Fluent Essays

BRIEF REPORT THE USE OF ERRORLESS LEARNING PROCEDURES IN TEACHING PEOPLE WITH A LEARNING DISABILITY: A CRITICAL REVIEW Robert S.P. Jones a n d C B . Eayrs University College of North Wales, Bangor, Gwynedd, LL57 2DG Abstract The use of procedures which minimise the making of errors is a popular method of teaching skills to people with learning disability. The origin of this approach can be traced t o two distinct sources: the work of B.F. Skinner on programmed learning, and the work of H.S. Terrace on discrimination learning. This early work is reviewed and research findings which highlight the negative side affects of an ‘errorless’ approach are discussed. The role of prompting, attention, reinforcement and generalisation is outlined. Recommendations for the development of teaching programmes are made. Introduction One of the most fundamental aspects of skill acquisition involves the development of stimulus control. It can be argued that the ability of an organism to discriminate between stimuli with which it is presented represents the most basic foundation of learning. The acquisition of an operant discrimination has been defined as the process whereby an organism comes to respond more frequently to a stimulus correlated with reinforcement, S + , than to a stimulus associated with non-reinforcement, S - (Terrace, 1963a). Historically, the learning of an operant discrimination involved the differential reinforcement of responses which occurred in the presence of the S + and the extinction of responses which occurred in the presence of S - . This procedure is most commonly known as ‘trial-and-error’ learning with any response in the presence of S- being regarded as an error. The making of errors in acquiring a discrimination was not only therefore extremely common in the development of stimulus control but it was frequently assumed that the making of errors was an essential occurrence if an organism was to master a discrimination (e.g. Keller & Schoenfeld, 1950). 09.52-9608/92/02 0204-9 $01.80/0 MENTAL HANDICAP RESEARCH 204 0 1992 R . Jones & C. Eayrs Vo1.5, No.2, 1992 USING ERRORLESS LEARNING PROCEDURES 205 In applied work with people with learning disabilities, discrimination learning using trial-and-error methods was limited by a number of factors which involved the making of mistakes or errors. It was felt that for some people with a learning disability the demands of learning by trial and error could provoke problem behaviours (Carr, Newsom & Binkoff, 1980; Weeks & Gaylord-Ross, 1981). Problems such as ‘apathy, aggression, self-injury, negativism and tantrumming’ have been cited as direct consequences of the trial-and-error learning approach (Touchette & Howard, 1984; Hamblin et a/. ,197 1). In the early 1960s H.S. Terrace conducted a series of now famous animal experiments which suggested that not only was the making of errors unnecessary in the development of learning but that learning was actually facilitated if no errors were made during the acquisition of the discrimination (Terrace, 1963a, 1963b,; 1964; 1966). Essentially Terrace’s early work involved the use of trial and error procedure designed to teach pigeons to peck a key illuminated with a red light but not to peck a key illuminated with a green light. When the pigeons pecked the red stimulus ( S + ) they received a food pellet and when they pecked the green stimulus (S - ), nothing happened. Both keys were present at the same illumination during the experiment. A second group of pigeons began the experiment with the green key dark and had the light behind it slowly ‘faded in’ during the training condition. By the end of the experiment the second group also had two illuminated keys of equal intensity present. The second group made very few responses to the green light. Indeed it was as if they hadn’t noticed its presence. The stimulus fading procedure (resulting in few, if any, responses being made to the S - ) was called errorless learning. Although Terrace’s work has not led to major changes in the theoretical explanations of discrimination learning (Robinson & Storm, 1978), this early work stimulated subsequent research which has had a major effect on strategies for teaching people with learning disabilities. In the years immediately following Terrace’s early work, errorless procedures were employed in teaching simple discriminations to children with learning disabilities (e.g. Sidman & Stoddard, 1966, 1967; Touchette, 1968). This work, supported Terrace’s claim that discrimination learning was facilitated by the use of errorless