The power of prediction
In this previous post, I discussed the intriguing processes of 'Reconsolidation': when memories are reactivated by retrieval they become malleable and may be re-consolidated in an updated form. The evidence regarding the reconstructive nature of memory may have possible implications for fields like mental health treatment, and importantly, education. Here I continue to explore reconsolidation by looking at recent research exploring the conditions for human reconsolidation, discussing possible applications in education, and raising some questions with an aim to further decipher the phenomena - at the cognitive and educational levels.
Open questions
In the previous page we saw that Reconsolidation is the post-activation re-stabilization of an existing memory. Some major open questions are whether reconsolidation is the mechanism for updating memories in the brain and whether it necessarily occurs after every reactivation, If not, what are the conditions under which reconsolidation occurs. A fascinating review of the topic by prof. Jonathan LC Lee (1) explores the evidence and suggests a compelling summary and interpretation:
Reconsolidation does not necessarily take place after every reactivation of any memory.
The evidence may suggest that reconsolidation occurs when the new experience is likely to update the existing one (rather than adding no new information or encoded separately).
It is suggested that a mismatch between a prediction that is based on the existing knowledge and the new experience (i.e. prediction error) triggers the reconsolidation of the former.
This interpretation of the evidence points to a role of reconsolidation in updating memories "to maintain their predictive and adaptive relevance".
Reconsolidation and prediction in humans
Reconsolidation was demonstrated in humans' episodic memory (conscious memory for events). A study(2) that was described in this post used two groups of objects learned two days apart, as the "Existing memory" and "new learning". The study findings highlight the importance of a reminder or reactivation prior to new learning and the passage of time for reconsolidation. Additional studies corroborated these findings and further explored the specific conditions for reconsolidation, for example, the nature of the reminder and whether it matches or mismatches with the new information(3). However, in most of these studies, highly discrete stimuli were used (unrelated objects and nonsense word pairs, respectively) without any meaningful connections between them. In an educational context ,we are interested in retrieving prior knowledge that will serve as a basis for new learning, and a lot of effort is invested in making these connections (read more here). In a recent study (4) the researchers took one additional step toward this goal: Alyssa H Sinclair and Morgan D Barense set to investigate reconsolidation in humans, using more naturalistic stimuli with a narrative nature, they set to study the conditions for reconsolidation, specifically addressing the contribution of prediction error.
The role of the reminder
In this study, 24 participants viewed a series of 18 short video clips of scenes that included action and its outcome. On the next day, they watched the same videos again as a reminder, before watching a second series of different videos of matching themes. On day 3, the participants were tested for their memory of the details from the first series of videos. The main question was whether the reactivation of series 1 videos prior to watching series 2 videos on day 2, would lead to the reconsolidation of the original memories (series 1), and whether new information would be incorporated into them.
The role of surprise
to specifically test the role of prediction error (when a new experience violates the memory-based expectations), the researchers divided the 18 series 1 videos into two experimental conditions: full reminder videos (full squares) and reminder videos in which the outcome part of the clip was cut out (corner-cut squares). The question was whether the element of surprise (or violation of prediction) has an effect on the way the original trace and how it is updated by new information.
Control group
To control for the reminder effect, a second group of 24 participants went through the exact same procedure but on day 2 they watched the series 1 videos after watching the series 2 videos, so for this group, there was no induction of memory by reactivation prior to new information presentation. All the participants were tested on series 1 videos on day 3.
How do the reminder and the element of surprise affect the memory of the original information?
Control Group (no reactivation)
Results
To find out, the researchers compared the number of correct details from series 1 videos, the number of errors (incorrectly remembered details) and the number of intrusions (details from series 2 videos that were incorrectly associated with series 1 videos), among the experimental conditions:
they found:
No difference in the number of correctly remembered details.
More Errors when in the reactivation vs. no reactivation condition.
Importantly: more intrusions in the surprising activation vs. not surprising activation and no activation (see figures: right & below)
These results indicate that when an existing memory trace is activated it becomes malleable and may be updated, however when the predictions that this memory yields are violated, it is more likely to be updated by new information. These results are in line with the idea that the reconsolidation process has a role in updating existing memories with new information and that the process is sensitive to the nature of the reactivation and learning experience.
