FAS Virtual Worlds Whitepaper
From FAS Virtual Worlds Almanac
By Henry Kelly, President of the Federation of American Scientists
Virtual Worlds have the potential to revolutionize the way we learn and the way learning systems are built - and do so in a way that is immediately accessible to the growing number of people worldwide connected to the internet. These systems can make it affordable to implement the kinds of learning experiences long known to be powerful but impractical because of their cost and complexity. They make it possible for a diverse collection of developers, instructors and others to collaborate in the design, evaluation, and use of such learning systems while providing the option for learners and educational institutions to choose among competing approaches.
Making this happen, of course, depends on creating a sizable community that can collaborate in the same Virtual World - or at least in worlds that can easily communicate with each other. Creating seamless interoperability among Virtual Worlds, however, is spectacularly (some would argue insurmountably) difficult. What is clearly needed is an HTML for Virtual Worlds, but the range of connections needed to make things connect in the 4D spaces of these worlds is many orders of magnitude more difficult than the page creation standards at the core of HTML. Heroic Department of Defense attempts to create Virtual World standards are already obsolete and have never caught on in commercial products. The most likely scenario for interoperability is one in which a single platform achieves such market dominance that all other platforms must find ways to be compatible. The path to dominance, however, is unlikely to be driven by the education and training markets since these will always represent a small fraction of the total market. However these institutions have a strong interest in encouraging the emergence of platforms that are open to additions and extensions needed to support learning applications.
Instead of waiting for a functioning standard to emerge, it makes sense for the education community to find a platform that can meet the greatest number of this community's requirements, that stands a good chance for capturing commercial markets and continuously improving its capabilities, and that offers the most powerful tools for building the extensions needed to meet the unique needs of education and training. No current platform perfectly meets these requirements. The issue at hand is whether one comes close enough that it can be chosen as the basis on which to build. The risk that the platform chosen will fail in commercial markets or that it will be overtaken by other methods is very real. But with some care the objects created, and the management lessons learned from active use and collaboration on a chosen platform can be moved with relatively little pain to other systems. There would presumably be strong commercial incentives to ensure interoperability with the emergent standard.
No Virtual World platform is the obvious choice for education and training and the process of choosing one will be vexing. The most exciting times in the history of a technology are also often the most frustrating.
The process must begin, however, by establishing a set of criteria that should be used to judge the candidates. That is what this paper seeks to do. It will be accompanied by a wiki-site that will post information about issues and criteria for judging every major Virtual World platform and invite comments and additions. The site will include information about topics listed in this paper and additional technical information and links to relevant websites. Functionality
While the success of any technology-based learning system will always hinge on the quality of the approach to instruction, there's no avoiding the fact that learners will be strongly influenced by the quality of the audio/visual experience that provides the background for the instruction. Expectations for quality simulations have increased at spectacular rates in recent years with playable levels, in-game cinematics and pretender cinematic sequences approaching photorealism. Graphics and Sound
The quality of real-time graphics in computer games will always be lower than those of animations created by movies since game graphics must be created in real time while movie animations may take hours per frame to create and are built using extremely powerful computers. But the quality of graphics on dedicated game consoles are never more than a generation behind. State of the art game devices like the PlayStation? 3 can produce 360,000 flat shaded polygons per second and roughly half as many textured, shaded polygons. With the use of normal mapped textures for characters and environments, that number of polygons can be tasked to appear much larger.
Graphic quality is further degraded when the display must receive information over a network connection. Instead of being able to pull all information about a new object from a local disk, some of the information must be imported over the communication link. This can create annoying latency delays. But the state of the art of online simulations is also advancing rapidly. A key innovation has been to minimize the amount of information that must actually be pushed through the server side of the network connection. Information about objects (such as a truck) are stored on the client machine and rendered locally using the client's graphics. Changes affecting the truck (such as the fact that another user steering the truck has decided to turn it to the left) must be sent from the server. In some systems all object information is stored in the client's machine - meaning that new objects can not be created or added in real time. In others, information about new objects is passed to the clients when they are about to come into view.
In addition, a new type of client-server platform is already on the horizon - server-based rendering and processing. Once any (usually 3D) object was rendered (i.e., created) on the server, the rendered result would be pushed to users (clients) in a video-like stream. This would make enhanced graphics more easily available, even to users with lower-end machines.
The central challenge remains one of finding clever ways to balance local and central processing. An even greater challenge is that almost no two systems use the same approach, making compatibility a difficult goal.