learning techniques. Perhaps the major impact of this work, however, was the finding that children with learning disabilities could be taught by errorless teaching methods in cases where such discriminations could not be acquired by traditional trial-and- error techniques (Walsh, 1985). Another influence on ‘errorless’ teaching technology came from Skinner’s pioneering research on applied behaviour analysis and 206 MENTAL HANDICAP RESEARCH programmed learning (Skinner, 1968). This research suggested that errors could be avoided by breaking tasks down into small steps, by the use of shaping through successive approximations and by the optimal spacing of reinforcers. As a result of these influences, the errorless learning approach was quickly extended beyond basic discrimination training to encompass a complete technology of teaching covering areas far removed from Terrace’s original work. An emphasis on an ‘errorless’ approach to teaching people with learning disabilities became increasingly popular and to this day many books and training manuals owe their origin, overtly or covertly, to this approach (Best, 1987; Donnellan, La Vigna, Negri-Shoultz & Fassbender, 1988; Foxen & McBrien, 1981; La Vigna & Donnellan, 1986; Smith, 1990; Zarkowska & Clements, 1988) Although there can be little doubt that the influence of errorless learning techniques on the training of people with learning disabilities has been of great benefit, a number of problems exist with this approach. Difficulties with Errorless Learning Narrowing of attention Conclusions from a number of studies have suggested that as early as 1966 some researchers were questioning the efficacy of errorless techniques in certain situations (cf. Walsh, 1985). The major conclusions from those studies suggest that fading techniques may confine the person’s attention to very narrow attributes of the stimulus associated with reinforcement (S + ). Richell, for example, questioned the adaptability of a person who had been trained through errorless procedures and particularly the ability to cope in new situations where errors would occur (Richell, 1966). Bijou & Baer (1966) also questioned the ability to transfer skills to other settings. Both of these researchers made these suggestions on theoretical grounds. In 1968, however, Gollin & Savoy conducted a series of experiments which provided experimental evidence for the assertion that in some situations errorless learning may result in a greater number of mistakes compared to trial and error learning once transfer of learning to a new situation is involved. Gollin & Savoy (1968) divided a sample of 52 children into two groups. Both groups were taught a discrimination task followed by a conditional discrimination task (where the nature of the response is dependent on the nature of the discriminative stimulus). One group was taught by a trial-and-error procedure and one by using an errorless fading USING ERRORLESS LEARNING PROCEDURES 207 technique. Although more children who were trained by the fading procedure performed without errors during training, more children in the trial-and-error group solved the conditional discrimination problem. Thus, it appeared that for simple discriminations requiring a single response to a stimulus, errorless methods resulted in fewer errors. When the task became more complicated, however, with the use of a conditional discrimination task then the errorless group did not perform as well. Similar findings have been reported by a number of authors, (e.g. Wolfe & Cuvo, 1978; Walsh, 1985). Walsh (1985) compared errorless and trial- and-error procedures o n a conditional discrimination test. The results of Walsh’s (1985) study were similar to those of Gollin & Savoy (1968). Again errorless learning techniques worked very well when the task was a very simple one requiring only a simple response. When the task became more complicated and necessitated paying attention to more than one stimulus, then errorless learning proved to be an inferior teaching technique when compared with trial-and-error learning. Walsh concluded that ‘under certain conditions fading techniques are not able to provide optimal conditions for learning a given task’ (Walsh, 1985:36). It appears therefore, that access to a combination of both S + and S - throughout training is important in all but the most simple discrimination tasks. New responses In situations where a completely new response was required (as opposed to strengthening an existing behaviour) it was assumed that the optimum mechanism for generating the new response was the