Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
The partial reinforcement extinction effect (PREE) Frode Svartdal University of Tromsø Extinction: Basics Extinction is defined in terms of a reinforcement process Extinction contingencies The stimulus (SR or US) is discontinued The learning contingency is discontinued Extinction process The conditioned response is reduced (strength, frequency, etc.) Relearning, … not forgetting Extinction: Basics Operant conditioning Catania, 1984) Extinction: Basics Classical conditioning Factors affecting the extinction rate In general: Fast acquisition / high rate of responding fast extinction Amount of reward Variability High fast extinction Stimulus Response Reinforcement = high ext. persistence Some forms of learning do not extinguish (easily) Evaluative conditioning (e.g., Diaz, Ruiz, & Beyens, 2005) Factors affecting the extinction rate Partial Reinforcement Extinction Effect Partial (Intermittent) Reinforcement (PRF) increased extinction response Continuous Reinforcement (CRF) reduced extinction persistence First demonstrations Operant conditioning; free operant; rats; Skinner (1938) 50% 100% Classical conditioning; blink response; students; Humphreys (1939) Free operant Ferster & Culbertson, 1975 Free operant Compared to CRF: PRF • higher asymptotes • more persistent responding under extinction EXTINCTION PRF CRF Rats, maze running speed under extinction (Weinstock, 1954) PRF (30%) CRF Classical conditioning (rats): PREE Extinction 25% PRF response rate LOWER than CRF response rate 50% 100% 15% Classical conditioning; eyelid; human subjects (Svartdal & Flaten, in prep.) Operant conditioning; humans; Svartdal, 2003, Exp. 4 Conclusions (… preliminary) PREE is a very robust outcome Measures & species Bar pressing, rats Maze running, rats Pecking, pigeons Blink reflex, humans, rabbits … Contingency Operant/instrumental Discrete trial Free operant Classical But… How general is the PREE? Alternative methods of analysis Reversed PREE observed under some conditions Generalized PREE observed under some conditions Nevin (1988): ”PREE is an artefact because of wrong method of analzing extinction performance” Response unit issue PREE or not dependig on how the response is defined (Mowrer & Jones, 1945! Reversed PREE What happens if the subject is exposed to a mixture of PRF and CRF contingencies? Reversed PREE Pavlik & Carlton, 1965: Rats; bar pressing, free operant Gr. 1: Single contingency; CRF Gr. 2: Single contingency; PRF Gr. 3: Two signalled schedules alternated for the same subjects; CRF + PRF Reversed PREE Conventional PREE Reversed PREE Reversed PREE Reversed PREE Pavlik & Carlton (1965): Single reinforcement schedules (CRF vs. PRF) in betweengroups experiments PREE Two schedules (CRF vs. PRF) for the same subjects Reversed PREE Other research Reversed PREE observed Generalized PREE (overall increased persistence, but no difference between conditions) Conventional PREE rarely if ever observed in within-subjects manipulations of CRF - PRF PREE as a generalization: Ecological validity If applied to a situation with a very specific schecule for a specific behavior PREE Example: Single mother – child is begging for toys only from mom If applied to various situations with mixed contingencies Reversed PREE Generalized PREE Example: Mother and father – child begs for toys from both PREE as a generalization Relevance to ADHD Complex schedules RPREE or GPREE, not PREE Factors associated with slower learning slower extinction Attentional problems, difficulties with concentration, memory, … Would add to biological factors Response unit issue Free operant responding: What is the response unit? Mowrer & Jones,1945: What should be counted as the response unit - single responses or the unit of responses required for reinforcement? Free-operant Intermittent reinforcemet, e.g., FR4 Response unit FR4 Reinforced responses PREE Total responses Reversed PREE Total responses / reinforcement ratio Nevin: PREE is an artefact PREE: Alternative analyses Nevin, 1988: Behavioral momentum • ”RPREE” is the rule – the response is stronger following CRF • • • in free-operant responding (but not in discrete-trial experiments) following extended training Extinction performance • • Traditional measure: Number of responses Nevin: Slope of the extinction curve SHORT LONG Nevin, 1988 Absolute number of responses PREE Relative to initial ext response level RPREE PREE vs. RPREE – important variables Dependent measure Type of situations Free operant vs. discrete trial Complexity of situation No. of responses vs. relative change One vs. more schedules (e.g., multiple schedule) Design Between groups vs. within subjects PREE typically observed Measure Number of responses Situation Discrete trial Schedule Single Design Between-groups manipulation of reinforcer rate Other CRF schedule must be 100% PREE: My interests Interaction PREE & Reversed PREE Cognition (verbalization) related to behavioral PREE The experimental situation ”Computer responses” presented Left, right Subject responses recorded Left, right The experimental situation Task Complete a four-response chain of responses started by the computer ”Obtain as many correct answers as you can.” Rules (depending on