Sociability and Scalability in Online Learning Environments
1. The Problem: Sociality is Blooming
Instructional designers are frequently asked, “to which learning theory do you subscribe?” And like the famously unanswerable question “do you still beat your wife?,” no answer to the query generally seems satisfactory. Behaviorism, cognitivism, and social constructivism present significantly differing views of the educational universe. And though persuasive arguments are made that integrity of character requires an educator to adhere permanently to one view or another, I believe an individuals’ choice of a learning theoretic view of the world must always be as transient as it is pragmatic.
1.1 The Sociality Axiom
In the past few years I have come to realize that there is a reason I move fickly from learning theory to another as my instructional design framework. My seeming inability to commit is best explained by reference to Bloom’s taxonomy (Bloom and Krathwohl, 1956 ). When faced with the design challenge of helping students learn material at the bottom of Bloom’s taxonomy, behavioral methods seem most appropriate. As the subject of the design challenge climbs Bloom’s taxonomy, behavioral approaches give way to cognitive methods in my design inclinations. By the time the target material has reached the top of Bloom’s taxonomy, cognitive approaches have completely acquiesced to my social constructivist ideas about facilitating learning.
Imagine the following two non-examples of these connections between the elements of Bloom’s taxonomy and the popular learning theories. First, imagine using a social constructivist approach to helping students learn a long lists of facts. Would you personally want to learn the capitals of the 50 states through a process of discussion and negotiation in small groups of five to seven peers? Probably not; though this approach may eventually be effective, it would be neither efficient nor appealing. At the other end of the spectrum is using behavioral methods to teach students to make complex evaluative judgments. Can you imagine using only flashcard-style drill and practice methods to help students learn to evaluate the ethics of the current US war in Iraq? This approach would probably completely fail to be either effective, efficient, or appealing.
There are many implications to draw from this linking of Bloom’s taxonomy to the historical progression in learning theory, but the implication most important to this chapter I posit as an axiom (for short, the Socialization Axiom): the further up Bloom’s taxonomy a desired learning outcome is, the more important social interaction will be in promoting student achievement of the outcome. (Importance here is judged in terms of helping instruction meet Reigeluth’s (1999) desiderata for instructional designs – that they be as effective, efficient, and appealing as possible.)
1.2 The First Corollary
As instructional designers in settings from military to higher education seek to make their educational materials scale to larger numbers of learners using automated database-driven software or intelligent tutoring systems, a problem arises. The degree to which instructional episodes utilize automated agents (i.e., the degree to which human interaction is removed from these instructional episodes) is the degree to which these episodes are restricted to effectively, efficiently, and appealingly facilitating learning at the bottom of Bloom’s taxonomy. In other words, the extent to which one automates their instruction is directly proportional to the extent to which their instruction is confined to the bottom rungs of Bloom.
“But wait!” a contending voice clamors. “Intelligent tutoring systems can teach incredibly complex material very effectively!” Yes they can, particularly material that is meaningfully computable (mathematics, physics, chemistry, bioinformatics, music theory, etc.). But complex learning outcomes are not necessarily the same as learning outcomes at the top of Bloom’s taxonomy. As we see in self-organization, extremely rich complexity can emerge from the application of a small number of very simple rules. NetLogo, a popular educational simulation environment, contains a sample model which simulates complex flocking behavior in birds. The model’s documentation explains:
The flocks that appear in this model are not created or led in any way by special leader birds. Rather, each bird is following exactly the same set of rules, from which apparent flocks emerge. The birds follow three rules: “alignment”, “separation”, and “cohesion”. “Alignment” means that a bird tends to turn so that it is moving in the same direction that nearby birds are moving. “Separation” means that a bird will turn to avoid another bird which gets too close. “Cohesion” means that a bird will move towards other nearby birds (unless another bird is too close). When two birds are too close, the “separation” rule overrides the other two, which are deactivated until the minimum separation is achieved (NetLogo, 2004).
Through the simultaneous application of these three simple rules, great complexity and even seeming intelligence emerge. Computing systems are very capable of generating and interpreting this sort of complexity, and therefore they are capable of modeling, teaching, and assessing similarly complex learning outcomes. But as the NetLogo example shows, extremely rich complexity requires moving no further up Bloom’s taxonomy than application. Therefore, I posit the First Corollary to the Socialization Axiom – automated systems are only capable of effectively, efficiently, and appealingly facilitating the achievement of learning outcomes in proportion to the learning outcomes’ proximity to the bottom of Bloom’s taxonomy.