prompting of that response and the gradual fading of the prompt as the behaviour became established. No direct experimental evidence was obtained which indicated that this method was, in fact, the optimum means of generating new responses. Rather, this appears to have been an intuitive ‘common sense’ decision and it is possible that the emphasis placed on avoiding errors in accounting-for successful teaching, and the consequent generalisation of the methodology, might be misplaced. Although the work of Skinner (1968) is perhaps the most relevant to the operant teaching of new skills, the term ‘errorless’ has been traditionally applied to the work of Terrace rather than Skinner. The danger here is that this confusion could lead to the assumption that the methodology which is appropriate for teaching pigeons to peck red rather than green illuminated keys will also be appropriate for teaching self-care skills to people with a learning disability. A closer examination of the procedures used, however, MENTAL HANDICAP RESEARCH reveal a number of factors, apart from ‘errorlessness’ which could account for success or failure. In the typical two choice discrimination procedure, the animal or person typically makes a simple ‘pointing’ response. This may either be a pecking response (Terrace, 1963a) or a gestural indication (Cullen, 1978). The prompt to respond is the actual stimulus which is designed to gain control over the behaviour once learning has taken place. Furthermore, at some level a choice is being made from the outset. This can be conceived of as a trial-and-error scenario with the dice loaded towards success. When this procedure is compared with the use of ‘errorless’ procedures to teach self-help skills to people with learning disability a number of differences become apparent. Firstly, the motor behaviour is frequently far more complex than simply ‘pointing’ (e.g. brushing teeth, feeding with a spoon). Secondly, the form that the prompting takes is different. The avoidance of errors is achieved by the physical prompting of movement by the trainer. The person has no choice about what he/she does. At its extreme, the person may be so passive in the procedure that he/she may initially be little more than an extension of the trainer’s own musculature. Thirdly, the discriminative stimuli during training are not always the same as those which are t o gain final control at the end of the training procedure. For example, the final discriminative stimulus for handwashing would normally be the presence of dirt. During training, however, it is the presence of the trainer, and the use of a prompting procedure which acts as a discriminative stimulus. It is these aspects of the training procedure which are so difficult to ‘fade out’. Fourthly, in Terrace’s errorless discrimination learning procedure the S + remained constant throughout whilst S - was ‘faded in’. In the self-help skills teaching situation, however, S + continually changes as prompts are ‘faded out’. Thus, it may be an oversimplification to simply compare errorless versus trial-and-error training unless these other aspects of the environment are controlled. Generalisation A further disadvantage with the errorless approach seems to lie in the area of generalisation. As was mentioned earlier, many errorlessly trained discriminations require a prompting component. In such cases, the person’s ability to generalise a learned skill is wholly dependent on the success with which the prompting component can be gradually withdrawn or faded out. Unfortunately, ‘it is often the case that students who respond appropriately when prompted founder when the prompt is removed’. (Touchette & Howard, 1984:175). In practice, the fading of prompts is a USING ERRORLESS LEARNING PROCEDURES 209 very difficult part of teaching. Premature removal of prompts can lead to persistent incorrect response patterns which preclude acquisition of the target repertoire (Sidman & Stoddard, 1966; Touchette, 1968). If, on the other hand, prompts are presented for an unnecessarily extended period then the person may become dependent on the prompt. Either of these extremes will preclude successful generalisation of learning to novel situations. It is, nevertheless, all too common t o find that a person with a learning disability can only display a learned skill with the aid of a prompt, despite many unsuccessful attempts to fade out prompting. As Touchette & Howard (1984) have said ‘. . . research has not yet resolved the question of how to produce a successful transfer from prompted to unprompted responding’ (p. 175). Reinforcement One of the major advantages of an errorless approach lies in the high frequency of reinforcement available to the person. In teaching people with a learning disability many authors (e.g. Foxen & McBrien, 1981), have advocated that each prompted response should be followed by reinforcement. This continuous reinforcement (CRF) schedule, however, is unlikely to lead to spontaneous generalisation to new situations. There exists a wide variety of research data which suggests that once continuous reinforcement is withdrawn, the acquired behaviour quickly extinguishes (Kazdin & Polster, 1973; Koegel & Rincover, 1977; Kazdin, 1984). Although it has been advocated that, as soon as a behaviour is acquired on a CRF schedule, reinforcement should be switched to an intermittent schedule, there have been a number of difficulties with this procedure (Tierney & Smith 1988; Dehn, 1969). Dehn (1969) found that the smoothness of transfer from continuous to intermittent schedules is a key factor in the success of the procedure. Hamblin et al. (1971) in discussing this work, concluded that ‘after acquisition during the transition from continuous to intermittent reinforcement, negative behaviour is increased in proportion t o the haste with which the transition is made’ (p. 153). Research carried out with children with learning disabilities (Tierney & Smith, 1988), demonstrated that responding on an intermittent schedule during the maintenance phase of a training programme was more likely if initial response acquisition was programmed using a partial schedule than if response acquisition was initially programmed using a CRF schedule and then switched t o an intermittent schedule (Tierney & Smith, 1988). This suggests that even from the beginnings of teaching a discrimination, not all correct responses should be reinforced if optimal response generalisation is 210 MENTAL HANDICAP RESEARCH to be attained. Clearly, this would mean that one of the major aspects asso- ciated with an errorless approach (each response is correct and leads to rein- forcement) could not occur. A dilemma between speed of initial response acquisition and optimum generalisation of responding could thus be engendered. A further issue with regard to reinforcement and subsequent generalisation involves the timing of reinforcement administration. It is frequently assumed that the time delay between responding and contingent reinforcement should be as short as possible and this is usually advocated in teaching materials (e.g. Foxen & McBrien, 1981). In the natural environment, however, there is often a delay between responding and reinforcement and it has been suggested that such a delay may actually facilitate generalisation (Kazdin, 1982). Summary In summary, although errorless learning has undoubtedly provided a technology of teaching which has led to significant advances in the education of people with a learning disability, a number of difficulties remain with this approach. Firstly, fading techniques appear to result in a narrowing of attention which may inhibit the subsequent learning of more complex behaviours. Secondly, where errorless prompting procedures are used, difficulties may arise in the removal of the prompts due to an overdependence on their availability. Thirdly, and perhaps most importantly, there are difficulties with the subsequent maintainence and generalisation of behaviours acquired through errorless methods. Conclusions While the literature clearly points out that a learned discrimination can be achieved and maintained by specifying the setting conditions (prompts) and locating the maintaining events in the natural environment, it is suggested that prompting should be as infrequent as possible and that where possible the learner should have access to a wide range of stimuli in addition to the S + . It can be argued that errorless learning methods should only by used in situations where trial-and-error methods have proven unsuccessful. The preferred treatment approach should be the shaping of the desired response, by successive approximations in the natural environment. Thus, an attempt should be made to teach using trial-and-error methods in the first instance before fading procedures are employed. USING ERRORLESS LEARNING PROCEDURES 21 1 Once the decision to use errorless methods has been made, specific strategies need to be incorporated to take account of prompting, attention, reinforcement and generalisation. Preventing errors at all costs does not seem to merit the same emphasis that Terrace’s early work might