We often get the question: how does a modern learning platform actually use artificial intelligence to improve the user experience? In the current landscape, AI is no longer a futuristic buzzword; it is a practical and powerful tool that is fundamentally reshaping how people learn, develop skills, and advance in their careers. The days of a one-size-fits-all digital library of courses are over. Learners today expect and deserve an experience that is as dynamic, relevant, and personalized as the consumer technology they use every day. This is where AI becomes not just a feature, but the core engine of a truly effective learning ecosystem. At our organization, we have incorporated artificial intelligence into our learning platform in meaningful and measurable ways. We believe in a culture of continuous experimentation, testing, and innovation, which allows us to learn and then share those insights with our community. We are proud to be at the forefront of providing AI-driven, transformative learning experiences. Our efforts in this space have even been recognized by external bodies for excellence in artificial intelligence. This commitment is not about technology for its own sake, but about solving the real, practical challenges that learners and organizations face every day.
Why AI is Essential for Effective Learning
The traditional approach to corporate learning has a fundamental flaw: it scales by treating everyone the same. It relies on static learning paths and vast, uncurated libraries where learners are expected to find what they need on their own. This results in low engagement, wasted time, and a disconnect between learning and on-the-job performance. AI is the key to solving this. It has the power to shift the paradigm from a “content-centric” model to a “learner-centric” model. Instead of forcing a learner to navigate a sea of information, AI can act as a personal guide. It can understand who a learner is, what they know, what their goals are, and what they are interested in. It can then sift through tens of thousands of learning assets—videos, courses, books, and labs—to find the exact piece of content that will help that learner move from Point A to Point B in the most efficient and effective way possible. This creates a learning experience that feels relevant, supportive, and truly individual.
Our Four Pillars of AI Strategy
We use artificial intelligence to accomplish four primary goals that benefit not only the users of our platform but also the users who access our content through integrated learning management systems. These four pillars are the foundation of our AI-driven approach. First, we aim to personalize the learning experience for each user, helping them progress efficiently. Second, we work to improve search and discovery, making it easier for learners to find precisely what they need, when they need it. The third pillar is to link skills, roles, and learning together, which allows us to guide learners along tangible career paths. And fourth, we use AI to help generate new content and automate curation. This enables users to assess their skills, build confidence, and receive relevant learning paths for new and trending topics. These four pillars work in concert to create a smarter, more responsive, and more valuable learning ecosystem for everyone.
Pillar One: Personalizing the Learner Journey
Personalization is the first and most critical pillar. AI is the engine that allows us to move beyond a generic, one-size-fits-all homepage. We use it to personalize the learning for each user based on a rich set of data points, including their stated interests, their job role, their search behavior, their assessment results, and their past learning activity. This allows the platform to be proactive, not just reactive. Instead of presenting every learner with the same “popular” courses, the platform can surface a highly relevant piece of content that is specific to that user’s needs. In addition to personalized learning paths that are generated from our skills assessment tools and stated interests, we use AI in several key ways to make the entire experience feel like it was built for an audience of one. We will explore these methods in greater detail in the second part of this series.
Pillar Two: Revolutionizing Search and Discovery
A vast library of content is useless if learners cannot find what they need. Our second pillar focuses on using AI to solve the “finding” problem. Traditional search engines on learning platforms were notoriously simple, relying on exact keyword matches. This meant that if a user did not know the precise title of a course, they might never find it. Today, we have adopted the advanced AI models used by major search engine and technology organizations to power a much more intelligent search. Our platform’s search function understands semantics and intent, not just keywords. It can deliver more targeted and relevant results, even when a learner uses multi-word phrases or ambiguous terms. We also use AI to solve the “last mile” problem of discovery: ensuring our content is easily found even when it is delivered through a third-party learning management system (LMS) that may have a less sophisticated search function. This ensures learners can get to the right content quickly, reducing friction and frustration.
Pillar Three: Connecting Learning to Careers
Learning is not just about acquiring knowledge; it is about building a career. Our third pillar involves using AI to connect the dots between skills, job roles, and the specific learning content that bridges the gap. We maintain a proprietary, complex skills taxonomy, and all of our content and assessments are meticulously mapped to it. This skills framework is the “brain” that allows the system to understand the relationships between different competencies. AI is being evaluated to help manage this complex “sea of skills.” This includes normalizing and de-duping skills from different sources, categorizing skills into meaningful clusters, and even auto-curating role-based skill profiles from job descriptions. The end goal is to provide a clear, guided path for a learner, showing them the skills they need for their current role and helping them build the skills they need for their next one.
Pillar Four: Generating and Curating Content
The fourth pillar focuses on using AI to augment our content creation and curation processes. The world is changing so quickly that new topics and skills emerge almost daily. AI models are being used to generate new types of content and images, and perhaps most importantly, to curate content to meet specific customer needs. For example, AI can help us build a brand new learning path on a trending topic faster than ever before by analyzing consumption data and expert-curated content. We are also experimenting with AI to generate a deep pool of assessment questions, which are necessary for allowing learners to test their skills multiple times. This allows us to scale our ability to help users assess their skills as they build confidence. This AI-assisted generation and curation does not replace our human experts; it enhances their capabilities, allowing us to be more agile and responsive to the needs of the learner.
Moving Beyond the One-Size-Fits-All Model
For decades, digital learning was defined by a static, library-based model. Every user who logged in saw the same homepage, the same categories, and the same “featured” courses. This one-size-fits-all approach is deeply inefficient. A new manager, a senior software engineer, and a marketing specialist all have vastly different learning needs, yet they were all presented with the same interface. This is the fundamental problem that AI-driven personalization aims to solve. It allows a learning platform to treat every user as an individual, with unique goals, interests, and existing knowledge. Our platform uses AI to transform this static library into a dynamic, personalized experience. The goal is to ensure that from the moment a user logs in, the content they see is immediately relevant to them. This personalization is based on a wide range of factors, including their declared profile interests, their job role, their search behavior, their learning activity, and the results from any skills assessments they have taken. All of these data points create a unique “learning fingerprint” for each user, which our AI models can then use to serve up the perfect piece of content at the perfect time.
