What is possible for the future includes ample computing resources for every student ... tapping electronically many resources outside the classroom. It includes the idea of a personal factotum that could serve as a knowledgeable intermediary ... to bridge the gap between the classroom and the external world ... Virtual field trips linking libraries and museums will have their holdings available in electronic (or photonic) form ..." (Nickerson, 1988, p. 312).

We've seen where ITS research and development has been, and we've discussed a few of the systems that have been evaluated in controlled studies. We'll now examine some of the conceivable futures for these systems. Given the diversity of researchers in the area, and the great differences among learners, there will be, in reality, many different streams of research co-occurring and the most likely future is probably a composite of them all.

19.6.1 Future 1: Immersive Learning Environments Evolve from ITS

Alden (age 11) walks into his cubicle at school and excitedly puts on his VR body-suit. Today's itinerary (jointly produced by Alden and his main teacher) is teeming with new learning adventures. After taking a Dramamine, he boards a boat heading up the Nile. This trip (and his on-line tour guide) will help him learn about East Africa's geography, flora, and fauna as he cruises, observes, hears, and smells things along the world's longest river. When the trip concludes, he plans on visiting Olduvai Gorge for some archeological excavations (after all, he's already in Africa). Specifically, Alden will get a chance to help dig out some early human remains. Then, for change of pace, Alden and his VR-pal Rafael, who lives in Mexico City, will meet in a happenin' space station they programmed together. They are learning each others' language and culture--Rafael speaks English and helps Alden learn Spanish, while Alden speaks Spanish and assists Rafael with his English. Following a real lunch (not a virtual one, as all this learning makes one hungry), Alden concludes his day on an artistic note. He's creating a VR masterpiece representing his interpretation of the classical score, "The Wall" by the noted composer Roger Waters, designing virtual sculptures, their choreography, and musical arrangement.

This imagined future using Immersive Learning Environments can attain its instructional goals as follows. As a ghost presence, the tutor in these new systems can interact with a student through digital speech, through text that floats in the air, or through replays. As an embodied presence, the tutor can vary in reality from a stick figure to a realistic mannequin, with facial expressions and voice. The possibilities for realistic guidance that is as believable and as forceful as a real tutor may be quite difficult to achieve, but it can be dramatic in implications. The believability of these new systems hinges on the quality of the immersive experience they provide. The differences between an Immersive Learning Environment and its 2D simulation counterpart depends upon the results of immersion and in the different ways that students can interact with the world. Instead of moving a mouse or a joystick, learners can move their own hands to pick something up(see 8.5.2). Although they might not feel the object accurately, there are enough cues to provide the sensation of picking things up. First, they see it happening, and vision clearly dominates other senses to provide a compelling illusion. Contact and force can be provided realistically with expensive force-feedback devices, or suggestively with sounds, such as a ping that denotes collision or touching(see 13.5.4, Chapter 15, 29.5).

VR also opens the opportunity for providing handicapped or disabled people an experience of unfettered motion; or new interfaces to control the world with minimal movements. It can make invisible forces like gravity and air pressure visible and hence, more comprehensible to students. For instance, Minstrell (1988) pointed out that high school students go through a period of misconceptions during which they confuse gravity and air pressure; so that when air is pumped out of a bell jar, objects inside it are expected to become lighter or even float. VR offers an opportunity for doing a set of experiments in which the forces of gravity and air pressure could be made visible through graphic icons, such as colored arrows or textures. As the gas is removed from a bell jar, it could be visible as a colored gas flowing out. Students could actually reach into the bell jar and manipulate the objects as the gas is removed. They could even adopt the point of view, or frame of reference, of an object inside the bell jar, and experience the change in forces directly. Making these forces visible in a multitude of lifelike and believable environments may have profound effects on children's understanding of science.

It should be noted that the same problems that plague ITS are relevant to VR. That is, the emphasis needs to perhaps shift away from omnipotent VR systems, toward a collection of specific mini-systems and goals (e.g. teach the knowledge of X, the skill of Y, and provide the kinesthetic feedback for Z).