Long vs. Short term
One last criterion to consider is the passage of time. Reconsolidation is a biological process that requires time: its effects should only be evident in the long term. To further support the thesis that the updating process was dependent on reconsolidation the researchers set to compare the 1-day delay to the test (considered long-term) with an immediate test. To that end, they repeated the same experiment with some modifications, the major one being that the test took place on day 2, immediately after learning (rather than one day later). The results support the reconsolidation thesis: no effects were evident in the short term, demonstrating dependence on the passage of time. In the charts below (adapted from the original study, with the generous help of first author Allie Sinclair) you can see the patterns of results in the original (1-day test) experiment and the second (immediate test). Note how the differences between the control and experimental groups as well as between the full and 'no outcome' reminder conditions disappear. Please check out the original study (4) for complete analyses, additional results and in-depth interpretation.
Figures are adapted with permission from Sinclair and Barense, 2018.
Conclusions
This study replicated previous results (e.g. 2), demonstrating a process of updating existing memories with new information that is dependent on both reactivation prior to new learning and the passage of time. Furthermore, it highlights the value of surprise, or prediction error: memories are more likely to be updated when the predictions they yield are violated. In this study, as in the previous one (reviewed on this page), the memory for correct details remained intact after reactivation, while new details were added: some errors but mostly information from the new experience. These new details are called "intrusions" as the research focus is on the original memory traces and the changes they undergo. In educational settings, we are more likely to think about retrieving relevant prior knowledge prior to acquiring new information. I hope it becomes clear how this line of research can advance our understanding of the intricacies of the learning process, and enrich our ability to both evaluate and formulate effective approaches for teaching. In what follows I try to highlight those aspects and ask questions for future discussion and research.
From research to education
The important role of prediction makes a lot of ecological sense - the very essence of learning is to be able to store relevant information for future use. In instances when the predictions this stored knowledge produces are detected as inaccurate, they are more likely to undergo modification and updating. We are already familiar with a set of important conditions for acquiring new knowledge. This research validates some of them and highlights a possible role for prediction, which I find particularly illuminating.
Acquiring new knowledge
To demonstrate the conditions for acquiring new knowledge I use the pyramid model for knowledge (introduced here) where every triangle is a 'piece of knowledge'. Turquoise triangles are existing knowledge, Orange triangles are new knowledge, and the correct way to structure them one on top of the other represents meaningful connections. After learning, the structure is dependent on the process of consolidation (and reconsolidation) for long-term storage and possible future use.
The suggested conditions for the successful acquisition of new knowledge:
1. Prior knowledge exists- any new information is learned on the basis of existing knowledge. We need to make sure that the relevant prior knowledge already exists (read more).
2. Prior knowledge is active - this study and many others clearly and directly show that the activation (i.e. recalling the information) prior to learning is essential for knowledge integration. We can think about it this way: in order to build the pyramid, the bricks should be in place, not just in storage.
3. Trigger a prediction - while retrieving the knowledge, we should think of how we can use it and what predictions can we make. What questions can we answer, what questions are open and how new information can help us use the current knowledge more effectively? It is like considering where the new bricks can or should be placed.
4. Make meaningful connections - using the open questions and the focus on missing information we can now add the new information while creating connections that make sense in light of what is already known. In the model, a new brick should be carefully placed in position, rather than wherever it falls. At times, the new information will not be aligned with predictions and may require additional construction: either teaching more information or recalling more relevant information from memory.
5. Allow time for consolidation - To complete the process of integration and updating the traces are dependent on consolidation and reconsolidation. It, therefore, makes more sense to check for understanding, the ability to recall the new information in terms of the existing, in the longer (days) term rather than immediately.