1.3 Teacher Bandwidth
For years I have described the bandwidth problem in distance education as having nothing to do with bits, bytes, or fat data pipes. Rather than worrying about how many full-screen videos we can simultaneously stream to students’ desktops, we should concern ourselves with a pedagogical aspect of the system which is far more constraining – the number of students we are capable of serving with our distance education offerings. Preserving classroom-like teacher to student ratios in online courses prevents organizations – be they universities with extension or outreach commitments, corporations with significant training needs, or militaries with demands for high levels of readiness – from “teaching” as many students as need to learn. In order to reach all the learners necessary, many organizations have selected automating instruction and feedback as the prime candidate for solving the teacher bandwidth problem.
The automated solution which relies on learning objects, standards, or other enabling technologies displays the strengths implied by the First Corollary – to the extent that the learning outcomes to be taught by the system are classified close to the bottom of Bloom’s taxonomy, the automated system is theoretically capable of facilitating this learning with effectiveness, efficiency, and appeal. However, to the extent that the learning outcomes to be taught by the system are classified close to the top of Bloom’s taxonomy, the automated system suffers the weaknesses described in the First Corollary. The question becomes, “how can we scale educational opportunities for large numbers of learners while keeping the learning environment highly social, so as to enable the achievement of learning goals near the top of Bloom’s taxonomy?”
2. Scaling Socially
The dichotomy between meaningful teacher-to-student interaction and the complete automation of teaching and feedback is false in a number of respects. Foremost among the faults of this thinking is the obvious overlooking of students as a valuable source of socialization. While some might argue that interaction with peers is vastly inferior to teacher-to-student interactions in facilitating learning near the top of Bloom, research indicates otherwise. Lave and Wenger (1991) relate that in traditional apprenticeship learning situations, apprentices spend the majority of their time working with other apprentices, as opposed to the idealized one-on-one relationship between the apprentice and master. In distinguishing between “authoritative discourse” and “internally persuasive discourse,” Wertsch (1998) argues convincingly that students learn more effectively from peers whose statements and feedback are expected to be judged on their own merits than they do from professors who expect their statements to be blindly accepted due to their positions of power in the classroom.
2.1 Environmental Affordances and Group Size
One could easily imagine that if instruction were designed in such a manner as to utilize students as an educational resource, traditional problems with oversized classes might disappear. However, physical constraints frequently prevent teachers from taking effective advantage of the significant amount of expertise distributed throughout their classrooms. Whether a student sits in a class of 30 or a lecture hall seating 300, when the teacher asks students to take 10 minutes to discuss an important topic, any given student can only feasibly converse with the students in their immediately surroundings. This “affordance” of the face to face environment works against the discovery of teaching methods that leverage large group or class sizes.
Online settings, however, provide instructors with a very different set of environmental affordances. In an asynchronous environment individuals have the capacity to interact with any person in the group, opening access to the expertise of the entire collection of students as opposed to the expertise of those seated around an individual student.
Another hypothesis can be stated here: just as there exists a class of instructional approaches which lose effectiveness as the number of students involved grows very large (think of just about any traditional teaching strategy here), there also exists a class of instructional approaches which lose effectiveness as the number of students involved grows very small. Due to the affordances of physical spaces we have never been able to witness or study this second class of methods. However, online environments like massively multi-player games demonstrate that significant learning can take place in very large groups when their communication is mediated by networked, as opposed to physical, space.
“But wait!,” one might object. “There certainly would be access to more expertise, but there would also be hundreds of thousands of voices all sounding simultaneously. How would an individual learner cope with what would have to be a very low signal-to-noise ratio associated with such a group?” Several years ago a website called Slashdot faced that very question.
2.2 Of Slashdot and k5
Slashdot (http://slashdot.org/) is a website which carries items of interest to “geeks.” The site’s tagline reads, “News for Nerds. Stuff that Matters.” Launched in 1997, today Slashdot boasts over 350,000 registered users posting over 6,500 messages per day in the site’s threaded discussion areas. As Slashdot grew to very large number of users, in the Fall of 1999 the signal-to-noise ratio became unacceptably low for site users and administrators. With the number of posts pouring into the site, it was impossible for the site’s administrators to monitor each post.