19.6.2 Future 2: Traditional ITS Disappear, Specific Cognitive Tools Dominate

Whitney (age 14) arrives in her classroom and takes a seat at her learning station, a large comfortable desk with an embedded computer. The touch-screen is divided into many different areas that have distinct functions (e.g., graphics, spreadsheet, sound analyzer, dozens of databases). From the front of the class, a visiting detective (serving as the day's teacher) accesses the international police database (IPD) and obtains details surrounding a grisly murder that happened the previous month in a small Italian city. She electronically transmits all of the information to the students, which includes electronic photographs of the physical evidence (e.g., the body and the weapon), psychological profiles of the victim and 11 suspects, recorded interviews, alibis and motives, phone logs, and so on. The students have to engage in a variety of coordinated cognitive activities to solve the murder mystery. Whitney first brings up the psychological profile of the dead man. After reading the file, she notes in her electronic scratch pad that the victim had a history of drug abuse and depression. On another part of her 25" screen, she accesses a 3D photo of the victim, zooms-in on his arms, and sees evidence of two recent intravenous injections. The pathology report from the coroner's office concluded that the victim died from a gunshot wound to his heart, but traces of a narcotic substance were also found in his body. Playing the interview tapes on her "stress analyzer," Whitney discovers that two of the suspects are clearly lying. Throughout the day, puzzle pieces slowly come together, the detective-teacher offers a few suggestions, and finally, Whitney figures out whodunit (with .93 probability of accuracy).

In this vision of the future, "omnipotent" intelligent tutoring systems have been replaced by collections of specialized educational or cognitive tools--technological devices that help people to perform cognitive tasks (i.e., help them know, think, or learn). For example, simulators, smart spreadsheets, and extensive databases are cognitive tools available within classrooms. Apprenticeship training is envisioned as the main source of imparting skill, in conjunction with the supplemental simulator and associated tools for the apprentice to employ during learning. The training situations relate to real-world events, thus placing learning within a meaningful context.

One reason that ITS may disappear in the future is that, while many researchers agree that intelligence in an ITS is directly a function of the presence of a student model, the student model may, in fact, be the wrong framework around which to build good learning machines. Derry and Lajoie (1993) presented six reasons why the student modeling paradigm is problematic: (1) In complex domains, the student model cannot specify all possible solution paths, (2) One cannot determine or induce all possible "buggy" behaviors, (3) "Canned" text is antithetical to principles of tutorial dialog, (4) Reflection and diagnosis should be performed by the student, not the tutor, (5) Implementing the student modeling approach is very difficult, technically, and (6) Model-tracing is only applicable to procedural learning, but the focus should be on critical thinking and problem solving.

A second factor that could contribute to the decline of ITS is that the term "intelligent tutoring system" is associated with philosophical issues relating to the nature of intelligence. Many people associate intelligence with awareness and, since no AI system could be said to have achieved awareness, these people would not grant that any ITS had ever been developed. Nevertheless, dozens of "intelligent" tutoring systems have been routinely reported in the literature, and even more discussed at conferences. So, the name (and hence, the whole enterprise) may be inappropriate or misleading. Simply put, ITS may promise too much, deliver too little, and constitute too restrictive a construct. Gugerty (1993) summed it up best as,

There is a sense in which the goals of traditional intelligent tutoring systems are both too ambitious and too narrow. Most traditional ITS... are designed to provide tutoring in a stand-alone setting... This ambitious goal requires that the ITS handle all aspects of the very difficult task of tutoring, including expert problem solving, student diagnosis, tailoring instruction to changing student needs, and providing an instructional environment... On the other hand, the goal of developing very intelligent stand-alone ITS is narrow in the sense that it limits our conception of how intelligence can be incorporated into computer-based training and education (p. 3).