Personalization Through Collaborative Filtering
One of the primary ways we personalize the learning experience is by recommending content based on a user’s recent activity. This is achieved through a sophisticated AI model known as collaborative filtering. This approach is modeled after the recommendation engines used by major e-commerce retailers, which you might have seen in sections like “people also bought.” Our platform’s engine, however, is tuned for learning, not shopping. It works by learning a user’s behavior, such as the courses they have taken, the videos they have watched, and the topics they have searched for. The system then finds other, similar users in our global user base. It identifies learners who have a similar role, similar interests, and a similar learning history. The AI then looks at the content that these “most similar” users consumed and found valuable. It recommends this content to the original user, providing a powerful, data-driven suggestion for what they might want to learn next. This approach is incredibly effective because it leverages the collective intelligence of millions of learners to surface relevant content that a user might never have found on their own.
The Impact of Personalized Recommendations
The results from this personalization strategy are both clear and compelling. We have found that these AI-driven recommendations are used most often by our learners. More importantly, we have measured that users who access these personalized recommendations spend, on average, 59% more time learning on the platform. This is a significant metric. It indicates that the recommendations are not just being clicked, but that they are leading to deeper, more sustained engagement with the content. This signals that the recommendations are truly valuable and relevant to the learner. This personalization is not just on the homepage. It is also used to power our automated re-engagement notifications. If a user has been inactive, the system can send a highly personalized email or notification that, instead of a generic “come back and learn” message, can suggest a specific, highly-relevant piece of content based on their past activity. This dramatically increases the likelihood that the user will re-engage with the platform in a meaningful way, helping them build a consistent learning habit.
The Next Evolution: Sequential Consumption Patterns
We are continually experimenting and testing to improve our personalization capabilities. Our AI specialists have recently been testing and training a new collaborative filtering model that adds another layer of sophistication. This new model does not just recommend content based on what other people have consumed, but also on the sequences in which they consumed it. It analyzes the common paths and sequences inherent in the way learners have successfully built a new skill. For example, the model might learn that for the topic of “Data Science,” learners who first watch a video on “Python Basics,” then take a course on “Data Wrangling,” and then read a book on “Statistical Models” are far more successful than those who consume the content in a different order. The AI can then recommend this specific, proven sequence to a new learner who expresses interest in data science. This moves personalization from simple “topic” recommendation to sophisticated “learning path” recommendation.
AI-Powered Learner Support: The Chatbot
Personalization is not just about the content we recommend; it is also about the support we provide. Our learning platform has a chatbot, which is currently deployed to select customers, that provides immediate, AI-driven Level 1 customer support. This chatbot is trained to understand and answer the most common questions that learners have, such as “How do I reset my password?” “How do I find my learning transcript?” or “How do I get a certificate of completion?” This provides an instant, 24/7 support channel for learners, allowing them to get answers to simple questions without having to file a support ticket or wait for a human response. This frees up our human support team to focus on more complex and high-value learner issues. This chatbot will be made available to all of our customers in the near future, further personalizing the support experience.
The Data Driving AI Support
The effectiveness of this chatbot is directly tied to its AI model. It uses a sophisticated natural language processing engine to understand the user’s question, even if it is phrased in an unusual way. The AI can parse the request, identify the user’s intent, and provide the correct answer from its knowledge base. The data we have collected from the customers who are currently using this feature is very promising. We have found that for these participating customers, 45% of all Level 1 common questions were successfully answered by the chatbot without any human intervention. This is a high success rate that demonstrates the power of AI to provide immediate, scalable, and personalized support. It shows that AI can be a powerful partner in handling the logistical side of learning, reducing friction for the user and letting them focus on what matters most: the learning itself.
The Future of Personalized Learning Paths
The ultimate goal of all these AI efforts is to create a truly individualized learning journey for every user. In the future, this will go beyond just recommendations and chatbots. We are moving toward a model where AI can help a user build a comprehensive, long-term learning plan from the ground up. By combining data from their skills assessments, their stated career goals, and their on-the-job performance, the AI will be able to suggest a complete path of courses, videos, books, and hands-on practice labs. This path will be dynamic, adjusting in real-time as the user’s needs change or as they master new skills. This is the true promise of AI in learning: an experience that is not just personalized, but predictive, adaptive, and deeply supportive of each user’s unique journey from Point A to Point B, whatever that journey may be.
The ‘Finding’ Problem in Corporate Learning
One of the most persistent challenges in corporate learning is what we call the “finding” problem. Modern learning libraries are vast, often containing tens of thousands of individual learning assets, including courses, videos, books, and audio summaries. While this breadth of content is a strength, it can also be a significant weakness. When a learner is faced with a library of this size, the simple act of finding the specific piece of content they need can be overwhelming. It is like trying to find a specific book in a massive library without a card catalog or a librarian to guide you. This search-and-discovery challenge leads to frustration, and frustrated learners stop learning. They may try to find something, fail, and then give up, assuming the content they need does not exist, even when it does. Traditional search engines on these platforms were a major part of the problem. They typically relied on simple, outdated keyword matching. If a user’s search term did not perfectly match the title or metadata of a course, the search would fail, returning no relevant results.
From Keyword Matching to Semantic Understanding
To solve this problem, we have moved far beyond simple keyword matching. We have embraced the revolutionary advancements in artificial intelligence, specifically in the field of natural language processing. The major search engine and social media organizations have made massive investments in AI models that can understand the meaning and intent behind a search query, not just the keywords. This is the difference between “keyword search” and “semantic search.” We have adopted these models and deployed them within our learning platform to deliver a world-class search experience. Our platform’s search engine now incorporates advanced, transformer-based language models. These models are pre-trained on a massive corpus of text, allowing them to understand context, nuance, and the relationship between words. This means a learner does not have to know the “magic” keyword. They can type a query in their own natural language, using multi-word phrases to find something specific, and the AI will understand what they are looking for.