Slashdot administrators struck upon the idea of distributing the workload for monitoring comments over the entire community. Based on criteria defined by them, “moderators” were selected and given access to a comment rating system, by which they could reward excellent comments with points and punish ill-meaning comments by assigning negative points. Add to this distributed rating system a real-time filter by which users could determine the level of comments they wanted to see (e.g., only comments rated 3 or higher on a scale of –1 to 5) and the signal to noise problem on Slashdot largely disappeared. The secret of the solution was to turn the burden into a boon – more user posting more comments? Great, that means there are more moderators, too!
Another website called kuro5hin (http://kuro5hin.org/) or k5 has taken the distribution of work over the group one step further. In addition to enabling users to rate comments posted on the site by other users, k5 users also completely control all the content which appears on the site through a voting system. Any time new content is recommended for the site (content about which threaded discussion will shortly occur), the entire community votes on whether to accept the content or not. Votes of –1, 0, or +1 are tallied in real time and acceptance or rejection is based on the total score crossing a threshold calculated according to the total number of registered users. For example, for a given number of registered users, if a submission’s total score drops below –20 it is rejected. On the other hand, if its total score surpasses +75 it is accepted.
As described in the literature, self-organization sounds very similar to the manner in which these large social websites function. Whitaker (1995) outlined many of the facets of self-organization:
- self-creation – the notion that a given system’s origin is somehow determined by its character or the specific circumstances in which it occurs.
- self-configuration – the notion that a given system actively determines the arrangement of its constituent parts.
- self-regulation – the notion that a given system actively controls the course of its internal transformations, typically with respect to one or more parameters.
- self-steering – the notion that a given system actively controls its course of activity within some external environment or a general set of possible states.
- self-maintenance – the notion that a given system actively preserves itself, its form, and / or its functional status over time.
- self-(re-)production – the notion that a given system generates itself anew or produces other systems identical to itself.
- self-reference – the notion that the significance of a given system’s character or behavior is meaningful only with respect to itself.
Very large websites like Slashdot and k5 exhibit many of these characteristics, as do many online massively multiplayer game environments. These similarities led Wiley and Edwards (2002) to describe very large groups of individuals who gather in online settings to provide peer support for problem-solving and other learning goals as “online self-organizing social systems” or OSOSS.
Self-organization as an explanatory framework for social phenomena is most often used in the context of social insects such as ants or bees. How is it that a hole full of ants is capable of carrying out tasks necessary to the colony’s survival without direction from a central coordinating authority? Who guarantees that the jobs get done? As in Slashdot and k5, the answer is no one and everyone. The key which enables the ant collective to self-organize is the massive number of interactions that occur between individual ants. Some of these interactions are direct, others are not:
Self-Organization in social insects often requires interactions among insects: such interactions can be direct or indirect. Direct interactions are the “obvious” interactions: antennation, trophallaxis (food or liquid exchange), mandibular contact, visual contact, chemical contact (the odor of nearby nestmates), etc. Indirect interactions are more subtle: two individuals interact indirectly when one of then modifies the environment and the other responds to the new environment at a later time. Such an interaction is an example of stigmergy (Bonabeau, Dorigo, and Theraulaz, 1999; p. 14).
Slashdot and k5 allow for a massive number of both of these types of interaction between its users. Direct communication takes place as individuals post messages and replies to other messages. Indirect communication occurs through the rating of comments, as one user modifies a comment’s score and another users responds by ignoring or reading the comment based on that modification.
The number of interactions is important as well. Just as 20 ants wandering in a 100 square foot area may never interact with each other, the success of an OSOSS is heavily dependent on a critical mass of participants. There must be sufficient direct and indirect interaction between system users for self-organization to occur. This recalls the hypothesis posited earlier in this section – that there are strategies for working with large groups of students that were completely indiscoverable until technology mediated a minimum number of interactions among a minimum number of people.
But why is self-organization so important? Why should it be a desideratum of large online groups? When online groups are small, they can be centrally controlled, the way a moderator or instructor directs a 25 student chat in an online course. As the group grows in size, it becomes impossible for a small number of people to focus its activities in a single direction (it also becomes prohibitively expensive, as per the teacher bandwidth discussion above). For the group to remain organized, cohesive, regulated, and on a steady course, an organizing principle must obtain in the group. In other words, for learning environments to scale to numbers larger than faculty can control, and still remain necessarily social, we must rely on principles of self-organization to emerge within the group.
