Hostname: page-component-745bb68f8f-lrblm Total loading time: 0 Render date: 2025-02-12T01:22:41.144Z Has data issue: false hasContentIssue false

Teacher and learner: Supervised and unsupervised learning in communities

Published online by Cambridge University Press:  08 June 2015

Michael G. Shafto
Affiliation:
Cognitive Science Associates, 2850 Easy St., Ann Arbor, MI 48104.
Colleen M. Seifert
Affiliation:
Department of Psychology, University of Michigan, Ann Arbor, MI 48109-1043. piltdown@gmail.comseifert@umich.edu

Abstract

How far can teaching methods go to enhance learning? Optimal methods of teaching have been considered in research on supervised and unsupervised learning. Locally optimal methods are usually hybrids of teaching and self-directed approaches. The costs and benefits of specific methods have been shown to depend on the structure of the learning task, the learners, the teachers, and the environment.

Type
Open Peer Commentary
Copyright
Copyright © Cambridge University Press 2015 

In a bilingual public school classroom near Silicon Valley, one can observe a variety of teaching and learning practices occurring simultaneously. The teacher has a queue of five students at her desk, and she helps them individually. Thirteen other students are working “quietly” at their desks, and two are being tutored by adult volunteers, focusing on reading in English and Spanish. Perhaps the most striking observation in this classroom is the irrepressible tendency of the students to request and receive help from each other. Another is the participation of volunteers, many of whom work at Google, NASA, and Stanford. In addition, there is a broad range of unsupervised and tacit learning, including Spanish pronunciation and grammar, beliefs about the supernatural, entry-level soccer skills, gender roles, Minecraft, and so on. This suggests that teaching and learning are far richer phenomena than depicted in the target article. Who is a teacher and who is a learner, and does teaching require intention?

We suggest that useful contributions to Kline's framework may follow from examining the extensive work in the field of cognitive science, especially in the subfields of machine learning (ML) and intelligent tutoring systems (ITS). This research has addressed teaching in terms of four major types of variables: learner, environment, teacher, and content. The results demonstrate the limits of “what works” for effective human learning and teaching (though other work addresses artificial agents and nonhuman animals as well). By avoiding discussion of research on formal instruction, the Kline framework limits the landscape of what we know about teaching.

In the field of cognitive science – especially in the subfields of ML and ITS –theories and models of teaching have been characterized in terms of supervised and unsupervised learning. In the case of unsupervised learning, the learner has some means of encoding input from the environment. In the case of supervised learning, environmental input is selected, filtered, or otherwise enhanced by a teacher (Duda et al. Reference Duda, Hart and Stork2000). The teacher may choose to apply a wide range of different strategies in order to increase the learner's rate of progress. Settles (Reference Settles2010) and Zhu (Reference Zhu2008) review work on (hybrid) semi-supervised learning, which focuses on the effectiveness of specific combinations of methods.

These notions have been made precise in recent work on machine learning. They have been subjected to extensive empirical evaluations. In general, the effectiveness of various teaching strategies – mainly hybrids of supervised and unsupervised approaches – depends strongly on the structure of the environment and of the learning task (content). As a starting point for continuing interdisciplinary dialogue among cognitive science, anthropological, and evolutionary perspectives (cf. Atran & Medin Reference Atran and Medin2008; Beller et al. Reference Beller, Bender and Medin2012; Kline's target article), we present a few informative examples from the ML and ITS literature.

Kline discusses a child cooking alongside her parent, whereas research in cognitive science examines teaching in adult learning. Hutchins' (Reference Hutchins1995a; Reference Hutchins1995b; Reference Hutchins2005) research has emphasized the universal importance of teaching by apprenticeship. His work on navigation tasks has included Micronesian small boats, piloting of large naval vessels, and crew-automation interaction in advanced aircraft. Apprentice learning appears to work best in small teams where less experienced members learn by performing less demanding aspects of real tasks while supervised by expert members (Seifert & Hutchins Reference Seifert and Hutchins1992). For example, Hutchins (Reference Hutchins1995b) suggests the importance of the “horizon of observation,” or access to visual information about other activities occurring around the learner. Participation through “overlooking” (as in “overhearing”) may play an important role in preparing for future learning.

Cooperative learning is also raised by Kline, and recent work (Resendes et al. Reference Resendes, Chen, Acosta, Scardamalia, Rummel, Kapur, Nathan and Puntambekar2013) suggests that its benefits begin accruing early in life. Resendes et al.'s study of second-grade students claims that “the continual give and take of ideas to advance community knowledge is a foundational principle upon which knowledge building communities operate” (p. 396). In the study, they designed and evaluated tools to support behaviors like sharing information and word learning. In classrooms with team-based learning among 7-year-olds, they found that individual vocabulary and community (shared) knowledge both increased. Peers are important sources of knowledge for human learners in both informal and formal educational settings (Boud et al. Reference Boud, Cohen and Sampson2014).