The Impact of Semantic Search on Learners
This shift to semantic search has a tangible, positive impact on the learner experience. It delivers more targeted, more relevant, and more useful results. For example, let’s consider the source’s example of a search for “micro-services” versus “micro-behaviors.” A traditional keyword-based search might get confused by the overlapping term “micro” and return a jumbled, irrelevant mix of results for both topics. A semantic search engine, however, understands that “micro-services” is a specific software architecture concept and that “micro-behaviors” is a leadership development concept. It can distinguish the two and provide highly relevant results for the learner’s specific query. This AI is also less sensitive to common human errors. It can understand that a learner searching for “url” is looking for the same thing as a learner searching for “URL.” The older, case-sensitive models would have treated these as different queries. By handling these nuances, the AI-powered search reduces friction and makes the learner feel like the system is working with them, not against them, to help them find what they need.
The Next Generation: Continual Search Improvement
Our commitment to improving search and discovery is ongoing. Our search and recommendations team is currently in the process of upgrading our platform’s search capabilities from one advanced AI model to an even newer, more sophisticated one. We are moving from a well-known model to a different model that was trained in a unique way by another major technology organization, and our internal tests have shown its superiority. Our test results showed that this newer model produces even more relevant results, particularly for shorter search phrases, which are very common learner behavior. It was also found to be more accurate in handling overlapping terms. This process of continually testing, evaluating, and deploying the latest AI models is a core part of our engineering culture. It ensures that our learners always have access to the best possible search experience, an experience that improves and evolves as the technology does.
The “Last Mile” Problem: Improving Search in an LMS
Our AI-driven search capabilities are a powerful feature within our own platform. However, many of our clients access our vast content library through their own, third-party learning management system (LMS). This creates a significant challenge. When our content is delivered in another company’s LMS, we cannot use our own advanced search engine. The learner is forced to rely on the search capabilities built into that LMS, which are often the older, less effective keyword-matching type. This can lead to a poor discovery experience, where learners cannot find our content. To solve this “last mile” problem, we use AI in a clever and proactive way. We know that these third-party systems often rely heavily on just two fields of data: the content’s title and its description. To improve the relevancy of our content in their search results, we have used AI to automatically create new, highly detailed descriptions for our content.
Using Generative AI to Create Richer Descriptions
We have a library of over 40,000 videos. Manually writing a unique, keyword-rich, detailed description for every single one of these videos would be an impossible task. Instead, we have used a large generative text model to do it for us. This AI model was trained to automatically generate detailed descriptions for each of our videos to improve the search results within a third-party LMS. We experimented with this process and learned a valuable lesson. The quality and yield of these AI-generated descriptions were strongest when we fed the model two sources of input: the full video transcript and the high-level course objectives. The transcript provided the “what” of the video, and the objectives provided the “why.” By combining both, the AI was able to generate a rich, relevant, and comprehensive description that dramatically improves the content’s “findability” in a simpler search engine. Our team will next experiment with even more modern generative AI models to see if we can improve this process even further.
Quality Control for Human-Authored Content
The same problem of “findability” also exists for our course-level content. As mentioned, learning management systems typically rely on course titles and course descriptions to drive their search results. Our content is authored by a wide range of brilliant, human subject matter experts, who create the course descriptions as part of their authoring process. Because these are written by many different people, the quality and completeness of these descriptions can vary. A great course on a critical topic might be hard to find simply because its description was written too briefly. Occasionally, our teams would trace reports of less relevant search and recommendation results back to the quality of a course’s description. To solve this at scale, we turned to AI. A generative model was fine-tuned for a new purpose: not to write descriptions, but to rank the quality of existing, human-written descriptions.
How AI Becomes a Quality Assurance Partner
This fine-tuned AI model was trained to read a course description and rank its quality on a predefined scale. This allowed us to automatically and programmatically check the quality of tens of thousands of course descriptions in a way that would be impossible for a human team. The results of this quality check were illuminating. The AI identified that about 10% of our course descriptions were of inadequate quality to perform well in a third-party search engine. This AI did not just identify a problem; it also gave us the tool to solve it. With the 10% of low-quality descriptions identified, we can now use the other AI model—the generative description writer described previously—to remediate them. This system will be used to automatically replace the lower-quality, human-written descriptions with new, high-quality, AI-generated descriptions. This is a perfect example of how different AI models can work together to solve a complex problem and ensure a better learner experience.
The Shift to a Skills-Based Economy
The modern economy is undergoing a fundamental shift. The old model, where an employee’s career was defined by their job title and their degree, is fading. It is being replaced by a more dynamic, fluid, and skills-based model. In this new world, what matters is not your title, but the skills you possess. This “skills-based economy” requires a new approach to learning and talent development. Organizations are no longer just hiring for a role; they are hiring for a set of skills. And employees are no longer just “doing a job”; they are continuously acquiring new skills to stay relevant and advance their careers. This shift presents both a massive opportunity and a massive challenge. The opportunity is a more agile, mobile, and equitable workforce, where people are valued for their proven capabilities. The challenge, however, is one of complexity. How does an organization define, measure, and manage the thousands of skills needed to run its business? And how does a learner navigate this complex “sea of skills” to build a meaningful career path? This is where artificial intelligence becomes a critical partner for organizational design and individual career guidance.
The Challenge of the “Sea-of-Skills”
The “sea of skills” is a real and difficult problem for almost every large organization. A typical company may have dozens of different systems—HR systems, learning platforms, recruiting tools—each with its own “skills dictionary.” One system may list “Project Management,” while another lists “Project Leadership” and a third lists “PMP.” Are these three different skills, or are they all the same? This lack of a common language creates chaos. It makes it impossible to get a clear picture of the skills the company has versus the skills it needs. For learners, the problem is just as confusing. They are told they need to “build new skills,” but they are given no clear guidance. They may not know what skills are required for their own role, let alone what skills they would need to get a promotion or move to a different department. Without a clear map, learners are left to guess, which leads to wasted time and a disconnect between learning efforts and business goals.