suggest. In particular, it is not recommended that overprompting is used t o avoid making mistakes. In terms of subsequent generalisation, it is probably more advantageous for a client to make a mistake in mastering a discrimination than to become dependent on prompting. Each individual, whatever his o r her degree of learning difficulty, brings a unique constellation of skills, experience, and talent t o any new learning situation. It is important to adapt all teaching t o be as flexible and individualistic as possible and not to attempt to fit the learner into a predetermined system of training o r instruction. References Best, A.B. (1987) Steps to Independence: Practical Guidance on Teaching People with Mental and Sensory Handicaps. Kidderminster: BIMH Publications. Bijou, S.W. and Baer, D.M. (1966) Operant methods in child behavior and development. In W.K. Honig (ed) Operant Behavior: Areas of Research and Application. New Y ork : Appleton-Centur y-Cro fts . Carr, T., Newsom, C. and Binkoff, J. (1980) Escape as a factor in the aggressive behavior of two retarded children. Journal of Applied Behavior Analysis 13, Cullen, C.N. (1978) Errorless learning with the retarded. Journal of Practical Approaches to Developmental Handicap 2, 21 -4. Dehn, J. (1969) An Investigation of the Development and Maintenance of the Negative Behavior of Autistic Children. Unpublished Ph. D. dissertation, Washington University (cited in Hamblin et al., 1971). Donnellan, A.M., LaVigna, G.W., Negri-Shoultz, N. and Fassbender, L.L. (1988) Progress Without Punishment: Effective Approaches for Learners with Behavior Problems. New York: Teachers College Press. Foxen, T. and McBrien, J. (1981) Training Staff in Behavioral Methods: Trainee Workbook. Manchester:Manchester University Press. Collin, E.S. and Savoy, P. (1968) Fading procedures and conditional discrimination in children. Journal of the Experimental Analysis of Behavior 1 1, 443 - 5 1. Hamblin, E.L., Buckholdt, D., Ferritor, D., Kozloff, M. and Blackwell, L. (1971) The Humanization Process: A Social, Behavioral Analysis of Children’s Problems. New York: Wiley-Interscience. Kazdin, A.E. (1982) Single-Case Research Designs: Methods for Clinical and Applied Settings. New York: Oxford University Press. -( 1984) Behavior Modification in Applied Settings. Homewood: Dorsey Press. Kazdin, A.E. and Polster, R. (1973) Intermittent token reinforcement and response Keller, F.S. and Schoenfield, W.N. (1950) Principles of Psychology. New York: 101 - 17. maintenance in extinction. Behavior Therapy 4, 386-91. Appleton-Century-Crofts. 212 MENTAL HANDICAP RESEARCH Koegel, R.L. and Rincover, A. (1977) Research on the difference between generalisation and maintenance in extra-therapy responding. Journal of Applied Behavior Analysis 10, 1 - 12. LaVigna, G.W. and Donnellan, A.M. (1986) Alternatives to Punishment: Solving Behavior Problems with Non-aversive Strategies. New York: Irvington. Richell, M. (1966) L’apprentissage sans erreurs. L’Annee Psychologique 2, 535-43 (cited in Walsh, 1985). Robinson, P.W. and Storm, R.H. (1978) Effects of error and errorless discrimination acquisition on reversal learning. Journal of the Experimental Analysis of Behavior 29, 517-25. Sidman, M., and Stoddard, C.T. (1966) Programming perception and learning for retarded children. In N.R. Ellis (ed.) International Review of Research in Mental Retardation Vol. 11. New York: Academic Press. - (1967) The effectiveness of fading in programming a stimultaneous form discrimination for retarded children. Journal of the Experimental Analysis of Behavior 20, 3- 15. Skinner, B.F. (1968) The Technology of Teaching. New York: Meredith Corporation. Smith, M.D. (1990) Autism and Life in the Community: Successful Interventions for Behavioral Challenges. Baltimore: Paul H . Brooks. Terrace, H . S. (1963a) Discrimination learning with and without errors. Journal of the Experimental Analysis of Behavior 4, 1-27. - (1963b) Errorless transfers of a discrimination across two continua. Journal of the Experimental Analysis of Behavior 6, 223-32. - (1964) Wavelength generalisation after discrimination learning with and without errors. Science 114, 78-80. - (1966) Stimulus control. In W.K. Honig (ed.) Operant Behavior. Areas of Research and Application (pp 271 - 344). New York: Appleton-Century- Crofts. Tierney, K.J. and Smith, H.V. (1988) The effects of various combinations of continuous and partial reinforcement schedules on response persistence in mentally handicapped children. Behavioral Psychotherapy 16, 23 - 37. Touchette, P.E. (1968) The effects of