As a parallel, consider what happened in the field of robotics. First-generation robots were constructed out of pure research curiosity. Then, after the initial flurry of excitement in the 1960s and early 1970s died down, emphasis shifted from building single-system robots, to more emphasis on building component parts. This trade-off was due to the problems associated with designing a system that has general-purpose problem-solving skills versus one with more focused expertise. The next generation of robots, arising from the work being done on the individual parts, may resolve this conflict by becoming an expert in a given domain, but also possessing a wide repertoire of general problem-solving skills. The same applies for ITS. Rather than attempting to build an omnipotent tutor, a more fruitful approach might be to create a coherent collection of computerized tools (i.e., a divide-and-conquer strategy).

19.6.3. Future 3: Distance Learning

Curtis (age 9) rolls out of bed, greets his parents (already at work in their cubicles), eats breakfast, glances at the sleet falling outside, then ambles over to his computer for his morning curriculum. Curtis "goes to school" in his home. When he logs onto the Public School System, he first checks his mail, then receives a menu of options for the morning's learning project: Would he like to learn about Tyrannosaurus Rex, the politics leading up to World War II, or what caused the California earthquake of 1994? All he has to do is tap into the appropriate database, travel to the correct geographical region and time period, and interact with these respective environments through the multi-media systems. The respective databases all include on-line hosts to narrate events and answer questions, movies to depict a range of relevant topics (from mundane to crucial), and simulators to allow Curtis to experiment within the different worlds. After choosing T. Rex as his learning project, the host narrates some basic declarative information (e.g., when they existed and for how long, size of the dinosaur, diet, mating habits, other co-existing plants and animals) then Curtis uses the simulator to manipulate geological events to see their ramifications on the dinosaur. The first thing he does is to reverse the advancing ice age (introducing a global warming trend in its place), and then sees its implications on not only the survival of the lizard king, but also on the evolution of other plants and animals on the planet. Periodically, the host asks for some predictions, Curtis responds, and receives feedback from the host. On occasion, other students in the same module communicate their findings and questions to him over the network lines.

As can be seen, this future is attractive for a lot of reasons. With distance learning, one can allow learners to stay at home or some other convenient learning location (saving time and transportation costs), and connect to a rich network of information and training software, available across an information super-highway(see Chapter 13). To achieve this future, expert systems--spanning a huge array of possible domains--are needed that present comprehensive information, as well as provide thought-provoking questions, and respond to student-directed queries. The network should also allocate nodes to which one's peers can be connected, thus providing for collaborative learning opportunities. Notice that this distance-learning future is not limited to accessing declarative knowledge from databases. Rather, software (e.g., simulators) should also be accessible to practice skill in any specific domain.

In this future, it is possible to quickly access on-line, digital-rich libraries with virtually limitless realms/databases for our personal learning pleasure. And while the educational horizon will invariably include VR technology as an important instructional medium (see Future 1), it will be just one of many media.

Finally, to attain this future and the metaphor and promise of the library as a knowledge space (i.e., the epitome of Carbonell's dream and the hypertext vision), we must first make a fundamental change how we think about education. Our narrow conception of education (e.g., "school"), only relevant for those between the ages five and eighteen, is no longer appropriate. Education should be for everyone, all ages, and available in all places.