Our Proprietary Skills Taxonomy
To provide this clear map, we have invested heavily in building a proprietary skills taxonomy. This taxonomy is, in effect, the “brain” or the central nervous system of our platform. It is a complex, hierarchical framework that defines and organizes thousands of skills, from broad competencies like “Leadership” down to very specific, technical skills like “Python Data Structures.” All of our learning content—every course, every video, every book, and every assessment—is meticulously mapped to this central skills framework. This mapping is what allows our platform to function at an advanced level. When a user takes a skills assessment, the system can pinpoint their exact skill gaps within this taxonomy. When a user consumes a piece of content, the system knows exactly which skills they are building. This common language of skills is what enables true personalization and, more importantly, the ability to guide a learner from one role to another.
AI as a Skills Management Partner
This skills taxonomy is the foundation, but it is not a static document. New skills emerge every day. Furthermore, our clients often have their own internal skills taxonomies, or they may use industry-wide frameworks (such as those for government or technology sectors). This is where AI is being evaluated as a powerful skills management partner. We are currently evaluating generative AI and large language models, in conjunction with other services, to help manage the immense complexity of our stock and custom skills dictionaries. The goal is to use AI to solve the “sea-of-skills” problem at scale. The specific areas we are evaluating are critical to making a skills-based organization a practical reality. AI can help us automate the process of normalizing, de-duping, and mapping roles and skills from different, conflicting dictionaries. It can take two separate lists of skills and find the overlaps, harmonize the language, and create one single, clean “source of truth.”
From Job Descriptions to Career Profiles
Perhaps the most exciting application we are evaluating is the use of AI to automatically generate custom taxonomies and role-skill profiles from job descriptions. Today, this is an intensely manual process. A company might spend months with consultants to define the skills needed for every role. Generative AI has the potential to automate and accelerate this significantly. An AI model can be trained to read a company’s entire library of job descriptions. From these descriptions, it can extract the key skills required for each role, and then “auto-curate” a skills profile for that role. For example, it could read 100 “Senior Software Engineer” job postings and conclude that this role requires an expert level in “Java,” a proficient level in “Cloud Computing,” and a foundational level in “Project Management.” This creates an instant, data-driven profile that can be used for hiring, development, and career pathing.
Categorizing Skills into Meaningful Clusters
Another area of evaluation for AI is its ability to categorize the vast sea of skills into meaningful, human-friendly clusters. A flat list of 5,000 skills is not useful to anyone. An AI, however, can analyze the relationships between these skills. It can see, for example, that skills like “Active Listening,” “Giving Feedback,” and “Empathy” are all highly related and frequently appear together in leadership roles. The AI can then “cluster” these together, perhaps under a broader competency like “Interpersonal Communication.” This clustering makes the skills framework far more intuitive and navigable for a learner. Instead of being faced with an overwhelming list, they can explore these meaningful clusters. This helps them discover new, related skills they may not have known about, and it helps them understand how different skills build upon one another, creating a more holistic and less fragmented learning experience.
The Future: Guided Career Paths
This all leads to the ultimate goal of our third pillar: using AI to provide guided, personalized career paths. By linking skills, roles, and learning content together, we can move beyond simple recommendations. The system of the future will be able to have a “career conversation” with a learner. A user will be able to say, “I am currently a Data Analyst, but I want to become a Data Scientist.” The AI, armed with the skill profiles for both roles, can instantly perform a “gap analysis.” It will be able to show the learner the skills they already have from their current role and the new skills they need to acquire for their desired role. Then, because all of our content is mapped to this same skills framework, the AI can automatically curate and recommend a specific, sequential learning journey, complete with courses, assessments, and practice labs, to help that learner bridge the gap and achieve their career goal.
Mapping Content to Custom Taxonomies
This AI-driven skills management is also a critical service for our clients. Many large organizations have already invested heavily in creating their own internal skills frameworks, and they need our learning content to align with their language. In the past, this required a massive, manual mapping project that could take months. AI is being evaluated to automate this. A client could provide us with their skills taxonomy, and an AI model could “read” our content and their taxonomy and automatically create the mappings. This would allow a client to have our platform and content, but have it speak their unique skills language. This deep, custom integration is what businesses are asking for, and AI is the only technology that can deliver it at scale.
The Dual Challenge of Content: Scale and Relevance
In the world of learning and development, content is king. But organizations face a dual challenge: they need a massive scale of content to cover the thousands of skills their employees need, but they also need that content to be highly relevant to their specific, unique business context. No off-the-shelf library, no matter how large, can perfectly meet every need of every organization. This is where the fourth pillar of our AI strategy—generating new content and curating existing content—becomes a critical value. We are using artificial intelligence, particularly generative models, to augment the work of our human experts. This is not about replacing human instructional designers or subject matter experts, but about giving them “superpowers.” AI can help us scale the creation of certain types of content, like assessment questions or images, and it can help us curate our existing, expert-created content in new and powerful ways, allowing us to deliver a more customized and relevant experience for our clients.
AI-Powered Curation for Custom Needs
The first and most practical application is in customer-specific curation. Every organization has its own culture, its own policies, and its own specific learning needs. A client might come to us and say, “We love your leadership development content, but our company has a very specific model for giving feedback, and we need to ensure our learners are not exposed to conflicting information.” In the past, this would require a human curator to manually review hundreds of courses and videos. Today, we can use a fine-tuned generative AI model to do this at scale. The customer can define their specific needs, which may include aligning content to their own internal taxonomy or, just as importantly, identifying content for exclusion from their programs. The AI model can then scan our entire library, ranking content based on how well it aligns with the customer’s needs. The human customer then just needs to review the AI-ranked content before deploying it, turning a months-long project into a days-long review.
