The Science Behind Mindstone

The Science Behind Mindstone

Joshua Wöhle - Published on Oct 10th, 2021

Mindstone uses scientifically proven methods to help users learn faster and remember more. In this article, we will review findings from the academic literature and explain how they have been applied in the development of our learning platform. We will also consider avenues for future development as the platform expands and new features are added.

Executive Summary

  • Spacing has been proven to dramatically improve recall and understanding. Mindstone uses a state-of-the-art algorithm for spaced repetition and allows you to create remember items while you are learning.

  • Interleaving: studying a range of related topics in parallel has been shown to improve learning. Mindstone uses interleaving in both the remember and highlights view features to help you learn faster and remember more.

  • Testing has been proven to help people learn better than just going over the material again. Mindstone’s remember feature takes advantage of this by explicitly testing users on what they’ve set out to learn.

  • Feedback is known to be beneficial both to the person on the receiving end and to the person giving it. By encouraging peer-to-peer feedback Mindstone supports faster feedback loops and, in turn, accelerated learning.

  • Modality matters: everyone learns better when they have the opportunity to engage with the text, images, audio, and video. Mindstone has been built from the ground up to amplify the learning experience across modalities.

Spacing: the not-so-secret weapon of science-based learning

The spacing effect is one of the most robustly demonstrated findings from the science of learning. It consists of two insights. Firstly, learning is more effective when there is a space between learning sessions, rather than studying content ‘all at once’ (as in cramming for an exam).(Bahrick, Bahrick, Bahrick & Bahrick, 1993) The second insight is that the effect is enhanced by expanding the gap over successive study sessions (Landauer and Bjork, 1978) so that a learner might return to a piece of content after (for example) two days, and then two weeks, and then two months later.

Although the majority of research into spacing has focused on memory, there is a growing body of research suggesting spacing can have a positive impact on other aspects of learning, including abstract reasoning and problem-solving (S Bird, 2010). Recent research has also sought to optimize the learning schedule, with algorithms such as SM2 (Wozniak, 1998) claiming a 95% improvement in retention and retrieval compared to control groups.

There are competing explanations for the efficacy of spacing as a learning strategy. An obvious and intuitively appealing explanation is that the ‘embedding’ of new knowledge is improved with each repetition, and expanding spacing simply tracks the ‘forgetting curve’(Ebbinghaus, 1885): each time content is revisited it takes longer to forget (Thios & D’Agostino, 1975). A more recent and counterintuitive explanation is that the increased difficulty of remembering on an expanding spacing-schedule actually drives better learning - that is, the expanding schedule works because it makes things harder to remember (Bjork, Kornell and Cheung, 2009, Rohrer & Taylor, 2006, Moulton et. Al, 2006).

Although there are many spaced-learning products on the market, Mindstone is the first to incorporate the creation of spaced repetition items into the flow of learning. While reading an article or reviewing highlights users can select items to remember and set questions to test themselves (see below for the impact of testing) which are automatically placed on a spaced schedule.

Although there are many spaced-learning products on the market, Mindstone is the first to incorporate the creation of spaced repetition items into the flow of learning. While reading an article or reviewing highlights users can select items to remember, and setting questions to test themselves (see below for the impact of testing) which are automatically placed on a spaced schedule.


Mindstone uses the SM2 algorithm to ensure ‘desirable difficulty’ (Bjork, 1994) in recall activities. Mindstone also supports a context-based workflow in which the learner simply selects something they wish to remember, and is tested using the surrounding context rather than a prompt question. This approach reduces the effort of initial prompt-creation, while increasing the ‘desirable difficulty’ of recall.

As the Mindstone platform continues to develop we will incorporate the remember feature into other modalities including audio and video sources (see learning modalities below) in order to broaden the range of contexts in which spaced learning can be applied. The learner’s engagement with the remember feature will also be incorporated into the meta-learning feedback the platform provides, helping learners to optimize their study-practices.