19.6,4.Future 4: Individualized Learning is Out, Collaborative Learning is In

Sierra, Nicole, Fernando, Sasha, Kevin, and Uri comprise "Team 3." They are between the ages of 18-22 (college sophomores). In their sociology class, there are two professors and five teams, each team reflecting an optimal mixture of aptitude, gender, learning styles, personality types, and ethnic backgrounds. They are all geared-up for their on-line VR lesson on "racial prejudice." The six students are transported to Birmingham, Alabama on a hot August day in 1951. In reality, only Sasha and Kevin are African-American, but in this lesson, all six kids are transformed into "Negroes" (as they're called in 1951). The lesson requires them to take a city bus to a "Whites Only" park that has a nice public swimming pool, try to swim in the pool, then go home to their impoverished residences on the outskirts of town. Problems arise immediately in this compelling simulation when they board the bus. Automatically, they all sit down in the front seats; after all, there are only four other riders on the bus, sitting in the middle section. The white bus driver rudely informs them to "move to the back" whereupon Sierra (Team 3's outspoken leader) politely asks "why?" When she gets slapped for her impudence, Nicole starts to cry. But Sierra persists. Then the bus driver utters some very ugly sentiments about them all, based solely on their skin color. They see by his reddening face and posture that he's about to strike out again, so they collectively decide to move quickly to the back of the bus. During the ride to the park, they discuss their experiences (what they feel, what they could have done differently, what caused this state of affairs, etc.). Sasha and Kevin contribute valuable information to the discussion from personal tales related to them by their grandparents and great-grandparents. Finally they arrive at the park, and things really go downhill from there. They're not allowed to enter the park or swim in the pool, they're called "dirty" and worse, and the simulation makes them all painfully aware of racial prejudice. Afterward, Team 3 reviews and discusses all of the events, and their professors provide information, as needed, about the historical roots of racial prejudice leading up to the situation they encountered in their lesson.

The motivating force driving this future is the belief that collaborative learning is superior to individualized learning(see also Chapter 35). That is, learning may be invaluably enhanced from conversations with those who have differing opinions, backgrounds, or skills, know more about some topic, or who can ask perceptive, thought-provoking questions. Basic research is being conducted in cognitive and social psychology that seeks answers to questions pertaining to the optimal compositions of learner groups. Some of these research questions include: Is it better to mix genders, or have more homogeneous groupings? When establishing groups based on aptitude levels, is it better to match highs with highs, or a high with a low? What are the optimal coordinations of affective characteristics (e.g., passive with gregarious)? And what other cognitive/social considerations should be made (e.g., letting individuals self-select their group vs. being assigned)? According to Resnick and Johnson (1988), sociological studies show that most people prefer personal sources of information, and computers can enhance such communications.

Technology is evolving to the point where computer systems can routinely contain learning environments that support a high level of social interaction. This important technology facilitates effective learning, especially within the classroom. The atmospheres in the classrooms containing the connected computerized environment are boisterously controlled, similar to what Feurzeig (1988) found in a collaborative mathematics course that was "...more like a beehive than a math class." (p. 117). These collaborative classrooms can even support networked VR, which means that students, trainees, and experts can interact between schools and remote sites, and that trainees and instructors can share the same experience. Learners can work collaboratively on the same project. On the other hand, different students can work on the same project at the same time, without awareness of each other's presence, but with some invisible instructor lurking over their shoulders. The number of combinations are staggering, and their learning/training potential is unknown.

The other person in the networked world could also be an autonomous agent, or cyborg, part real and part synthetic. This idea raises a whole new set of possibilities for a computer coach, explanations, and guidance. "Social interface agents" (Thorisson, 1993) have progressed steadily as information about how to direct gaze, when to use paraverbals (hmmm, uh ...) and when to take turns in a dialogue, all become better understood. Improvements in modeling human actions and planning (e.g., Badler, Phillips, and Webber, 1992), including natural language interaction, will soon lead to the development of virtual agents that can coach and guide learners' actions within carefully planned learning activities. Some of these interactions are already available in a text form (Curtis and Nichols, 1993). These virtual agents focus on students' errors by offering experts' stories (Kedar, Baudin, Birnbaum, Osgood, and Bareiss, 1993). Networked digital spaces, such as digital libraries, demand new techniques for navigating through these complex spaces without getting lost. Issues of how to maintain a sense of location (Benedikt, 1991) and how to best use these environments to support memory with the method of loci (Neisser, 1987) need more research.

As shown in the above illustration, VR provides a new saliency on the notion that some things (such as race and gender) are constructed, and that we can become what we play, argue about, and build. For instance, text-based VR already invites the participation of women and girls in social interactions in ways that adventure games like dungeons and dragons did not (Turkle, 1993). Turkle points out that MUDs (i.e., multi-user dungeons) are easily used for gender swapping. When gender roles are switched, sexist expectations and overt demands that might be ignored in daily life become highly visible and reactive, and they are openly discussed. The MUD then becomes an evocative object for a richer understanding not only of sexual harassment, but of the social construction of gender.