Generating Assessment Questions at Scale
Another key area of focus is content generation. Our platform currently assesses skills and knowledge using many different methods, including our comprehensive skills benchmarks, end-of-course assessments, final exams for our structured career journeys, and mobile flashcards for learning reinforcement. To be effective, these assessment features require a very deep and very broad pool of high-quality assessment questions. This is necessary to enable each user to assess and reassess their skills without seeing the same questions over and over. Manually writing, testing, and validating tens of thousands of multiple-choice, in-depth questions is a significant bottleneck. To solve this, we are using AI to generate new assessment questions. Our teams are using fine-tuned generative text models to create a “first draft” of questions, which are then reviewed and refined by our human subject matter experts. This dramatically accelerates the development pipeline and ensures our assessment engines are robust.
The Human-in-the-Loop: Lessons from Question Generation
This experiment in AI-generated assessment questions has also provided one of our most valuable lessons. The quality of AI-generated content is highly variable and requires a “human-in-the-loop” to be effective. Our initial experiments with a fine-tuned generative model produced mediocre results. We found that only about 50% of the AI-generated questions were usable as-is. Another 40% needed significant edits from an expert, and 10% were simply incorrect or unusable and had to be discarded. We also tested a newer, zero-shot generative model, which is a model that requires no special training. The results from this model were even worse, yielding only about 21% acceptable questions. This taught us that, for a high-stakes, nuanced task like creating a valid assessment, AI is not a “magic bullet.” We are now exploring different techniques, such as using the AI to reflect on and grade its own output, which is showing promise in improving the yield. But for now, the human expert remains the most critical part of the quality control process.
The Future of Automated Learning Path Curation
In addition to specific content, we are also experimenting with using AI to curate entire learning paths. Currently, the creation of our structured, multi-modal learning paths is done by expert curators in each domain. These instructional designers base their work on sound instructional design principles, sequencing content in a way that builds a learner’s skills from foundational to advanced. This human expertise is invaluable and creates a very high-quality product. However, we have also collected a tremendous amount of data about how users actually learn on our platform. We can see the paths they take, the content they use, and the skills they build. This data can be used to complement our human experts. We are experimenting with AI models, including both collaborative filtering and generative models, to automatically curate new learning paths for new or trending skills. This AI could, for example, analyze a new technology trend, identify all the relevant assets in our library, and sequence them based on the consumption patterns of our most advanced users, creating a “first draft” of a learning path in minutes.
Using AI to Create Engaging Visuals
Learning is not just a cognitive process; it is also an emotional and visual one. An engaging, eye-catching thumbnail image or a relevant blog post graphic can make the difference between a learner scrolling past a piece of content and clicking on it. A good image can energize the learner and make the content feel more modern and appealing. However, creating or sourcing unique, high-quality images for thousands of pieces of content is a significant design and licensing challenge. To address this, we have begun experimenting with AI-powered image generation models. We used one of these text-to-image models to create the image for the very blog post this series is based on. We plan to continue experimenting with this technology to create unique, eye-catching thumbnail images for our courses. These AI-generated images would then appear in our platform and would also travel with the content when it is used in third-party learning management systems, creating a more engaging and visually consistent experience for the learner.
The Ethics and Quality of AI-Generated Content
As we explore these powerful new capabilities, we are also being extremely thoughtful and cautious. The rise of generative AI brings with it important questions about quality, accuracy, and bias. This is why our approach is one of “augmentation,” not “replacement.” Our human subject matter experts and instructional designers are, and will remain, at the center of our content strategy. Our experience with generating assessment questions is the perfect example of our philosophy. We did not simply turn on the AI and flood our platform with unverified questions. We tested, we measured, we learned, and we implemented a process that uses the AI to boost the productivity of our human experts, who are the ultimate arbiters of quality. This “human-in-the-loop” model ensures that we can leverage the scale of AI without sacrificing the quality that our learners depend on.
Our Culture of AI Innovation
Developing a world-class, AI-driven learning platform is not a single project. It is a continuous, iterative process that requires a deep-seated culture of innovation. It involves constant experimentation, rigorous testing, and a willingness to learn from both successes and failures. Our approach is not just to acquire and deploy new technologies, but to be an active participant in understanding how these powerful tools can be responsibly and effectively applied to the unique challenges of corporate learning. This means we are always in the lab, testing, learning, and sharing those insights. This culture of innovation extends to our own internal processes. We believe in “drinking our own champagne,” and our artificial intelligence team is already using the latest generative AI tools internally. This internal use serves two purposes: it improves our own productivity, and it gives us first-hand experience with the technology we are implementing for our learners. This hands-on, practical approach is a core part of our philosophy.
AI for Internal Productivity: A Case Study
A concrete example of this is how our own AI team has been utilizing large generative text models to improve their own productivity and accelerate our AI efforts. The team has been using these models to generate code to implement new features for the learning platform. For example, when building enhancements to the AI-based collaborative filtering model or improving the platform’s search features, such as auto-suggest and type-ahead, these AI tools can write boilerplate code or suggest solutions to complex programming problems. Our engineering teams have found this to be quite effective, allowing them to focus on the more unique, high-level architecture rather than the commodity code. This accelerates our development cycles and allows us to deliver new features and improvements to our learners more quickly. This internal use case proves the power of AI as a productivity partner, a benefit we hope to pass on to our learners.
The Critical Importance of Copyright and Ethics
This internal experimentation also serves as a critical proving ground for ethical AI use. As we use these tools, we are exercising great care and developing strict policies to ensure we are not violating copyrighted code or introducing insecure or biased code into our systems. The current generation of large language models was trained on vast amounts of data from the public internet, which includes copyrighted material. We are actively engaged in the legal and ethical discussions surrounding this technology. We are building processes to ensure that any AI-generated code is properly vetted, that its sources are understood, and that we are respecting the intellectual property of others. This careful, deliberate approach is a non-negotiable part of our strategy. We believe that AI can only be successful if it is built on a foundation of trust, and that trust must be earned through responsible and ethical implementation.
