The history of computing shows how technological breakthroughs have characterized its different eras, re-shaped industries and transformed society, till unmet needs have joined hands with human ingenuity to push for the next frontier of innovation.
The Tabulating Systems era began in the 1890s with the invention of the tabulating machine – electro-mechanical systems that could summarize information stored in punch cards. This marked the beginning of large-scale semi-automated data-processing (e.g. the US census or social security administration); but new computations were unwieldy, often requiring extensive reconfiguration of wires carried out by services personnel.
The 1940s saw the beginning of the Programmable Systems era – digital computers that could automatically perform a variety of computations expressed (“programmed”) in a language they could understand. Massive mathematical calculations became possible at the blink of an eye and processes started getting codified through known ‘if-then’ scenarios. However, the real world is incredibly complex, making it impossible to anticipate and program all possible scenarios in advance.
Many decades of innovations later, we now stand at the threshold of the Cognitive Systems era. The emergence of data as the next natural resource, coupled with tremendous progress in speed, scale and affordability of computing have empowered us to start creating cognitive systems that can “learn at scale, reason with purpose and interact with humans naturally”. Such systems can now assist with or enact complex decision making even in uncertain scenarios with probabilistic outcomes – whether in the diagnosis of diseases, in self-driving cars, or a host of other sophisticated applications.
Compared to the history of computing, the history of electronic or eLearning is, of course, much more recent. However, like the eras of computing, the evolution of eLearning to cognitive eLearning has progressed through stages, each building on the earlier one through technical breakthroughs that have addressed unmet learning needs, and taken us closer to the end goal of personalized learning at-scale. This journey – which straddles the eras of programmable and cognitive systems – is helping shape what we understand and expect of personalized learning in the years to come.
So, if one were to look back at the journey of eLearning towards personalization, what would be its different stages? If we loosely define eLearning as the use of electronic technology to assist in the acquisition and development of knowledge, then the first modern computer-based training system is generally accepted to be PLATO (Programmed Logic for Automated Teaching Operations) which originated in the 1960s and supported programmatic design of new lesson modules. However, the real discontinuity in eLearning was unarguably triggered by the advent of the internet. So it would be more interesting to begin our review with that as the starting point. Figure 1 depicts this journey, which we explain in more detail below.
Figure 1. The Journey towards Personalization
So, if one were to look back at the journey of eLearning towards personalization, what would be it’s different stages? If we loosely define eLearning as the use of electronic technology to assist in the acquisition and development of knowledge, then the first modern computer-based training system is generally accepted to be PLATO (Programmed Logic for Automated Teaching Operations) which originated in the 1960s and supported programmatic design of new lesson modules. However, the real discontinuity in eLearning was unarguably triggered by the advent of the internet, so it would be more interesting to begin our review with that as the starting point. Figure 1 depicts this journey, which we explain in more detail below.
Anytime, Anywhere Learning: In the first wave of eLearning following the advent of the internet, the focus was on addressing the unmet need of learning access at-scale. The strategy was to use the reach of the internet to deliver learning content/courses widely (and at lower cost), especially to individuals who might not be well-served by the traditional educational systems. Naturally, distance learning courses got a new lease of energy, novel forms of blended learning emerged, and over time, new virtual eco-systems of content providers, consumers and accreditation agencies started to appear. With computers shrinking to fit in the palm of our hands and internet access costs reducing year over year, this has eventually led to anytime, anywhere learning – the first major technology-enabled breakthrough towards true democratization of learning.
However, while these advances allowed individuals to learn at their own pace, they did not provide any support to learn in their own way. In other words, the learning experience while digitized, was learner-agnostic and mostly passive. An individual had to go through pre-programmed learning pathways, or had to decide on her own which learning resources in which sequence was best suited to reach a learning goal – clearly a difficult task that mostly led to ad-hoc and ineffective learning. Thus, realization dawned that eLearning was good, but not a silver bullet.
But, with a global network of learners connected through technology, the seeds for the next era towards greater personalization driven by learner data, had already been planted.
Adaptive Learning: To personalize learning, we first need to understand how individuals learn. In the digital world, we may do so by:
• observing their actions while they are engaged in a learning activity e.g. in what sequence are they reviewing content or navigating learning objectives, how much time are they spending on an assessment activity etc.
