'll break, for the same reasons that centralized databases will break. We must leave a door open to discovery; in a highly distributed, networked virtual world - which is where VRML is going - we must have an architecture which allows people to publish into a global database without explicitly adding their content to the database. That way, I can build my own house, and keep it on my own server at home; when you come touring through my neighborhood, you access my server - though you don't know this - and find out what my house looks like.

This means that the idealized scene graph has only a root node, and other nodes are added or taken away from it based upon the location of the virtual egocenter. The browser accesses a distributed database to determine the hosts which contain the elements of the scene graph of proximal relevance, and loads them.

The protocol which undergirds this distributed database is known as cyberspace protocol (CP), and it was the focus of the first World Wide Web paper on VRML. In fact, VRML was invented as an afterthought, to demonstrate the efficacy of CP. This is not widely known, because Brian Behlendorf asked me to keep quiet about networking issues until we as a community had a solid foundation in graphics. That's now well established, and now it's time for us to move into scaleable network architectures, to support our scaleable graphics architecture, and provide a strong trunk to support a broad tree.

Branch: Shelter from the Storm

Things are very murky now, or rather, we might fear being crushed between large stones. Our work, which two years ago was described as pie-eyed lunacy, has been proclaims as the central technology of the twenty-first century. That's all well and good, but those are someone else's words, not ours, uttered for the benefit of media, analysts, and clients.

We're at a point where our safe island of www-vrml is rapidly being overrun; the number of marketing-based posts and corporate arguments about this versus that have skyrocketed in the last few weeks.

This is the VAG's fault, while some proclaim that it's been ineffective as an organization, it's actually done so much to establish a strong foundation for VRML that it's chased some birds from the bush. As these birds fly around the room and create chaos, it's good to remember that this is a function of the success of the VAG efforts, rather than their failure.

Last week two announcements shook the VRML world; first the announcement of the SGI/Netscape/Sun alliance to produce a Java/VRML hybrid for VRML 2.0, and second, the Microsoft announcement of Active VRML, positioned to be a VRML 2.0 candidate. That both of these proposals are serious indicates how seriously we're being taken; the VAG is being lobbied by all parties, and is being called upon to make strategic decisions based upon technologies that are at best, immature. We've even been told that the future of 21st century computing hangs in the balance, and, if we believed that hype, we'd probably be afraid to make any decisions at all.

But VRML did not originate with Sun or Netscape or Microsoft, or even SGI. It comes from and belongs to the communities which have given it shelter - the web community, the graphics community, and the simulation community. It belongs to us - to everyone in this room who has ever posted to www-vrml, or thrown a BOF, or written a white paper, or lobbied an organization to support VRML in some way. It is the Web's shining example of a grass-roots effort that works, and you should be quite proud of that. You've unleashed a revolution that will be as profound as the Web itself, and uncorked the genies of cyberspace.

So, you need to ask yourselves if you're willing to hand over your leadership position to someone else - some company, or some set of companies, and let them dictate to you what you can and can not do. That's really what this fight is about, going on somewhere over our heads - they're telling us that they know best, and that we should do as they say. But how can we question them? How can we be sure that our voices will always be heard, and that the concerns of the community will be placed before the concerns of commerce? VRML wouldn't be where it is today without both commercial support and a legion of enthusiasts - like the Web, it's been built on the efforts of a legion of pioneers. Is it really time to cut them out?

How do we protect the community? The branches of the tree offer protection to many species who live in its boughs; most of us have small companies, and we need the protection that these branches can offer from the billion-dollar storms swirling outside. If we want an ecology - an environment which is diverse and subtle and comprehensive - we'll have to take steps now, to ensure it.

For all of these reasons I'm announcing an initiative to form a VRML Consortium.

Over the last months since SIGGRAPH - when we tried this and failed to win significant support for it - I've given serious thought to the appropriate role of a consortium, and I've talked to other consortium heads - including Tim Berners-Lee, so that I might understand how they see their relationship to their respective communities. Over and again its been made clear to me that the consortium acts as a factor which moderates the influences of corporate behemoths, and creates a climate of stability for an entire field. The X Consortium

is a great example of what happens when this works well.

The function of a VRML consortium can be broken into three broad categories; first, research and development, second specifications and standards practices, and finally, conformance and compliance.