A Look Back: The Lessons We’ve Learned
This entire journey of integrating AI has been one of continuous learning. We have learned that some applications of AI provide immediate and clear value. For example, our data shows that personalized, collaborative-filtering-based recommendations are highly valuable to users, evidenced by the 59% increase in learning time they generate. We have also learned that AI-powered chatbots are remarkably effective at handling simple support tasks, successfully answering 45% of common Level 1 questions and reducing friction for learners. We have also learned that AI is not a magic wand. Our experiments with generating assessment questions taught us that the quality from some models is mediocre at best, with only 50% of questions being usable. This highlights the critical, non-negotiable need for a “human-in-the-loop.” We also learned that the quality of AI-generated content can be dramatically improved by providing the model with better, more contextual inputs, as we saw when we combined video transcripts with course objectives to create high-quality descriptions.
The Power of Iteration and Experimentation
These lessons all point to a central truth: progress in AI is built on a foundation of iteration and experimentation. You must be willing to try things, measure the results, and learn. Our teams are constantly testing new models, new inputs, and new techniques to see what works. We are not afraid to find that an experiment failed, because that failure provides a valuable lesson. The discovery that a certain generative model produced only a 21% yield on acceptable questions was not a failure; it was a critical data point that saved us from deploying a low-quality feature and refocused our efforts on more promising techniques. This iterative approach is what allows us to move forward with confidence. We can separate the real-world, practical applications of AI from the over-hyped, futuristic promises. This focus on practical, measurable value is what has allowed us to be at the forefront of this field, earning recognition for our AI-driven learning experiences.
The Human-AI Partnership in Learning
The future of learning, as we see it, is a deep and collaborative partnership between human intelligence and artificial intelligence. Our strategy is not to replace the human element, but to augment it. AI is at its best when it is doing what humans cannot: sifting through petabytes of data, finding subtle patterns, and performing repetitive tasks at scale. This frees up human learners, human instructors, and human curators to do what they do best: be creative, think critically, ask insightful questions, and connect with each other on a human level. Our platform is the embodiment of this partnership. It uses AI to handle the complex, logistical “plumbing” of learning—personalization, search, and skills mapping. This creates a frictionless, supportive environment where the human learner is free to focus on the hard work of learning, and where our human experts are free to focus on creating the highest-squality content.
The Future Vision for AI-Driven Learning
The landscape of education is undergoing a profound transformation, driven by advances in artificial intelligence and machine learning technologies. As we stand at the threshold of this new era, we envision a future where learning becomes more than just consuming information. It becomes an intelligent, adaptive, and deeply personalized journey that responds to each individual’s unique needs, pace, and learning style. This vision represents not merely an incremental improvement over existing systems, but a fundamental reimagining of how knowledge is acquired, processed, and applied in the modern world.
A New Paradigm of Personalized Learning
Traditional educational models have long struggled with a fundamental challenge: how to effectively teach diverse groups of learners who possess different backgrounds, abilities, and learning preferences. The conventional one-size-fits-all approach has proven inadequate in addressing the nuanced needs of individual students. While personalization has been a goal in education for decades, the practical limitations of human instructors managing large groups of students have made true individualization nearly impossible to achieve at scale.
The integration of artificial intelligence into learning platforms promises to break through these limitations. By leveraging sophisticated algorithms and data analysis, AI-driven systems can track and understand each learner’s progress with unprecedented precision. These systems can identify patterns in how individuals absorb information, recognize when a concept hasn’t been fully grasped, and adjust the teaching approach accordingly. This level of personalization goes far beyond simply recommending the next lesson based on quiz scores. It involves creating a dynamic, responsive learning environment that evolves alongside the learner.
Beyond Multiple-Choice: Authentic Assessment Methods
One of the most significant limitations of traditional online learning has been the reliance on standardized testing methods, particularly multiple-choice questions. While these assessments offer convenience and ease of grading, they fall short in measuring true understanding, creativity, and practical application of knowledge. The ability to select the correct answer from a list of options does not necessarily indicate that a learner can apply that knowledge in real-world situations or demonstrate genuine mastery of a skill.
The future of AI-driven learning envisions assessment methods that are far more sophisticated and authentic. Imagine a system capable of analyzing a complete project submission, whether it’s a piece of writing, a coding project, a design portfolio, or a business case study. Through advanced natural language processing and computer vision technologies, AI can evaluate not just the correctness of answers, but the quality of reasoning, the creativity of solutions, and the depth of understanding demonstrated in the work.
This project-based assessment approach mirrors how skills are actually utilized in professional environments. Rather than testing isolated facts or concepts, learners are evaluated on their ability to synthesize information, solve complex problems, and create tangible outputs. The AI system can provide nuanced feedback that goes beyond right or wrong, offering insights into the strengths of the work, areas for improvement, and suggestions for further development.
Real-Time Coaching and Immediate Feedback
One of the most powerful aspects of human tutoring is the immediate feedback loop it provides. When a student struggles with a concept, a skilled tutor can recognize the difficulty instantly and adjust their teaching strategy on the spot. They can ask probing questions, offer alternative explanations, and provide encouragement precisely when it’s needed most. This real-time responsiveness has been difficult to replicate in digital learning environments, which traditionally rely on periodic assessments and delayed feedback.
The next generation of AI-driven learning platforms will fundamentally change this dynamic. By continuously monitoring learner interactions, these systems can detect confusion, frustration, or disengagement as it happens. Advanced natural language processing allows AI tutors to engage in meaningful dialogue with learners, answering questions, providing clarifications, and offering guidance in conversational, natural language. This creates a learning experience that feels less like interacting with software and more like working with a knowledgeable mentor who is always available.
The immediacy of this feedback is crucial for effective learning. Research in cognitive science has consistently shown that timely feedback significantly enhances retention and understanding. When learners receive corrections or confirmations immediately after attempting a task, they can adjust their mental models while the context is still fresh in their minds. AI-powered systems can deliver this feedback at scale, providing every learner with the attentive, responsive coaching that was previously available only to those with access to personal tutors.