19.6.5. Future 5: The ITS Approach Continues, Becoming Truly Intelligent

Wesley (age 10) arrives at the math lab where he sits in front of a computer that is going to help him learn to solve algebra word problems better. Today's focus is on those troublesome distance-rate-time problems. After stating his name, the computer accesses Wesley's records, flagging his salient strengths and weaknesses (i.e., not only his higher-level aptitudes, but also the low level productions that he's acquired and not yet acquired). Beginning with a review of concepts and skills that he learned the day before, the ITS generates a problem which is just a little bit out of his grasp. The ITS then works out the correct solution to the problem, along with an alternative solution that Wesley is very likely to come up with based on its student model of him. In fact, he solves the problem exactly like the tutor predicted. As part of its student model of him, the ITS "knows" to instruct Wesley with an emphasis on a graphical representation of the problem to clarify the discrepancy between the correct and incorrect solutions and facilitate the formation of a functional mental model. Thus, the tutor presents two animated trains appearing on opposite sides of the screen that converge at a point almost in the middle of the screen. They travel at different rates of speed. The problem statement stays up at the top of the screen, and the tutor points out, as it periodically pauses the simulation, what elements should be attended to and when. Wesley states that he understands the mapping between the explicated mental model, the appropriate equation, and the relevant parts of the word problem. So the ITS presents an isomorphic word problem. This time he solves it correctly, without any supplemental graphics. Wesley exercises an option to play around with some trains, missiles and boats on his own for a while to test his emerging understanding. He views his "score" of curricular elements acquired, and seems a little frustrated about his progress, but the ITS reassures him that he is proceeding at a reasonable rate. Instruction and learning continue.

For ITS to evolve to the point seen in the above scenario, more controlled research must be conducted in three areas of intelligence: the domain expert, the student model, and the tutor. First, the subject matter must be understood by the computer well enough for the embedded expert to draw inferences or solve problems in the domain. Next, the system must be able to deduce a learner's approximation of that knowledge. Finally, the tutorial strategy must be intelligent to the point where an on-line tutor can implement strategies to reduce the differences between the expert and student performance (Burns & Capps, 1988).

Solutions to problems involving difficult AI, psychology, and pedagogy will emerge from research endeavors that yield information about effective and efficient ways to (a) represent, utilize, and communicate domain knowledge, (b) represent an individual's evolving knowledge state (for both declarative knowledge and procedural skill), and (c) instruct the material most effectively for a particular learner. Some specific research questions include: How can computers better understand natural language (input as well as output)? What kinds of inference mechanisms can optimally model students' knowledge status? How can computers be programmed to understand "semilogical" reasoning (including intuitions, pet theories, prior experiences)? What are the specific characteristics of learners who perform better in certain types of learning environments and not in others? Are certain domains better suited for specific instructional methods? When should feedback be provided, what should it say, and how best should it be presented? How much learner control should be allowed?

Some additional limitations of current ITS have already been mentioned (e.g., student models cannot specify all possible solution paths in complex domains, model-tracing is only suitable for procedural learning). One possible solution would be to use a kind of model-tracing approach for instructing well-defined procedural skills, using an underlying expert and student model that are primarily rule-based. And for instructing declarative information or complex, ill-structured domains, the ITS may include a knowledge base that is a semantic net with extensive indexing (like CBR).

Whatever future ultimately evolves from ITS, the fields of AI, education, and psychology have profited enormously from the contributions made in the ITS arena. Learning theories have been tested; individual differences issues have been validated against complex, real-world learning tasks (e.g., ITS, in contrast to artificial laboratory tasks); AI programming techniques have been refined; different instructional approaches have been compared, controlled studies conducted of aptitude-treatment interactions, and so forth. So, in terms of research vehicles, ITS are greatly underestimated. But for purposes of education, their time may be limited; maybe not.