These stages take place naturally in many thinking processes, when we encounter a question, a problem or new information we naturally try to assess what we know, how we can use this knowledge and what is still missing in order to process the new information meaningfully. However, as teachers, we should not assume that these processes take place naturally or automatically for all students. It is beneficial if we consider them when planning how a new concept is learned. These explicit stages can shape the way we use several effective teaching strategies, and raise questions and direction and future research (in the field or in the lab):
Teaching strategies
Review - the first three stages above highlight the value of reviewing prior knowledge immediately before learning. Advanced students may intuitively engage in this process effectively (retrieving information and trying to make predictions), but novice students would benefit from guidance on this process by using several strategies to prompt retrieval and trigger predictions, for example:
Retrieval practice - an effective way to activate existing knowledge prior to new learning. The accumulating evidence about reconsolidation may suggest that not all retrieval attempts are equal: Unsuccessful or incomplete attempts are more likely to promote learning and importantly this depends on the experience that follows the retrieval attempt: the correction or learning something new. This is also in line with Bjork's desirable difficulties idea, the "difficult parts" are the aspects that yield prediction errors and hence promote the integration of new information.
Pretesting - Several studies show that attempting to answer test questions before learning improves memory for learned content (e.g. 5), however other studies (6) did not find such benefits. The findings regarding prediction error may help us refine the possible benefits of this effect: I want to suggest that pre-testing is beneficial mostly in cases where making predictions is possible. That is, when learners have enough prior knowledge and when the questions trigger prediction. In addition, the congruence between the prediction and the new knowledge is expected to play a role, as well as the time to the memory test. Interestingly, the effects seen in (5) were targeting unsuccessful retrieval attempts, and the non-effects (6) were evident in short-term test.
Questions Formulation Technique - recently I have come across this technique developed in detail by Dan Rothstein and Luz Santana from the Right Question Institute that is focused on guiding students in asking questions and making predictions as an effective way to ignite learning: the Question Formulation Technique is explained in detail in their book (7) and on their website. In short, the teacher presents a carefully selected piece of information as a trigger (may be a picture of a statement) and the students engage in an intensive guided session of generating questions, sorting and prioritizing them. This quick and carefully structured method, aimed to exhaust students' efforts to use existing knowledge for making predictions and articulating uncertainties and specific questions, structuring the basis layer for new learning which is as wide as possible, active, explicit and transparent to both learners and teachers.
Each of these strategies, when considered in light of the findings described above about prediction and reconsolidation, triggers a set of refined practice-oriented research questions that may shed more light on theoretical ad practical aspects of learning. For Example:
Retrieval practice as a trigger for knowledge acquisition - the evidence suggests that retrieval practice is effective for strengthening memory for previously learned material, it is interesting to ask how retrieval for previously learned material affects learning of new information. It would be valuable to systematically evaluate the contribution of retrieval practice for previously learned content A and newly learned content B, by comparing retrieval practice for A at the end of lesson A and beginning of lesson B, with and without sequential teaching of content B.
Conditions for the PreTesting effect: the evidence reviewed here and their interpretation suggests that the following factors will probably influence the benefits of the pretesting effect: 1) amount of previous knowledge on the subject. 2) Temporal proximity between the pretesting and learning. 3) Retention time between learning and memory testing. I predict that more prior knowledge, greater proximity, and longer-term memory testing would yield better results.
Value of triggering prediction - More specifically, both the evidence and the practice lead to the question about the power of prediction: what is the value of triggering predictions about specific aspects of the to-be-learned material. what is the value of specific prediction and how it can be differentiated from the effects of retrieving previously learned material and what role does it have in the preTesting effect.
Contribution of congruence - as the cognitive neuroscientific research suggests that congruency between predicted and experienced information is valuable, it would be valuable to address this issue in each of the directions suggested above.
Until we have (hopefully) all these answers I think it is still valuable to think about all these stages of the learning process and design learning activities that would address them, I would go with a retrieval practice activity at the beginning of a lesson to reactivate and evaluate previously learned material, with a quick whole-class feedback, followed by questions or activity to trigger predictions about the core aspects of the newly learned material. There are more than a few examples of such sequences: e.g. this formulation of Barak Rosenshine's "Principles of instruction" (8) by Tom Sherrington, illustrated by Oliver Caviglioli and the chapters about "Retrieving" and "Predicting" in James Lang's Book 'Small Teaching' (9).