Research and Development

A primary focus of a VRML consortium must be to maintain and add to a strong base of technology which form the core of the community's knowledge-base. In the same way that first CERN and now W3C maintains libwww for a wide range of hardware platforms, the VRML consortium would maintain QVLib - for real, instead of the haphazard support we get now - along with the reference implementations of VRML browsers, and tools that form an invaluable resource for developers. And, just as W3C thinks about the "hard problems" in HTML and on the Web, the problems that don't make anyone any money, but need to be dealt with nonetheless, the VRML Consortium would serve as the fertile ground for innovations whose main benefit is not economic.

We need to have a place where the brightest among us can come and teach the rest of us; and a place where those same can have room to play, and give expression to their wildest fantasies. VRML is a fantasy come true - Tony and I worked alone for six months before our work garnered significant attention outside of a tiny portion of the Web community; who knows what we can engender if we create a place where great things can grow?

Specifications and Standards

The core of the VRML Consortium is the ability to create and maintain a cogent body of standards; that's one thing we don't have right now. All VRML design work is entirely unofficial; the VAG is a self-appointed body, doing its best to serve the needs of a community, but without any ability to enforce its decisions, or brand them with any particular authenticity. That's where the VAG must become one part of the VRML Consortium, because it forms the core of the Specification and Standards arm of that body.

The VAG needs to explode into at least four directions, each of which will have a particular focus. A graphics group, focusing on the core issues in visual scene description; a networking group, focusing on distributed simulation and its relation to the Web; a browser group, focusing on languages, interfaces, and interactivity; and a multimedia group, focusing on sound, video and synchronization in the virtual environment.

In all of this work, we'll have help - the VAG will grow to fifty or more members of the next year, and if that slows our process down, it's still better than the alternative - a process dictated by the largest players. And we can expect assistance from other, affiliated groups, like W3C and the IETF - for networking expertise, and the MPEG 4 standards group, who will be integrated VRML into their specification, even as we integrate their synchronization techniques into ours.

Conformance and Compliance

The only power that a consortium really has is the authority of authenticity. The VRML Consortium will have the legal power of the phrase "VRML", and can choose to label applications as VRML-compliant, or deny applications this label. That's an important power, which needs to be wielded wisely, to keep the market from becoming confused by near-VRML work-alikes, and to ensure the integrity of the specification process. Already we're encountering the frustrating problem of VRML tools that should interoperate but don't; this will be compounded a hundred-fold, over the next year, unless some body can certify that products do indeed conform to the VRML specification.

The certification process has two components; the first is a resource designed for the VRML developer, to explore the outer boundary of VRML compliance. This is analogous to the POSIX or C++ test suites, and just as necessary. The VAG has found it necessary to refine the VRML 1.0 specification several times, because ambiguities in the specification can create huge differences in implementation. Conformance tests save time and money, and preserve the reputation of VRML as a solid specification.

Finally, there's the act of certifying a VRML-compliant product. That's no small task; VRML is subtle enough to create a broad spectrum of implementations. It'll take a real lab, well-equipped and well-staffed, to ensure that products which bear the VRML name actually do conform to the VRML specification. In itself that's a great help to the developer, who will find problems during certification that may have been missed during the development process, saving money and saving face.

Crown: Shining Glories

The VRML Consortium would be the crown upon our efforts as a community; to emerge, from a mailing list, into a force for coherence and sensibility in cyberspace is no mean feat - but it's one that we should strive for, and once achieved, one we could be proud of.

It won't be cheap - nothing is - and one of the reasons that the community is fraying a bit at the edges is because everyone's doing this when and where they can. But now, with our enormous success, we all find ourselves busier than ever before; how can we make sure that VRML grows into any future? It'll take some cash.

Nothing is set in stone - these are proposals, not proclamations, and they're as open for discussion and debate as any specification or question that has ever concerned the community. These questions won't go away, and I'll be working with this community to put the consortium together, so it'd be wise to consider what you'd like to get out of a consortium, and, what you feel you can put into it. We need this common ground - without it, all of our work might soon mean nothing at all, as our spec is ripped in two by giants who fight over the meaning on a scrap of paper.