This discussion has focused primarily on visual effects, but similar challenges apply to sound patterns. In many cases the emotional impact of a situation depends more powerfully on sounds than on what you see. In principle, simulations should match good 3D visuals with good 3D sound. You should hear a ball and bat collide, and hear the pitch of a train change as it passes you. Simulations
In addition to the graphic displays, a realistic virtual environment needs to ensure that the objects behave appropriately. This means that cars need to drive over hills not through them and that balls rolling off a tabletop drop and bounce as per their physical material. But this can pose an extremely complex processing task when many objects are involved or when the phenomena simulated are complex (e.g. fluid dynamics or softbody objects that flex and twist). One of the challenges of an instructional environment is that these simulations need to be technically correct. Many game systems greatly simplify the underlying science or intentionally exaggerate its causes and effects. A good example is frequently seen in games dealing with water - an ocean meeting a coastline is most normally created with a series of overlapping textures that look like breaking waves, rather than the more accurate simulation of fluidic particles hitting the sand and responding appropriately.
In technical terms the issue is whether the platform includes simulations as a part of the basic software and whether it supports "application programming interfaces" or APIs that allow them to connect to simulations created by others. Many platforms have simple "physics engines" that ensure that basic movements and collisions in a simulated 3-D world are correct. But they face huge challenges when many different objects must interact with each other and their surrounding environment in real time. Even the seemingly simple problem of ensuring that a large number of objects collide appropriately (or steer around each other and obstacles in the environment if they're acting with artificial intelligence) is a huge challenge. The problem becomes even more complex when the instructional environment demands precise engineering models, models of fluid flows through complex equipment or over complex landscape, chemical reactions, biological synapses etc.
Modeling the behavior of both player and non-player controlled (NPC) game characters requires another form of simulation. In many situations it is necessary for users to interact with human or animal forms that react intelligently to the user's actions. The use of artificial intelligence scripting and behaviors, motion capture and hand animated moves, proper physics and Inverse Kinematic movement, as well as "attitude" all come into play when dealing with characters. Add more characters and the processing challenge is multiplied. Clearly it will be a long time before all such phenomena can be modeled correctly in educational platforms. The early courses supported will simply be ones that don't require such sophistication.
A final issue is the extent to which the 4D Virtual Worlds can connect seamlessly with the 2D world of the existing web. It is important that people operating in the world be able to see videos, read text, see weather maps, and enjoy all existing web based tools without compromising the quality of these materials. Interface Tools
The quality of the user's experience will also be strongly affected by the ease of moving within the Virtual World and their ability to interact with it. The generation of Americans who've been weaned on computer games have no trouble navigating the landscapes and understanding how to manipulate objects. But many challenges remain. Tools to pick up and manipulate objects are very artificial. Use of gloves and exotic manipulators has largely failed to gain markets - in part because they were uncomfortable, expensive, performed poorly and were not marketed well. The new Wii interface shows, however, that innovative approaches to new interface tools can have great appeal if they're easy to use. Haptic devices may yet emerge in an improved and widely marketed version in the future.
A user's experience, at least with tools, will depend both upon how easily they can manipulate objects created by others and upon how easily they can manufacture objects of their own. Modern 3D creation tools (i.e. Autodesk Maya, 3D Studio Max and CAD tools) are extremely complex to use - although recent experiments in the US and the UK clearly demonstrate that middle-school children feel perfectly comfortable using simplified modeling tools to create very sophisticated 3D objects. One difficulty with the current generation of Virtual Worlds is that most use their own authoring tools and many store the data in nonstandard formats, making interoperability difficult and forcing users to learn new authoring systems for each Virtual World.
Similar problems are faced in writing scripts or inserting new simulations to guide objects following artificial intelligence rules or behaviors governed by a science or engineering model (e.g. structural integrity of a bridge).
Another critical interface involves communications between characters - that is between the learner and the other humans - artificial and otherwise - who occupy the virtual world. Systems designed for social networking have placed great importance on these capabilities and many offer both text messaging and sophisticated audio/voice system that makes it easy to determine who is talking.
The quality of the experience of using and building objects also depends on the ease with which they can be located. Searching for complex objects in a Virtual World can be frustrating since no Google tool has been developed. On the other hand, web portals are currently available to help address the need for "Googling" virtual objects. Sites such as SLExchange.com and OnRez? .com allow users to search and purchase virtual objects today. Electric Sheep "scrapes" the entire Second Life grid for objects marked for sale and offers results on the web (without, it should be noted, the permission of the users who created them).
The task of making virtual worlds accessible to people with disabilities is challenging and has not been adequately addressed. Techniques are available for making the world accessible to people with limited hearing (many worlds communicate entirely through text), and interface devices can improve access for people with motor disabilities. Virtual Worlds, however, depend heavily on visual clues making them all but inaccessible to people with limited sight. To our knowledge this remains an unsolved problem. Performance Issues
The practical utility of Virtual Worlds depends not only upon the quality of what they deliver, but also the reliability of delivery. It can be impractical and frustrating to the end user to use software that isn't stable and crashes frequently. Similar frustrations would occur if the software is unable to scale gracefully to accommodate large numbers of users and large numbers of objects, or major changes are constantly being made, forcing massive, time-consuming and possibly data-unfriendly updates.