Interleaving: mixing things up to maximize learning

Although spacing is great for long-term learning, it runs up against real-world constraints: if your exam is in two weeks, a six-month spacing schedule is not going to work for you. Interleaving confers many of the benefits of spacing but in much tighter time-frames. The idea is simple: instead of focusing on a single area of study for a whole study-session, shuffle your learning between a range of (preferably related) topics.

The benefits of interleaving related topics are well-supported by the literature across a range of learning activities, from mathematics (Rohrer & Taylor, 2007) to motor-skills (Shea & Morgan, 1979) to translation (Richland, Bjork, & Finley, 2004). In addition to helping learners remember what they have learned, interleaving improves the ability to transfer skills and apply them in novel situations. The challenge of switching between different kinds of learned content is thought to allow learners to extract general rules and apply them in different settings. Moreover, the additional cognitive load of this kind of task-switching between related topics has been shown to improve inductive learning (generalizing from individual examples to make predictions and infer underlying rules)(Kornell & Bjork, 2008).

A second form of interleaving involves task-switching between wholly unrelated topics. Here there is much less scope for transfer and generalization, and it might be expected that the benefits of interleaving would disappear. But this is not entirely the case: although switching to unrelated topics is effortful, it seems to share some of the ‘desirable difficulty’ effects of spacing: the subjective difficulty of recall is increased, but recall is nevertheless better than with blocked study (where the learner studies a single topic for the whole session)(Alter, Oppenheimer, Epley & Eyre, 2020).

Mindstone’s remember feature supports this second kind of interleaving by default: information is presented that the learner has selected from multiple sources, increasing the challenge and improving recall. The highlights view feature supports both forms of interleaving: learners can choose whether to filter their highlights by tag, thus narrowing the view to related topics, or to view their highlights unfiltered. As with the remember feature, the ability to view highlights either with or without context helps the learner to make links to their prior learning.

In the future Mindstone will allow users to select for long-term learning using spacing or short-term learning (e.g. to prepare for exams) using interleaving, and will nudge learners towards long-term learning habits which will reduce the need for blocked-study.

Testing improves learning, even for things that weren’t on the test

As most students know, testing yourself is a great way to find out what you know and where you need to improve. Something that’s less widely known is that testing yourself is often better for your learning than additional time spent studying (Roediger & Karpicke, 2006). Furthermore, a study in 2006 found that testing can even improve recall of content that wasn’t in the test (Chan, McDermott & Roediger, 2006). It is thought that this is due to the learner retrieving information in order to rule it out, thus improving subsequent recall of that information despite it not being part of the test (Little & Bjork, 2010).

Mindstone’s remember feature explicitly tests learners rather than simply presenting them with information. First, users set themselves questions for which a selected piece of information from their reading is the answer. Next, in their daily sessions, they are prompted with the question and then shown the answer to check whether they got it wholly correct, partly correct, or incorrect. Finally, the SM2 algorithm updates the spacing schedule accordingly, ensuring that the learner will be presented with the question again at the best moment for ‘desirable difficulty’. The task of setting questions is itself a learning activity, as the learner must think about what question to set themselves and what they think they will be able to remember the next time they are presented with the question.

As the platform grows, Mindstone will allow for peer-to-peer testing and question-sharing, which will in turn enable the platform to provide suggested questions appropriate to the learner’s needs. We will also incorporate multi-modal testing, including video and audio tests.

Giving feedback can be as beneficial for learners as receiving it

The value of feedback for learning is a well-established finding in the scientific literature: feedback is more strongly and consistently related to achievement than any other teaching behaviour (Bellon <em>et al, </em>1991), and in adult learners feedback is strongly correlated with student retention and course completion (York, 2002). In web-based learning feedback is associated with a higher sense of agency and confidence, which in turn drives successful learning outcomes (Wang &amp; Wu, 2007). These findings have been confirmed by subsequent research, which found that in peer-to-peer feedback, the person giving the feedback experienced a similar improvement in their learning to the person receiving it (Li, Liu, &amp; Steckelberg, 2010).

