Information-Sharing in Three Interacting Minds Solving a Simple Perceptual Task

Information-Sharing in Three Interacting Minds Solving a Simple Perceptual Task Michal Denkiewicz (michal.denkiewicz@gmail.com) Department of Psychology, University of Warsaw Stawki 5/7, 00-183 Warsaw, Poland Joanna Rączaszek-Leonardi (raczasze@psych.uw.edu.pl) Institute of Psychology, Polish Academy of Sciences Jaracza 1, 00-378 Warsaw, Poland Piotr Migdal (piotr.migdal@icfo.es) ICFO–Institut de Ciencies Fotoniques 08860 Castelldefels (Barcelona), Spain Dariusz Plewczynski (darman@icm.edu.pl) Interdisciplinary Centre for Mathematical and Computational Modelling, University of Warsaw Pawinskiego 5a, 02-106 Warsaw, Poland Abstract We study three-person groups solving a simple, two alternative forced choice task of perceptual nature. The group members provide individual answers and afterwards discuss and reach a joint decision. Different models of information sharing describe the theoretical relationship between group and individual performance. Experimental data shows that average-performing members can benefit from cooperation, but the groups do not outperform their best members. Results point to voting as the best explanation of the behavior of the groups. Keywords: group decision making; distributed cognition; information sharing Introduction Whether “two heads are better than one” is no settled question in psychology. Many studies report groups to be less proficient than their most capable members (Corfman & Kahn, 1995), or that there is no benefit (Heaney, Foster, Gregor, O’Neill, & Wood, 2010). Groups are often regarded as source of negative influence on individual performance, stemming from conformism (Asch, 1951) or social loafing (Allport, 1924). There are, however, studies that report benefit from cooperation (Kerr & Tindale, 2004; Hastie & Kameda, 2005). In one of such studies Bahrami (B. Bahrami et al., 2010) determined that dyads can outperform their members in a simple two-choice perceptual task, provided that those members have similar individual effectiveness. This group benefit disappeared when communication was forbidden; hence free information sharing was a key factor. This result is interesting because there is no obvious reason for group benefit to occur in such a simple task. For certain types of tasks, such as concept mastery, concept attainment or learning, group members can pool cognitive resources, or utilize complimentary skill or knowledge. Then a group can provide better solutions in terms of quality, though not necessarily efficiency (Steiner, 1966; Hill 1982). A notable exception is brainstorming, where participating in a group has a negative impact both on quality and quantity of creative solutions (Taylor, Berry, & Block, 1958). If resources cannot be shared, perhaps solutions can be. For so-called “eureka-tasks” a solution can be demonstrated in objective terms, thus (at least in theory) a single participant, who finds the correct answer, can easily persuade other members, leading to a correct group solution. Hence, the chance of group solving a task grows with its size. Examples of such task are Remote Associates Test (Laughlin & Bitz, 1975), or simply scrambled letters/anagrams. In Bahrami’s study neither of these conditions was met. The participants first performed the perceptual task on their own, without the ability to divide it into parts. Then a group decision was made, based solely on what the participants perceived individually. No reasoning or previous knowledge or was of any use and the only thing, that the participants could communicate was their subjective idea, of what they think the answer was. Still not only did the pairs perform better, than chance (which in this case means the averaged effectiveness of the two participants), but they also outperformed the better of the two members. Bahrami tested several theoretical models of information sharing in communication, developed in the spirit of signal detection theory, and concluded, that dyad members communicate their own relative confidences, which allows the group benefit to occur. Later these models were theoretically extended to groups of arbitrary size by Migdal (Migdal, Rączaszek-Leonardi, Denkiewicz & Plewczynski, 2012). “Aggregation of decisions” was also considered, that is a situation when subgroups of a larger group first reach their decisions, and then try to convince one another. The topic of this paper is an experimental attempt to verify the applicability of these theoretical models to three-