Logan: Basic Principles of Learning PRINCIPLE: PARTIAL REINFORCEMENT (PRF) If reinforcement is given irregularly on only part of the trials, conditioning will occur at a slower rate than with 100% reinforcement; ultimate performance is lower in classical conditioning and higher in instrumental conditioning. In classical conditioning, it is not necessary for the US to occur on every trial that the antecedent stimulus occurs; for example, a conditional salivary response becomes associated with a tone that is only followed by food half the time. Similarly, reward need not be given for every instrumental response; rats will run down an alley if they are given food on only half the trials. Such conditions are referred to as PARTIAL REINFORCEMENT (PRF) because reinforcement is given on only part of the trials. (Note: It does not mean giving only part of the reinforcement.) The results that are obtained with PRF are not surprising in the case of classical conditioning: Performance is lower with PRF than with continuous 100% reinforcement (CRF), and the lower the percent reinforcement, the lower the ultimate level of performance. In instrumental conditioning, however, performance with PRF exceeds that with CRF; rats fun faster if they are not given reward on every trial. (These results are with random PRF; if alternate trials are reinforced, most organisms are able to learn the sequence.) This latter observation is called the partial reinforcement acquisition effect (PRAE) and is a bit more complicated than just stated. Actually, performance is higher with PRF during the early and intermediate portions of an instrumental response, but ends up somewhat lower. Specifically, rats start to run more quickly and reach greater speed with PRF, but slow down more as they get near the goal. Overall, they take less time to reach the goal, but they appear to start out and run with overoptimism and end up with over pessimism. The PRAE does not occur if the frequency of reinforcement is very low, say below ten percent of the trials. Partial reinforcement is a special case of VARIED REINFORCEMENT (VRF) where the amount of reinforcement varies from trial to trial. The intensity of US in classical conditioning or the amount of reward in instrumental conditioning need not always be the same, and zero is simply the limiting case of very little reinforcement. Generally speaking, the more widely varied the amount of reinforcement, the greater the difference in performance from that obtained if a large reward is given every trial. However, the fact that rats run faster under PRF and VRF than under CRF does not mean that they prefer such conditions; given a choice, rats consistently choose an alternative in which they get a large reward every time. One likely account of the PRAE involves the FRUSTRATION EFFECT. If rats are run in two consecutive alleys such that the goal box of the first alley is also the start box of the second alley and they received PRF in the first alley and CRF in the second alley, they run faster in the second alley on those trials on which they do not received reward in the first alley. It is argued that not receiving reward when it is expected is frustrating and that frustration is a source of motivation. In effect, the organism is aroused by not getting a customary reward, and this arousal helps potentiate behavior. Now since the rats are rewarded in the second alley, they are reinforced for running when they are frustrated. In order to make the frustration effect account for the PRAE, it is necessary to add the assumption that frustration is a learnable response. That is to say, if an organism is sometimes not rewarded for making a response, there is not only an expectation of rewards but also an expectation of frustration. Then if the organism is sometimes reinforced for responding in the face of possible frustration, that response will be learned and actually potentiated by the additional motivational arousal of anticipatory frustration. The fact that partial reinforcement has different effects in classical and instrumental conditioning has led some theorists to conclude that these conditioning paradigms result in different learning processes. Specifically, classical conditioning might involve a cognitive S-S association while instrumental conditioning involves a behavioristic S-R association. Others have concluded that the nature of the learning process is the same, but that different principles apply. Specifically, contiguity might be sufficient in classical conditioning while reinforcement is also necessary in instrumental conditioning. For whatever reason, nonreinforcement appears to be frustrating in instrumental conditioning but not in classical conditioning. We are talking about more than frustration tolerance, which is the shrug of the shoulder, sometimes-you-win sometimes-you-lose, or learning to take "no" for an answer. If organisms merely tolerated frustration, they would perform equally with PRF and CRF since they would presumably simply dismiss the times of nonreinforcement with a "that's life" attitude. Certainly people do learn to tolerate frustration, and to try, try again. But the PRAE indicates that appropriately trained people try, try again--harder and harder. The appropriate conditions of training are that the person must sometimes fail to get the large reward desired and expected, and then sometimes get that reward for responding adaptively in spite of such frustration. If one sets a realistic athletic goal, that is, one that is reasonably within reach, then each failure will add inspiration for a greater effort on the next try. TERMS: Conditioning (classical, instrumental), effect (frustration, partial reinforcement acquisition), motivation, reinforcement (partial, varied).