Mindstone’s collaborative learning model encourages peer-to-peer feedback, with synchronous and asynchronous collaboration on comments as well as notes. Learners are able to ask and answer questions, and in so doing reflect upon their own learning. Shared workspaces provide a forum in which peers can contribute resources and work together in the extraction of insights and understanding, writing notes together or in teams. The notification system ensures that learners are apprised of new comments on their discussions, and are able to immediately respond to questions they have been asked.

In the near future Mindstone’s peer-to-peer feedback will be enhanced by the addition of an automated learning-behaviour feedback system. Learners will be guided towards more effective learning strategies by the app itself, and that guidance will include collaborative behaviours such as peer-to-peer feedback which are known to promote learning.

Learning modality makes a difference, but not in the way people expect

The ‘learning styles’ model suggests that there are different kinds of learners and that effective instruction requires some degree of personalization, for example providing kinaesthetic learning activities for people whose learning style is embodied learning, or images for visual learners (Dunn, 1984). Although this theory has very limited scientific support, it has persisted in the popular consciousness, and to this day many people think of themselves as visual, auditory, or kinaesthetic learners. During the development of Mindstone we conducted interviews with many students who described themselves in these terms.

There are two truths, one half-truth, and one outright falsehood in the learning-styles model. First the truths: people do have preferences about how they learn, and these preferences can have an impact on learning outcomes. For example, the belief that you are bad at one thing and good at another may influence your behaviour, and in time become a self-fulfilling prophecy: Secondly, being given a choice about how to learn can improve motivation and the learner’s sense of autonomy, which is positively correlated with successful learning outcomes (Shulruf, Hattie, & Dixon, 2008).

The half-truth is that people benefit from personalised learning which incorporates multiple modalities. The brain excels at integrating information from multiple sources and forms deeper connections between ideas when those ideas have been communicated through more than one medium. It is therefore useful to learn from multiple sources and in different modalities (Shulruf, Hattie, &amp; Dixon, 2008).

And the falsehood? Put simply, there is no good evidence for the existence of different learning styles. Everyone is a visual learner, and everyone is an auditory learner, and everyone learns from embodied kinaesthetic experiences (Coffield et al. 2004, Lilienfeld, Lynn, Ruscio &amp; Beyerstein, 2010, Pashler, McDaniel, Rohrer &amp; Bjork, 2008). This is not to diminish the importance of different sensory channels or to ignore the effect of impaired hearing or vision upon learning experiences. But for most people, learning styles represent a difference of preference rather than capacity.

In light of these findings, Mindstone has been built from the ground up as a multi-modality learning environment. We have built tools to allow for annotation of text including web-pages and PDFs, but also video players. By allowing users to make choices about source-material and modality, Mindstone supports the beneficial effects of learning-styles without pigeon-holing users into artificial learning-style boxes.

In the future Mindstone will continue to expand the range of supported learning modalities, including audio sources such as podcasts, as well as embodied learning experiences using augmented and virtual reality.


Learning has historically been centered around the institution rather than the individual. In the 21st Century that is finally beginning to change, and Mindstone intends to lead. Our platform is designed to be content-agnostic, multimodal, and learner-centered. Every decision about the platform has been informed by our reading of the science, and everything we do is with the learner in mind.

As the science of learning continues to progress, new findings will undoubtedly raise fresh challenges and opportunities for our platform. Some aspects of the learning experience, such as embodied kinaesthetic learning, are currently under-served by the platform, due to the limitations of current VR/AR technology. More distant prospects include direct access to the brain, through trans-cranial stimulation or even cortical interfaces of the sort envisioned by Neuralink.

No matter where the scientific research and technological innovation lead, we are committed to a human-centered future, where learning is valued both for its transformative power in society and for the rewards it brings to the individuals who engage with it. We are at the beginning of an exciting journey, whose end goal is the unlocking of human potential.