From education to research - questions for further investigation
Finally, another set of questions explores the basic cognitive science from an educator's point of view:
Is curiosity dependent on existing knowledge and triggering predictions? in educational conversations curiosity and its relation to intrinsic motivation are common. Recently, cognitive researchers, Gruber, Gelman, and Ranganath demonstrated a relation between curiosity ratings of trivia answers and memory of the answers(10). My question is to what extent can we define curiosity in terms of 'having enough background knowledge about the subject combined with an opportunity to generate predictions' or identifying opportunities to learn something new and useful (and I think bout natural conditions like a tiger in the forest and educational conditions like students in the classroom at the same time). Is it possible to combine the research questions about curiosity and about prediction error, when concerning declarative learning?
How prediction error affects the acquisition of new knowledge? The study above was conducted with the malleability of the "existing" memory trace asking how it is updated after a surprising reactivation. It did not address the consequences of the new information that served as "interference". The results, however, raise many more questions about the conditions for acquiring new knowledge: how the new information is integrated with the existing, and how reactivation and prediction error affect the learning of the new information? Is there a dissociation in the kind of processes triggered by confirmation or violation of the prediction? (there is a good reason to believe that there is, see 11), What more can we learn when the "existing" and "new" materials are meaningfully (not just thematically or contextually) related to each other and what are the consequences for longer retention intervals?
These questions highlight the potential contribution of communication among cognitive psychologists, cognitive neuroscientists, and educators. It is not just about asking what is applicable in the classroom right now, but about unveiling more aspects of how we think, learn and remember, and asking questions that guide future research.
References:
Reconsolidation and Prediction Error
1. Lee, J. L. (2009). Reconsolidation: maintaining memory relevance. Trends in neurosciences, 32(8), 413-420.
2. Hupbach, A., Gomez, R., Hardt, O., & Nadel, L. (2007). Reconsolidation of episodic memories: a subtle reminder triggers integration of new information. Learning & memory, 14(1),
3. Forcato, C., Argibay, P. F., Pedreira, M. E., & Maldonado, H. (2009). Human reconsolidation does not always occur when a memory is retrieved: the relevance of the reminder structure. Neurobiology of learning and memory, 91(1), 50-57.
4. Sinclair, A. H., & Barense, M. D. (2018). Surprise and destabilize: prediction error influences episodic memory reconsolidation. Learning & Memory, 25(8), 369-381.
Pre-testing
5. Richland, L. E., Kornell, N., & Kao, L. S. (2009). The pretesting effect: Do unsuccessful retrieval attempts enhance learning? Journal of Experimental Psychology: Applied, 15(3), 243.
6. Hausman, H., & Rhodes, M. G. (2018). When pretesting fails to enhance learning concepts from reading texts. Journal of Experimental Psychology: Applied, 24(3), 331.
Questions Formulation Technique
7. Rothstein, D., & Santana, L. (2011). Make just one change: Teach students to ask their own questions. Harvard Education Press.
Principles of instruction
8. Rosenshine, B. (2012). Principles of Instruction: Research-Based Strategies That All Teachers Should Know. American educator, 36(1), 12. [PDF]
Using Retrieval and Prediction in Instruction
9. Lang, J. M. (2016). Small teaching: Everyday lessons from the science of learning. John Wiley & Sons.
Curiosity and memory
10.. Gruber, M. J., Gelman, B. D., & Ranganath, C. (2014). States of curiosity modulate hippocampus-dependent learning via the dopaminergic circuit. Neuron, 84(2), 486-496.
Schema and novelty
11. Van Kesteren, M. T., Ruiter, D. J., Fernández, G., & Henson, R. N. (2012). How schema and novelty augment memory formation. Trends in neurosciences, 35(4), 211-219.
I have discussed Reconsolidation and Power of Prediction and their implication to teaching approaches in ResearchED conference in Philadelphia, October 2018, the slides are available here, and also in ResearchED Home, June 2020, slides are available here. The page will be updated soon.
December 2018