Adaptive Learning Pathways
True adaptivity in learning systems goes far beyond simple branching logic or difficulty adjustments. It involves creating learning pathways that are genuinely responsive to each individual’s evolving needs, interests, and goals. Current personalization often relies on relatively crude metrics, such as whether a learner answered questions correctly or how much time they spent on particular content. While useful, these metrics provide an incomplete picture of the learning process.
Advanced AI systems can analyze far more nuanced indicators of learning. They can track the types of mistakes learners make, identifying whether errors stem from conceptual misunderstandings, procedural mistakes, or simple oversights. They can recognize when a learner is ready for more challenging material, even if they haven’t explicitly requested it. They can detect when a different teaching modality might be more effective, perhaps switching from text-based instruction to visual demonstrations or interactive simulations.
This deep adaptivity means that two learners pursuing the same overall learning goal might follow dramatically different paths to reach it. One learner might benefit from extensive practice with foundational concepts before moving forward, while another might thrive with quick exposure to advanced topics followed by deliberate practice filling in gaps. The AI system doesn’t force learners into predetermined tracks but instead crafts unique journeys that honor individual differences while ensuring comprehensive mastery of essential skills.
Responding to Emerging Skills and Knowledge
Perhaps one of the most exciting aspects of AI-driven learning is its potential to rapidly respond to emerging knowledge and skills. In today’s fast-paced world, new technologies, methodologies, and fields of study emerge constantly. Traditional educational institutions often struggle to update their curricula quickly enough to keep pace with these changes. By the time a new course is designed, approved, and delivered, the landscape may have shifted again.
Generative AI technologies offer a revolutionary solution to this challenge. These systems can automatically analyze emerging trends, synthesize information from multiple sources, and generate preliminary learning content addressing brand-new topics. Imagine a scenario where a groundbreaking technology is announced today, and within hours, a structured learning path is available to help people understand and work with that technology. This isn’t about replacing expert-created content, but rather about providing timely foundational materials that can be refined and enhanced by human educators.
The combination of AI-generated content and expert-vetted materials creates a powerful hybrid approach. Generative systems can quickly produce initial frameworks, explanations, and practice exercises for new topics. These materials are then reviewed, refined, and validated by subject matter experts who ensure accuracy, pedagogical soundness, and alignment with learning objectives. This collaborative process between artificial and human intelligence allows learning platforms to remain current and relevant in ways that were previously impossible.
Curating Comprehensive Learning Experiences
Content curation has always been a critical aspect of effective education. The internet provides access to virtually unlimited information, but this abundance creates its own challenges. Learners can easily become overwhelmed by the sheer volume of available resources, unsure which materials are reliable, appropriate for their skill level, or aligned with their learning goals. Expert curators have traditionally played a vital role in filtering and organizing content, but this manual process is time-consuming and difficult to scale.
AI-powered curation systems can process vast amounts of content, evaluating quality, relevance, and pedagogical value according to sophisticated criteria. These systems can identify complementary resources from diverse sources, creating comprehensive learning experiences that draw on the best available materials. They can recognize when a particular video explanation, interactive simulation, or written article would perfectly supplement the core curriculum, and seamlessly integrate these resources into personalized learning paths.
Importantly, this automated curation doesn’t operate in isolation. It functions as a tool that amplifies human expertise rather than replacing it. Learning designers and subject matter experts establish the standards and criteria that guide the curation process. They validate the AI’s recommendations, ensuring that suggested materials meet quality standards and align with educational objectives. This partnership between human judgment and machine efficiency creates a curation process that is both scalable and trustworthy.
Continuous Experimentation and Innovation
Building the future of AI-driven learning requires a commitment to continuous experimentation and learning. Education is a complex domain where human factors, cognitive psychology, technology, and pedagogy intersect in intricate ways. There is no single perfect solution that will work for all learners in all contexts. Instead, progress comes through systematic testing of new approaches, careful analysis of results, and willingness to iterate based on evidence.
This experimental mindset involves embracing a culture of hypothesis-driven development. Rather than assuming we know what will work best, we formulate specific hypotheses about how particular features or approaches might improve learning outcomes. We design experiments to test these hypotheses, collect rigorous data on learner experiences and outcomes, and analyze the results to draw meaningful conclusions. This scientific approach to educational technology development ensures that innovations are grounded in evidence rather than assumptions.
The process of experimentation also requires humility and openness to failure. Not every innovation will prove successful, and that’s not only acceptable but necessary for genuine progress. Each failed experiment provides valuable information about what doesn’t work and why, informing future efforts. By creating an environment where measured risk-taking is encouraged and learning from failure is valued, we can accelerate the pace of innovation and discover breakthrough approaches that might otherwise remain unexplored.
Sharing Insights with the Community
Knowledge and insights gained through experimentation have value that extends far beyond any single organization or platform. The broader educational technology community benefits when practitioners share their findings, both successes and failures. This open exchange of information accelerates collective progress, prevents duplication of effort, and helps establish best practices that can be adopted across different contexts.
Sharing insights involves more than simply announcing new features or celebrating achievements. It means providing transparent accounts of what was tried, how it was tested, what results were observed, and what conclusions were drawn. It means acknowledging limitations and uncertainties, not just highlighting positive outcomes. This level of transparency builds trust within the community and enables others to build upon our work, adapting successful approaches to their own contexts and avoiding pitfalls we’ve already encountered.
The commitment to community sharing also reflects a recognition that the challenges facing education are too significant for any single entity to solve alone. Improving how humans learn and develop skills is a collective endeavor that requires collaboration across organizations, disciplines, and perspectives. By contributing our learnings to the broader community, we participate in a larger movement toward transforming education for the benefit of all learners.
Building a More Human Way to Learn
There is a certain irony in using artificial intelligence to create a more human learning experience, but this apparent contradiction reveals a deeper truth. Technology at its best doesn’t replace human connection and understanding; it amplifies and extends our human capabilities. The goal of AI-driven learning is not to remove humans from the educational process but to free educators and learners from constraints that have limited what’s possible.