graduated stimulus change on the acquisition of a simple discrimination in severely retarded boys. Journal of the Experimental Analysis of Behavior 11, 39-48. Touchette, P.E. and Howard, J.S. (1984) Errorless learning: Reinforcement contingencies and stimulus control transfer in delayed prompting. Journal of Applied Behavior Analysis 17, 175-88. Walsh, P.G. (1985) Teaching colour discrimination t o the mentally handicapped. The Irish Journal of Psychology 7, 36-49. Weeks, M. and Gaylord-Ross, R. (1981) Task difficulty and abberant behavior in severely handicapped students. Journal of Applied Behavior Analysis 14, 449-3. Wolfe, V. F. and Cuvo, A.J. (1978) Effects of within-stimulus and extra stimulus prompting on letter discrimination by mentally retarded persons. American Journal of Mental Deficiency 83, 297-303. Zarkowska, E. and Clements, J. (1988) Problem Behaviour in People with Severe Learning Disabilities: A Practical Guide to A Constructional Approach. London: Croom Helm. JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR TWO "SYNTHETIC SOCIAL RELATIONS" B. F. SKINNER HARVARD UNIVERSITY About 10 years ago, two demonstration ex- periments were designed for a General Educa- tion course in Human Behavior at Harvard. They were briefly described in an illustrated weekly and are occasionally referred to in the psychological literature. It seems advisable to publish a somewhat more explicit account. THE "PING-PONG" PLAYING PIGEONS There were several versions of this ap- paratus, in one of which a motor-driven device returned the ping-pong ball to the playing surface so that the apparatus ran without at- tention. In a less mechanized version, the "ping-pong" table was approximately 8 in. wide, 16 in. long, and 8 in. high (Fig. 1). A pigeon standing at one end could conveniently peck a ball as it arrived at the edge of the table. If the ball rolled off the edge, it fell into a trough and tripped a switch which operated a food dispenser under the opposite edge and thus reinforced the pigeon which "won the point." Light metal rails prevented the ball from falling off the sides of the table. The surface was slightly canted, sloping from a center line toward each edge so that the ball would not stop on it. Wire barriers prevented the pigeons from jumping up on the table but did not interfere with play. In the finished performance, the demon- strator would start a ball near the middle of the table. It rolled to one edge and the pigeon on that side pecked it, driving it back across the table. At the other edge it was pecked by the other pigeon and thus returned. The pigeons usually watched the course of the ball as it crossed the table, and maneuvered into position to meet the return. They developed considerable skill in sending the ball straight across. Moving pictures show rallies of as many as five or six shots before a point was made. There is no evidence, however, that either pigeon reached the stage of placing or chang- Fig. 1. Two pigeons "playing ping-pong." 531 VOLUME 5, NUMBER 4 OCTOBER, 1962 B. F. SKINNER ing the pace of its shots so that the opponent would miss. Conditioning was begun with one pigeon at a time. A standard table tennis ball was fastened at the edge of the table, and a hungry pigeon was reinforced with food when it pecked it. At this stage the ball was not a powerful controlling stimulus; when it was moved to a different part of the edge, the pigeons often pecked the air where it had been. Eventually, however, they pecked the ball regardless of its position. The ball was then made free to roll away from the pigeon when struck. A mechanical reinforcing system was set up in which the ball, rolling up a slight grade, struck a cross-bar operating the food dispenser. The distance to the bar was gradually increased. If the ball failed to reach the bar, it rolled back and came to rest against a raised molding along the edge. The molding was later removed. As the distance between the pigeon and the reinforcing bar was increased, reinforcement was more and more delayed, and the behavior occasionally suffered. Eventually, however, mediating behavior arose to bridge the tem- poral gap. Even so, in the final game, in which two pigeons participated, the delay between striking the ball and the successful outcome of getting the ball past the opponent was occa- sionally troublesome. A deteriorating perform- ance could be rescued by reinforcing a pigeon with a hand-switch at the moment it struck the ball. Eventually the behavior was sustained not only for rallies of several shots at a time but for a full "game." The