One-on-one tutoring is such an obviously optimal teaching strategy that a good teacher will implement it (even with a 20:1 student–teacher ratio, as seen in our opening example). Clearly, 1:1 tutoring out-performs regular classroom or textbook-based instruction. And ITS have been found to perform as well as human tutors (VanLehn Reference VanLehn2011). But does it matter which strategies a teacher uses during tutoring? Studies show very little evidence that different tutoring approaches significantly affect learning outcomes (Chi et al. Reference Chi, VanLehn, Litman and Jordan2011). However, using an ITS for college-level physics (very advanced, abstract content), Chi et al. were able to demonstrate that theoretically superior tutoring strategies actually do out-perform others when analyzing at a micro-step level (Reference Chi, VanLehn, Litman and Jordan2011). These results, along with some of the examples in the target article, suggest that humans have evolved to be good teachers, but not optimal ones. Optimization may require detailed monitoring, decision-making, and control that lies beyond the cognitive capacities of the unaided tutor.

In closing, we raise an issue that may not surface in cognitive science without the benefit of an anthropological or evolutionary perspective. Much of the discussion of teaching and learning in humans focuses on providing basic skills, broadening horizons, or talent-scouting. But much actual teaching, in both Western and non-Western cultures, aims at instilling conformity by building resistance to change and maintaining conventional customs. Teaching can be aimed not at “utility by truth” (technology, nature, geography, etc.), but rather, at “utility for solidarity,” as in folklore, kinship, history, and mythology (cf. Roberts [Reference Roberts2013], for a contemporary Western example). Content may be quite tenuously linked to reality, yet it is often taboo to question this. From a cognitive perspective, this type of teaching seems anomalous; however, from an anthropological perspective, it may seem routine, and scarcely merit comment. It is clearly a key element in considering the (culturally) adaptive value of teaching.

References

Atran, S. & Medin, D. L. (2008) The native mind and the cultural construction of nature. MIT Press.Google Scholar
Beller, S., Bender, A. & Medin, D. L. (2012) tShould anthropology be part of cognitive science? Topics in Cognitive Science 4(3):342–53.Google Scholar
Boud, D., Cohen, R. & Sampson, J., eds. (2014) Peer learning in higher education: Learning from and with each other. Routledge.Google Scholar
Chi, M., VanLehn, K., Litman, D. & Jordan, P. (2011) An evaluation of pedagogical tutorial tactics for a natural language tutoring system: A reinforcement learning approach. International Journal of Artificial Intelligence in Education 21(1):83113.Google Scholar
Duda, R. O., Hart, P. E. & Stork, D. G. (2000) Pattern classification, 2nd edition. Wiley.Google Scholar
Hutchins, E. L. (1995a) How a cockpit remembers its speeds. Cognitive Science 19:265–88.Google Scholar
Hutchins, E. L. (1995b) Cognition in the wild. MIT Press.Google Scholar
Hutchins, E. L. (2005) Material anchors for cognitive blends. Journal of Pragmatics 37:1555–77.Google Scholar
Resendes, M., Chen, B., Acosta, A. & Scardamalia, M. (2013) The effect of formative feedback on vocabulary use and distribution of vocabulary knowledge in a grade two knowledge building class. In: To see the world and a grain of sand: Learning across levels of space, time, and scale: CSCL 2013 Conference Proceedings Volume 1 – Full Papers & Symposia , ed. Rummel, N., Kapur, M., Nathan, M., & Puntambekar, S., pp. 391–398, International Society of the Learning Sciences.Google Scholar
Roberts, S. L. (2013) “Georgia on My Mind”: Writing the “new” state history textbook in the post-Loewen world. The History Teacher 47(1):4160.Google Scholar
Seifert, C. M. & Hutchins, E. L. (1992) Error as opportunity: Learning in a cooperative task. Human–Computer Interaction 7(4):409–35.Google Scholar
Settles, B. (2010) Active learning literature survey. Computer Sciences Technical Report 1648. University of Wisconsin–Madison.Google Scholar
VanLehn, K. (2011) The relative effectiveness of human tutoring, Intelligent Tutoring Systems, and other tutoring systems. Educational Psychologist 46(4):197221.Google Scholar
Zhu, X. (2008) Semi-supervised learning literature survey. Computer Sciences Technical Report 1530. University of Wisconsin, Madison.Google Scholar