experiment) Subject: R L Instructed task: Identify and apply the functional rule(s) E.g.: Computer: L R ”Repeat computer sequence” ”Reverse computer sequence” Feedback (visual, autitory) for correct answer; nothing happens if answer is incorrect The experimental situation Manipulations (between groups and/or within groups) Rule Reverse (typically used) Repeat Contingency CRF (100%) PRF (20-60%) The experimental situation Reward rate manipulated Between groups Within subjects (multiple schedule) Discrete trial situation; fixed number of trials 180 acquisition trials 40 extinction trials Conventinal PREE; operant responding; students; Svartdal, 2003, Exp. 4 Reversed & conventional PREE; operant responding; students; Svartdal, 2000 Reversed PREE Purpose: Explore the relationship between PREE and RPREE PREE vs. RPREE: Contradiction or compatible effects? Method Independent groups: PRF and CRF Within: CRF and PRF Svartdal, 2000 ctd. Multiple schedule, alternating Group 40/40 PRF Group 80/80 Half trials (signalled): 40% Half trials (signalled): 40% Half trials (signalled): 80% Half trials (signalled): 80% ”CRF” Group 80/40 Half trials (signalled): 80% Half trials (signalled): 40% ”CRF” + PRF * No. of responses: RPREE * Relative change: No difference PREE 80% 40% Svartdal, 2000 ctd. Relationship between schedule components Simplest assumption: Modulation between component schedules: 60% + context = 60% reference 60% + context = 100% reduced persistence 60% + context = 20% increaced persistence Performance of a 60% schedule depending on other schedule = 100%, 60%, or 20% Svartdal, 2000 Svartdal, F. (2000). Persistence during extinction: Conventional and Reversed PREE under multiple schedules. Learning and Motivation, 31, 21-40. Cognition in PREE • Currently: Strong cognitive arguments to interpret conditioning in terms of cognition • • • • • Classical conditioning: Lovibond & Shanks, 2002 Operant conditioning: Shanks & St John, 1994 Implicit learning doubted: Shanks, 2005 Extinction: Lovibond, 2004 Basic argument: CONTINGENCY CONSCIOUS APPREHENSION BEHAVIORAL CHANGE CONTINGENCY CONSCIOUS APPREHENSION NO BEHAVIORAL CHANGE • Large number of studies supporting this assumption Cognition in PREE So, since the behvioral PREE is very robust, a ”cognitive PREE” must be easy to measure Basic prosedure: Behavioral acquisition under 100% vs. 60% reinforcer rate Measurement of verbalized PREE Cognition in PREE Prediction of persistence: ”How likely is it that you will continue responding if reward no longer appears?” Several experiments have demonstrated no sensitivity to learning history in predictions 3 extinction trials; immediate behavioral sensitivity No difference in predictions 1,0 0,9 0,8 0,7 0,6 100/100 60/60 0,5 0,4 0,3 Responses (proportion of possible) 0,2 0,1 0,0 Trl1 Trl3 Trl5 Trl7 Trl9 Trl11 Trl13 Trl15 Trl17 TExt Svartdal & Silvera, in prep. Cognition in PREE Retrospective judgments: ”How many responses did you emit after reward no longer appeared?” Subjects are very accurate in descrbing their own behavior, including their own extinction persistence Cognition in PREE Svartdal, F. (2003). Extinction after partial reinforcement: Predicted vs. judged persistence. Scandinavian Journal of Psychology, 44, 55-64. Meta-cognitive PREE? We all have long experience with various contingencies Maybe a ”meta-cognition” evolves: Uncertain outcomes Persist Certain outcomes Quit Meta-cognitive PREE? Scenarioes presented to subjects, manipulation Reliable outcome vs. Unreliable outcome Persistence judgments of behavior Meta-cognitive PREE? Naive students: No effect of outcome manipulation Meta-cognitive PREE? Psychology students Naive students (have read about PREE) Meta-cognitive PREE? Svartdal, F. (2000). Persistence during extinction: Are judgments of persistence affected by contingency information? Scandinavian Journal of Psychology, 41, 315-328. PREE: Theory Mowrer & Jones: Diskriminasjonshypotesen PRF: Læringbetingelsene ekstinksjonsbetingelsene Generalisering til ekstinksjon CRF: Læringbetingelsene # ekstinksjonsbetingelsene Liten generalisering til ekstinksjon PREE: Theory Amsel: Frustrasjonshypotesen PRF: Forventning om belønning frustrasjon når belønning uteblir Frustrasjons-cues assosieres med læringssituasjonen Under ekstinksjon: Frustrasjon pga uteblitt belønning Læringssituasjonen ekstinksjonssituasjonen CRF: Frustrasjon oppstår ikke under læring Læringssituasjonen # ekstinksjonssituasjonen PREE: Theory Capaldi: Sequential hypothesis PRF: Ikke-belønnede trials blir signal på at belønning snart vil følge: … N N N R N N N R … Dvs.: Det opparbeides en forventning om belønning når belønning uteblir Under ekstinksjon: Mange responser pga forventning om belønning CRF: Ingen erfaring med uteblitt belønning under læring Under ekstinksjon: Få responser PREE: Theory Status: Discrete-trial-situasjonen Diskriminasjonshypotesen står svakt Amsels hypotese står rimelig sterkt Capaldis hypotese står ganske sterkt Nevins modell: Ingen hypotese i vanlig forstand Capaldi og Amsel dominerende Fri-operant-situasjonen Svak teoretisk forståelse Evaluative conditioning Neutral stimulus ζ Neutral stimulus paired with US E.g., E.g., ζ + MURDER Extinction ζ alone Test ζ is evaluated in accordance with MURDER