demonstration offers a convenient ex- ample of competition. One bird is reinforced at the expense of another. If one is repeatedly successful, the other suffers extinction ("dis- couragement"). It was possible to maintain a reasonable balance in successful play by lower- ing the weight of the relatively unsuccessful bird or raising that of the successful, the prin- cipal effect being to sustain attention rather than alter accuracy or power. COOPERATING PIGEONS Two pigeons in adjacent compartments were separated by a pane of glass. Three red buttons were arranged in a vertical row on each side of the glass, as shown in Fig. 2. The buttons were approximately 10 in., 7-1/2 in., and 5 in. from the floor, respectively. By peck- ing a button the pigeon closed a switch. In the final performance, both pigeons were rein- forced with food (Fig. 2, below) when they pecked a corresponding pair of buttons so nearly simultaneously that the brief closures of the circuits (each lasting perhaps a tenth of a second) overlapped. At any given time, how- Fig. 2. Above: Two pigeons cooperating by pecking corresponding buttons at the same moment. Below: Pigeons eating from food dispensers. ever, only one pair of buttons was operative, and the effective pair was scheduled in a roughly random way. It was necessary for the pigeons to cooperate in two tasks: (1) discovering the effective pair and (2) pecking both buttons at the same time. In general, no pattern of exploration could be observed. The pigeons tested all three pairs of buttons in what was evidentally an unsystem- atic way. In general, there was a division of "SYNTHETIC SOCIAL RELATIONS" labor with respect to the two tasks. One pigeon (the "leader") explored-that is, it struck the three buttons in some order. A similar per- formance could have been generated in one pigeon alone in the apparatus by requiring simply that a given one of three buttons be struck. The other pigeon (the "follower") struck the button opposite that being struck by the leader. Similar behavior could have been generated in one pigeon alone in the apparatus if one button after another had been marked by a discriminative stimulus. A well-marked leader-follower relation could be established or reversed by altering the relative level of food deprivation, the more deprived bird assuming the position of leader by moving more alertly to the buttons. How- ever, even a decisive leader was probably to some extent following. A deprived pigeon would usually "wait to be followed" by one less deprived before exploring the buttons vigorously. Under levels of deprivation at which both birds responded quickly and with- out interruption, performance became so per- fect that it gave the impression of one pigeon seen in a mirror. The performance was established by con- ditioning each bird separately to peck the three buttons, reinforcement being roughly randomized. When sustained behavior oc- curred on all three buttons, two birds could be put in the adjacent spaces for the first time. The presence of another bird temporarily dis- turbed the performance, but both birds eventually began to respond to the buttons. At this stage responses to corresponding but- tons within, say, half a second of each other would trigger both food-dispensers. These contingencies sufficed to build cooperative be- havior without further attention. The visual stimulation supplied by one pigeon pecking on a button became a discriminative stimulus controlling a response to the corresponding button on the part of the other. Prolonged exposure to these conditions made pigeons strongly imitative in other re- spects. They would often drink from glasses of water in the compartments at the same time, for example. The extent to which their be- haviors were mutually controlled was infor- mally demonstrated when the experiment was shown to a group of biologists, one of whom suggested putting the birds in the opposite compartments. The birds immediately lined up alongside the glass plate, facing away from the buttons. They thus assumed their pre- viously effective positions relative to each other, but were now facing the audience through the transparent front wall of the ap- paratus. Though no buttons were available, they immediately began to cooperate in ex- ploring a corresponding area, bobbing up and down in a perfect mirror-image pattern under the control of each other's behavior. Possibly because the leader-follower relation had fre- quently been shifted, each bird was evidently largely controlled by the behavior of the other. Supported by NSF Grant G18167. 533