The phenomenal growth in interest in Virtual Worlds has created some expected growing pains. Some platforms can and do scale more gracefully than others because of a robust design. Constant updates, although sometimes annoying, are often expected and even anticipated by regular users, lest the product grow stale. A general rule of thumb among developers is that such updates should occur at least once every six months. They can also be a measure of whether the system is actually being used, developed and tested. It's easy to have stability with a few predictable users, but highly challenging when large numbers of users are finding problems and asking for new services. Rapid learning and innovation depends on a significant user base. A key issue in upgrades, however, is whether user investments in improving one's character or environment will be undone by new revisions of the software. Clearly a balance is necessary between the instability of upgrades and a dynamic and improving game experience.
The utility of the Virtual World platforms for education also depend on whether separate instances can be created to serve an individual group of students. If a virtual Venice is built, for example, must it be open to all users simultaneously or could an instructor use a copy (an instance) of the site for a specific class? Compatibility Issues
The first dimension of this issue deals with whether the system is compatible with the wide variety of platforms now available to students and whether it will be able to operate in a complex and continuously changing communication environment. This question involves whether the software can operate on multiple platforms (Windows, Linux, Macintosh), and whether it will be able to operate on wireless devices. It also involves the practical question of whether it will operate through the security systems and firewalls now needed by most installations - including schools. This issue is tightly linked to the issue of Trust and Identity that are included in a later indicator.
A second set of questions deals with its compatibility with conventional formats. The ability to import and export 3D objects from third-party software, user scripting for AI instructions and triggers, web pages, video and other material is a key interoperability issue Security Issues
The ancient tension between liberty and order plays out as vividly in Virtual Worlds as in the real. Networked transactions have evolved over the past decade to provide high levels of trust through encrypted communication, identity management, and other tools. But many problems still exist. Virtual worlds face an identical set of problems, but have not managed to build adequate tools for identity management and trust. Closed, proprietary game systems that require monthly fees and some form of verification, via credit card or similar payments, have comparatively few problems. It is possible to imagine a walled garden for education and training applications. Second Life, for example, allows users paying a fee to create protected areas where access is granted only to avatars given permission (assuming that they have not misled the owners about their identity). But the value of open Virtual Worlds, like the value of the web itself, is diminished if users have limited options.
As the internet itself has shown, the types of abuse are often completely unexpected. In World of Warcraft, for example, a group announced that it wanted to have a virtual ceremony to honor the memory of a player controlled by a person who had died. They were attacked viciously by a rival clan during the virtual ceremony. Second Life has been the victim of a large number of abuses both because of its philosophy of openness (users can enter without registering their identity or making any payment) and because of its huge growth. And as expected, abuses have occurred. Hackers have found ways to steal objects and vandalize others. Users are frequently subjected to abusive and malicious behavior in spite of Linden Labs, policy of banishing users exhibiting such behavior. Meetings have been disrupted by hecklers and obscenities. In another incident a vender in Second Life whose virtual objects had been stolen is attempting to sue the thief. This is, of course, difficult to do in Second Life since identity is difficult to establish given the registration requirements. But the thief was forced to use eBay to convert virtual income to real income thereby providing an avenue for detection.
Clearly there is an urgent need to address these issues before education and training institutions can make widespread use of Virtual Worlds. Installed Base
In the real world, the sale of each phone increases the value of every other phone since there are theoretically more potential choices for users about whom to call. The same phenomena occur in Virtual Worlds where the richness of the experience depends on a critical mass of participants actively building objects or are available for conversation or collaboration. The choice of a platform for education and training must reflect the number, quality and diversity of objects available for use, the number of people familiar with the platform and prepared to use it for educational purposes without a long learning curve, and the number of developers capable of using the authoring tools and quickly building new communities for educational purposes. One of the frustrations of making decisions about emerging technologies is that a system may be preferred if it has a rich set of assets, growing rapidly, and constantly adding new features may be preferable to a system with superior technical performance but few users and little real world experience. Legal & Management Issues
Virtual Worlds have been designed to serve a wide variety of markets - primarily gaming and social networking. While many could clearly be used to support complex education and training experiences, few have been designed for this purpose and none currently offer the full suite of services that would be needed to support education and training services.
An operational learning system built on a shared Virtual World platform must be constructed from at least five distinct services, each of which present different challenges for financing and policy on "openness":
1. Developing, upgrading, and maintaining the core software (client and server code) 2. Providing computing and data storage and management services running the server code provided by an operator (e.g. server farms) 3. Generic management and business services such as managing accounts, micro-payments, IP management, mechanisms for paying human experts, actors, counselors, etc. 4. Objects that can be used in the virtual environment (landscapes, buildings, equipment, AI avatars, simulation code). 5. Specific implementation of learning modules created by assembling components and building games, challenges, and assignments in the virtual world.