• reviewing these actions in the context of what we know about the learners e.g. their profile in terms of existing skills, interests, goals etc. and our knowledge of learning content or objectives such as the inter-dependencies that exist between these
• reasoning about the efficacy of these actions in the light of outcomes achieved or skills demonstrated for example, during assessments
We can then use these insights to adapt the learning experience for each learner, so that learning pathways are dynamic, personalized and optimized towards desired goals, rather than pre-programmed and static. For example, based on an individual’s learning trajectory so far, which is the next best action for him/her in terms of learning objective to target, learning resource to review, or assessment item to attempt? These decisions may be based not only on the learning history of this individual, but also from what we have learnt by observing and analysing the learning behaviour and outcomes of many other learners who have gone through this experience before. At its core, this is what adaptive learning is about.
To support this form of learning, the entire digital learning experience needs to be instrumented in a big way. Every click (or even mouse hover) of the learner in every session can potentially be of value, so it can be captured, contextualized and then used for analysis. Aggregate this across thousands or millions of learners, and we will have incredibly large data sets to store, analyse and learn from. However, what could have been a technical impossibility in earlier times, is now a reality driven by advances in Big Data, data mining and machine learning. This has led to a significant interest in adaptive learning in recent years, with many education technology companies promising personalized learning experiences driven by adaptive pathways, crafted using their own variants of mining, learning and optimization algorithms on datasets they have collected and analysed on the cloud. While empirical evaluation of the effectiveness of these approaches is still in its early days, clearly such approaches have huge potential to personalize learning and improve learner engagement.
But, can a pathway, however uniquely crafted, meet all the personalized learning needs of a learner? If a student who has mastered pre-requisites struggles to understand a new concept even after being presented with the available resources, what should she do? Directing the student to either a pre-requisite or a follow-on concept will not be meaningful or effective in such a case. Clearly, adaptive learning has its limitations – it can decide what’s the next topic or content a student should review, based on intelligence gained from tons of data; but it does not provide much support for the actual learning itself. This is the most fundamental unmet need of eLearning, which will define the next era of Interactive Learning, driven by conversational tutoring systems.
Interactive Learning: When a student struggles with a concept, what she needs is a personal touch – someone who she can have a conversation with, ask questions to get doubts clarified, or who can figure out her gaps in understanding and help bridge the same. In short, the student needs a personal tutor. Ideally, a human tutor, but given that human tutors/instructors can only be available for limited periods of time (and for many students worldwide, even that is a privilege they do not enjoy), can we turn to technology to seek assistance?
Intelligent tutoring systems – perhaps the ultimate goal in the Holy Grail of personalized learning - have been an active area of research for decades, and a few domain-specific tutors exist. Typically, these have been carefully programmed to capture all known domain rules using which they can assist students in solving problems or provide scaffolding when needed. A few tutors are also conversational in nature, in that they can carry out some form of limited dialog with students in the context of learning. However, these tutors mostly do not generalize or scale well. Consequently, they have not enjoyed widespread adoption or commercial success so far.
Two technical advances are likely to change the status quo going forward.
The first is the tremendous progress that has been made in the areas of natural language processing, speech recognition and computer vision in recent years, with the availability of ever-growing open corpora of unstructured content across a variety of domains fuelling algorithmic improvements via research and development.
The second is the progress in developing multi-modal conversational systems – whether through text, speech, images, gestures or other modalities – where virtual agents are being trained to not only recognize the content of a user input, but also its tonality or emotion, and respond appropriately.
The combination of these two advances suggests that we may be entering an age when intelligent tutoring systems that effectively augment the capacity of expert human tutors may become a reality. These intelligent tutors - with their ability to ingest, understand and reason about learning content, and then using this knowledge to engage in a meaningful dialog with learners to address their misconceptions and enhance mastery – can elevate personalization from unique pathways to uniquely tailored learning experiences.
In summary, what then is personalized learning? Numerous experts have answered this question in a variety of ways over the years, and perhaps there is not much harm in adding one more perspective shaped by the evolution of eLearning. Personalized learning happens at the confluence of anytime/anywhere learning, adaptive learning, and interactive learning. Together, they allow an individual to learn at her own pace, along pathways carefully crafted based on her needs and with uniquely tailored learning and pedagogical experiences along the way.
The author is STSM & Senior Manager, Cognitive Education and Interactions, IBM India Research.
Disclaimer: This article is published as part of the IDG Contributor Network. The views expressed in this article are solely those of the contributing authors and not of IDG Media and its editor(s).