3. Open Learning Support
In partnership with MIT’s OpenCourseWare project, members of the Open Sustainable Learning Opportunity (OSLO) Group at Utah State University are now piloting social software intended to facilitate self-organization among large groups of learners. MIT OpenCourseWare is a project in which MIT is making available over the Internet the materials supporting nearly all 2000 on-campus courses, for free. The OSLO Group’s Open Learning Support (OLS, http://ols.usu.edu/) system was integrated into seven MIT OpenCourseWare courses in April of 2004 and made available to the public.
Self-organization is, of a necessity, directed by its agent participants who make individual decisions based solely on the information available to them locally. While the characteristics of the supersystem will play a role in the direction in which an OSOSS evolves, to assume that a group can be coerced into self-organizing in a specific manner would be oxymoronic at least (and perhaps simply moronic). In designing OLS we have taken an “evolution-friendly” approach to software design. Linus Torvalds (2001) said it best in an email to the kernel development listserv:
And don’t EVER make the mistake that you can design something better than what you get from ruthless massively parallel trial-and-error with a feedback cycle. That’s giving your intelligence much too much credit.
Rather than assume that we could divine “in the beginning” all the features which the community of OLS users would need, we have chosen to implement very few features. In fact, we believe we have implemented the minimum feature set necessary to facilitate the emergence of an OSOSS around a collection of reusable digital educational materials.
- Login – creates stable identities which can accrue histories of activity
(Reifies the system agents which will interact with one another)
- Comment / reply – allows users to post questions, answers, and other messages (Enables direct interaction between system users)
- Kudos – single point award system by which one may thank another user for a useful post or answer to a question (Enables indirect interaction between system users)
By integrating individual OLS forums with individual collections of course material from MIT OpenCourseWare, we intend to encourage discussions to focus on specific academic topics like linear algebra and applied microeconomics. However, because our primary design criterion has been to facilitate self-organization among the very large group of MIT OpenCourseWare users, there is no certain way to predict where the groups will take the OLS forums.
3.2 How OLS Evolves
During the pilot phase of OLS, the OSLO Group will carry out two distinct sets of activity. First, the Research Group will carry out Computer Mediated Discourse Analyses (Herring, ref) of conversations in the OLS system to anticipate community needs and how they can be built into the OLS software. Second, the Engineering Group will build additional functionality so that it can be made available at critical junctions in the groups’ evolution.
Stable user identities and the OLS Kudos system enable many “advanced” features which we anticipate the community eventually desiring. For example, plain keyword searching can be enhanced by ranking results according to the number of Kudos returned comments have received. Tracking and displaying the number of Kudos awarded to a users’ messages facilitates a reputation management system (similar to Ebay’s color-coded stars) by which system users can gain a positive reputation for making significant contributions to the community. Also, individual users’ distribution of Kudos point awards can be mined to enable a collaborative filtering system (similar to Amazon’s book recommending feature) by which a user who awarded Kudos to 23 comments can be notified of 7 additional comments she might find useful based on similarities between Kudos she has awarded and those awarded by other OLS users.
3.3 The Future of OLS
Although OLS is currently integrated only with MIT OpenCourseWare materials, the OSLO Group is already working with other organizations to integrate OLS into their content collections. OLS is open source software. We hope that by enabling self-organization among large groups and organically growing site features as needed by the community we are able to demonstrate successful teaching and learning interactions within a very large, academically focused, self-organized group.
The achievement of higher-order learning outcomes, such as those near the top of Bloom’s taxonomy, requires social interaction to be an integral part of the learning experience. As institutions seek to scale their educational offerings over great distances to large numbers of people, social interaction has traditionally been seen as too expensive to include. While the automation or dehumanization of online courses does improve their scalability, it also hampers their ability to facilitate these inherently social higher-order learning outcomes. Online self-organizing social systems (OSOSS) are one method of scaling educational offerings to large numbers of people while keeping the educational experience very social. OSOSS may therefore be an important key to scalable online programs which are capable of facilitating the mastery of learning outcomes across the entire range of Bloom’s taxonomy.Originally posted by david at April 19, 2004 09:07 AM
Last modified 2004-09-23 12:33 AM