The power - and the meaning - is within us, as a community. We are the crown.

I want to propose a goal and a deadline - in five months, the next World Wide Web Conference takes place - this time it will be Paris in May, not Boston in December. I've already been asked by the program committee to develop a strong VRML focus at the Conference - there will be VRML tutorials, a VRML BOF, and, with your help, lots of papers and panels on VRML. We must work together to make that happen, and, beyond all that, we must have a VRML consortium set up and in place. During his opening address at the First International Conference on the World Wide Web, Tim Berners-Lee issued the first call for the formation of the World Wide Web Consortium; let's use this, our first meeting, to follow in his footsteps, and leave San Diego intent on building an organization to serve the needs of our community.

The VRML community is not a graphics community, it's not a networking community, it's not even a distributed simulation community - it's a cyberspace community, with prerogatives driven by William Gibson as much as Jim Kadjia, and Neal Stephenson as much Vinton Cerf. We're unique, unusual, and possessed with that rarest of qualities - a vision. That is the greatest of the jewels in our crown; every time we share that vision we put our jewels on display for the whole world to see. These gems are hard-won; we've fought first against indifference, then arrogance, and now the slings and arrows of outrageous fortune, to arrive where we are, firmly in control of our own destinies, beholden to none save ourselves as a community. It's time to spread out and share our wealth; spread our vision, so that people can understand that cyberspace isn't just good science fiction - it's good science - and that the future isn't just a simulation of fun, it's something as useful as the Yellow Pages or Wall Street's ticker.

My own work in VRML will press forward in two directions. First and foremost, I am continuing to focus on the creation of large-scale Earth-based modeling, such as those demonstrated in the T_Vision system; using the Web, I aim to create an infrastructure for planetary knowledge, as resource created and maintained by our community, for the benefit of humanity. Cyberspace is the premier tool for planetary management; in the 21st century we'll need these tools to understand the complex relationships between ourselves, our cultures and our biosphere. Cyberspace yields the ultimate "big picture", and using the Earth as the ultimate spreadsheet might teach us a lot about how to be kinder to it. The essence of cybernetics is the closed loop of monitor and modify - we will use cyberspace to bring us into a closer relationship with ourselves, our resources, and our planet.

I'm also continuing to explore new ways to sensualize the Web, using VRML. The WorldSong Project, a joint effort between myself, Dr. William Martens, and Paul Godwin, seeks to create a global soundbase, which maps sampled and real-time sound to specific locations on the planet. We're beginning to prototype this system in the new crop of interactive VRML browsers; within a few months, you'll be able to cruise through a model of the Earth, all the while, hearing the sound of communities that have added their own voice to the global world-song. We're demonstrating the power of interfaces which employ senses other than the eye, and are uniquely capable of reaching the heart.


The last two years have been an incredible adventure. The thrill of seeing cyberspace grow and mature surpasses any of my imaginings. It is a dream come true. All of this was built upon other, earlier dreams. Once I thought I could rule cyberspace, that I could own it, and, in the early years of this decade, formed a company to do just that. I named it Ono-Sendai, after Gibson's mythical manufacturer of cyberspace decks, and set out to become the mogul of cyberspace.

After three frustrating years of trial, anger, and failure, I threw in the towel, and left Ono-Sendai to pursue my own sanity. It was only then that I encountered the Web, and began creating the foundations of what we today call VRML. If there's one thing I've learned from that horrific experience, it's that no one can own cyberspace, any more than anyone can own the imagination. Don't even try. The dreams of megalomania are the shortcut to destruction; cyberspace is wild beast - none of us can hope to tame her, and most of us get only a brief ride.

But, put your dreams of hegemony aside, and the heavens will open up to you.

This is the closing talk, and these are the last words of that talk. In the Symposium program it's called the "Capstone", and it reminds me of a line from the Bible that I'm rather fond of: "The stone that the builders rejected has become the capstone". This project - which is anarchistic in form and utterly human in its intent - was dismissed and rejected, only to become the capstone of communications in the 21st century. I was - and still am - seeking a way to communicate the full depth of the human heart. That it leads to all of this is blessing upon blessing, and a clear indication that cyberspace is informed by faith, by hope, and by love.

Thank you.

11 - 14 December 1995
Boston to San Diego