One difficulty faced in comparing systems is that different virtual worlds provide different kinds of services. Croquet, for example, provides or will provide core software for building and sharing virtual worlds but does not provide computing support for business services or computing. Second Life, Olive and Entropia provide a broad range of services. In all cases education and training providers would need to build and operate their own learning systems (level 5) A core problem facing education and training providers is how to take advantage of the investment being made by commercial Virtual World products by building new services within them without being trapped in a proprietary system.
There is also a separate vision - that these educational systems can exist in a world entirely supported by volunteer effort in "creative commons" using "free software"•. The extent to which this is practical or financially viable remains an open question. Like most systems it's likely that a hybrid will evolve with some parts open and transparent while other services are proprietary. Standards for interoperability are essential for this to work. It is perfectly reasonable today, for example, for a provider to create a web page using proprietary software as long as there is a guarantee that the page can be read by all users and any text or graphics converted and reused within limits imposed by the creator's license.
There is even the possibility that the server and data management services could be provided at little or no cost if advertising or promotions are tolerated (something that the education and training community have never formally accepted). The extensive use of proprietary server networks that provides the backbone for Google searches and Google Earth, for example, are free because Google has found a way to earn revenue from advertising. Modzilla manages to support and maintain a "free software" browser because of revenue derived when users select Google or other search engines from the browser and by selling services (approximately $50 million in 2005). Open Source Software
The software used to build virtual worlds range from entirely closed, proprietary systems to completely open ones. All of Croquet is available under a license that allows essentially unlimited use - including incorporating the software into proprietary products. The client-side software for Second Life is available under a GNU "copyleft" license but the crucial server side software is not. Olive and Entropia are entirely proprietary. IP Rights
The rules governing ownership of user generated content varies enormously. There are two issues: (a) controls imposed by the provider of the Virtual World (e.g. Second Life), and (b) controls imposed by users on their own creations in the Virtual World. It's important to make the distinction since if users own the title to their creations and are free to convert their material for use on other platforms, the penalty paid for using a proprietary software platform are greatly reduced. A key related issue is the extent to which the systems provide protection for intellectual property. Second Life has experienced the theft of materials and made repeated upgrades to increase protection (see Identity and Trust below) Croquet has no content controls whatsoever although each user must create their own inventory of objects and find ways of sharing them. Second Life provides built-in tools for sharing objects with all users and gives users ownership of all their creations but retains the right to use the material for software testing and promotion. It appears that the user has complete right to take objects they create in second life and convert them to a form usable on other platforms. Entropia claims ownership of all user developed content.
Users who create their own content, however, are not necessarily willing to share their creations under a Creative Commons or other license. Many charge fees for using their material. An educational community would need to come to some agreement on the rights claimed. Separate protection may be sought by education and training institutions when assembling objects in the Virtual Worlds, creating challenges and assignments, creating tools for measuring individual student progress, providing advice and instruction through teachers and experts providing real-time guidance, and other services. Rights to the created content could be claimed by individuals, universities, or companies. Indeed, some may contain proprietary or even classified information and necessarily be restricted. The precedent of the "open courseware" initiative can have some bearing in this level but the complexity and cost of developing courses in Virtual Worlds is considerably greater than the task of putting readings and videos of lectures online. Identity and Trust
A successful education and training system must provide ways to ensure the identity of participants is known so that records can be maintained. Issues of identity and trust are essential to support any financial transactions and they are essential to provide credentials for students. Detailed data can be collected about each individual that can be extremely useful to instructors and to any AI system. This can range from which languages the person speaks to their level of expertise and mastery of learning objectives. But this information must be carefully protected and made available only to trusted agents (instructors and others) at the discretion of the student. Protection Against Abuse and Offensive Behavior
The quality of an experience in a virtual world can be compromised by malicious or annoying behavior by other participants. This is, of course, a price of openness, and the steps needed to get protection from hackers, spammers, and distributors of viruses are all too familiar to users of the current internet. The platforms available for Virtual Worlds provide a wide range of solutions to these problems, but few have had to contend with the level of threats that will be faced as large numbers of users start to participate. Platforms where the identity of each participant is known to the system manager because of credit card information have the most options for dealing with malignant behavior. At present - and probably in the future as well - the best protection against such behavior will be for training and education to be developed in silo environments where so-called griefing (deliberately attempting to ruin the online experience of another user) and outright theft can be greatly reduced. These two areas of concern, at least, should not have to wait until these specific issues are addressed by other means.