When AI systems handle the routine aspects of teaching, such as delivering content, tracking progress, and providing basic feedback, human educators can focus on what they do best: inspiring curiosity, fostering creativity, providing emotional support, and helping learners develop the metacognitive skills needed for lifelong learning. The technology handles the scalability challenge, ensuring that each learner receives individualized attention, while human mentors provide the wisdom, empathy, and inspiration that no algorithm can replicate.
A more human way to learn also means acknowledging and honoring the full complexity of human learners. People are not simply information-processing machines that need the right data input to produce desired outputs. They are emotional beings with hopes, fears, motivations, and contexts that profoundly influence their learning. Effective AI systems must be designed with this understanding, incorporating elements that support motivation, build confidence, acknowledge struggle, and celebrate achievement.
Creating Inclusive and Accessible Learning
One of the most promising aspects of AI-driven learning is its potential to make high-quality education more inclusive and accessible. Traditional educational systems have often failed to adequately serve learners with diverse needs, whether those differences stem from learning disabilities, language barriers, economic constraints, or geographic isolation. AI technologies offer powerful tools for breaking down these barriers and creating learning experiences that work for a broader range of people.
For learners with different abilities, AI can provide accommodations that are seamlessly integrated into the learning experience rather than being awkward additions. Speech-to-text and text-to-speech capabilities can assist learners with visual or motor impairments. Content can be automatically translated into multiple languages, making educational resources available to non-native speakers. The pace and complexity of instruction can be adjusted to match individual processing speeds and prior knowledge levels. These adaptations happen dynamically and individually, ensuring that every learner receives the support they need without stigma or delay.
Economic and geographic barriers can also be reduced through AI-enhanced learning platforms. High-quality educational experiences that were once available only to those who could afford expensive private tutors or attend elite institutions can be delivered at scale and low cost. Learners in remote or underserved areas can access the same sophisticated learning technologies and expert-curated content as those in major urban centers. This democratization of educational opportunity represents one of the most significant potential impacts of AI in learning.
Developing Essential Meta-Skills
While knowledge and technical skills are important, the rapidly changing nature of work and society means that meta-skills, the skills that help us learn and adapt, are becoming increasingly critical. These include abilities such as critical thinking, problem-solving, self-directed learning, collaboration, and emotional intelligence. Traditional educational approaches often struggle to effectively teach these higher-order skills, focusing instead on content delivery and knowledge retention.
AI-driven learning platforms can be specifically designed to foster these meta-skills. By presenting learners with open-ended challenges, collaborative projects, and opportunities for reflection, these systems can help develop the cognitive and emotional capabilities needed for lifelong success. The AI can provide scaffolding and support as learners tackle complex problems, gradually reducing assistance as capabilities develop. It can facilitate peer collaboration by intelligently forming learning groups and guiding collaborative processes.
Perhaps most importantly, AI systems can help learners develop metacognition, the ability to think about and regulate their own thinking. By making the learning process more visible and explicit, AI tutors can help learners recognize their own patterns, understand what strategies work best for them, and develop greater agency over their own learning. This metacognitive development is perhaps the ultimate goal of education, preparing learners not just with specific knowledge but with the capacity to continue learning effectively throughout their lives.
Ethical Considerations and Responsible Development
The tremendous potential of AI-driven learning comes with significant ethical responsibilities. As we develop systems that have increasing influence over how people learn and develop, we must carefully consider the values embedded in these technologies and their potential impacts on individuals and society. Questions of privacy, algorithmic bias, data security, and equitable access must be addressed proactively rather than as afterthoughts.
Privacy concerns are particularly acute in educational contexts, where detailed information about learners’ abilities, struggles, and progress is collected and analyzed. Strong protections must be in place to ensure this sensitive data is used only for its intended purpose of improving learning outcomes and is not exploited for commercial gain or other purposes that might harm learners. Transparency about what data is collected, how it’s used, and who has access to it is essential for building and maintaining trust.
Algorithmic bias represents another critical concern. AI systems learn from data, and if that data reflects existing societal biases, the systems may perpetuate or even amplify those biases. In educational contexts, this could mean that certain groups of learners receive different quality instruction or have different opportunities based on factors that should be irrelevant to their learning. Rigorous testing, diverse development teams, and ongoing monitoring are necessary to identify and mitigate these biases.
The Commitment to Transformative Education
The vision for AI-driven learning outlined here is ambitious, challenging, and transformative. It requires not just technological innovation but fundamental rethinking of how we approach education. It demands collaboration across disciplines, commitment to ethical principles, and willingness to challenge established practices. Most importantly, it requires keeping learners at the center of all decisions, ensuring that technological capabilities serve human needs rather than the reverse.
This is more than building a platform or developing new features. It represents a commitment to reimagining education for a new era, one where learning is personalized yet scalable, rigorous yet supportive, technologically advanced yet profoundly human. It’s a commitment to ensuring that every person, regardless of their background or circumstances, has access to high-quality learning experiences that help them develop their full potential.
The journey toward this vision will be long and complex, filled with both breakthroughs and setbacks. But the destination, a world where learning is more effective, accessible, and empowering for all, is worth the effort. By combining the best of human expertise with the capabilities of artificial intelligence, we can create learning experiences that were previously impossible, helping individuals and societies thrive in an increasingly complex world.
Conclusion
The future of learning is not about replacing human teachers or reducing education to purely technological interactions. Instead, it’s about creating a powerful partnership between human wisdom and artificial intelligence, leveraging the strengths of each to overcome the limitations of traditional educational approaches. Through adaptive personalization, authentic assessment, real-time coaching, and rapid response to emerging knowledge needs, AI-driven learning platforms can provide experiences that are simultaneously more effective and more human.
This vision requires ongoing commitment to experimentation, learning, and sharing insights with the broader community. It demands attention to ethical considerations and responsibility in development. Most fundamentally, it requires maintaining focus on the ultimate goal: not just building better technology, but creating a smarter, more accessible, and more humane way for people to learn and grow. This is the commitment that will guide the continued evolution of AI-driven learning, ensuring that technological progress serves the timeless human quest for knowledge, skill, and understanding.