Reimagining learning with AI

At its core, intelligence can be viewed as a process that converts unstructured information into useful and actionable knowledge.

The promise of artificial intelligence (AI), is that we may be able to synthesize, automate, and optimize that process, using technology as a tool to help us process data more effectively.

Powered by AI, the domain of learning is being reimagined at its core — accelerating time to mastery, improving engagement, and delivering rich learning analytics.

In this article, I’ll highlight some examples of what problems AI could solve for learners, educators, and authors.

Learning Assistant

No two learners are the same, which is why over 60% find misalignment between learning and their skill gaps to be a core learning issue.

A learning assistant identifies where skill gaps exist, and what learning is needed to reach mastery.

With adaptive assessment, learning, and review, the learning assistant personalizes the path to the individual needs of the learner — surfacing just what they need, right when they need it.

Through natural language processing, the learning assistant goes beyond hand-curated skill taxonomies to recommend assets related to your mistakes.

This has proven to reduce time to mastery by over 50% while improving knowledge retention by over 3x.

Smart Nudging

Lack of coaching and pedagogical support is one of the core challenges as we move to online-first learning.

Smart nudges deliver tiny, scientifically-based behavioral and pedagogical interventions — synthesizing a range of signals to increase course completions.

They leverage similar learner’s behavioural data to: set personalized goals; encourage you to reach them; surface supporting material when you stumble; and help you celebrate when you succeed.

For Coursera, such smart nudging increased course completions by over 10%.

Question Answering

Receiving real-time answers to your questions is critical when you’re stuck on a problem.

To follow how you naturally ask for answers, the search box moves from simple keyword matching to being able to answer complex questions about any content on the platform and understand the context of the words in your query.

To truly understand the nuances of how users search and ask questions, the language model we use was trained on a dataset of 8 million web pages learning 1.5 billion parameters.

This allows you to get just-in-time answers to your questions such as “where is blood pumped after it leaves the right ventricle?” and query your knowledge assets in the same way that you would ask questions to a tutor.

Learning Analytics

According to over 70% of C-Suite executives, inability to measure learning is the main workforce issue.

Powerful learning analytics allow you to evaluate and benchmark your teams’ strengths, weaknesses, and progress with unmatched clarity.

Go beyond false proxys like time spent and course completions with item response theory (IRT) to precisely measure the abilities and knowledge gaps of your learners.

Further, natural language search such as ”show me the average mastery of mechanical ventilation by hospital” is intuitive, suggests the next best question to ask, and provides immediate answers.

Actionable Insights

It’s not easy to know where to look in your data for useful insights on learning outcomes, let alone act on those insights effectively.

Using the power of AI, actionable insights are automatically personalized and delivered in real-time. Identifying unusual patterns that do not conform to expected behavior; consistent patterns in noisy data; similarities and differences between groups of data; and causal relationships between learning drivers and outcomes.

These insights give you the power of a thousand analysts, generating recommendations you can act on in a single tap.

Content Insights

Today course improvements stop once we hit publish, leaving ineffective content live with learners.

As you collect interactions from your learners, content insights find and rank opportunities to improve content iteratively over time.

Notably, content insights highlight which questions are confusing, which explanations need improvement, and even provide AI-generated suggestions of how to improve your content such as “remove double negatives from your question to effectively measure knowledge transfer”.

Content Generation

Creating courses and generating questions to comprehensively measure learning outcomes is a time-consuming task.

Smart authoring saves you time by automatically splitting static formats like PDFs into bite-sized content and generating questions based on your learning material.

Once generated, you can publish it immediately or edit starting with the AI-generated suggestions. Utilizing machine learning, the platform analyses which suggestions you approved and improves its suggestions over time.

Conclusion

The past decade has seen remarkable advances in AI. Computers now have the ability to see, hear and understand language better than ever before (see overview of important advances of the last decade).

These advances are inevitably going to transform how we learn from, access, and create knowledge assets, saving time while driving better learning outcomes.

As we adopt these technologies, privacy must remain at the heart of our approach. To this end, I am not only excited by how far we have come, but by where we are yet to go — especially as federated learning with differential privacy, capable of leveraging centralised insights across decentralised devices, gains traction.

Thank you Marc Ramos, Lori Niles, and Oscar Täckström for reviewing drafts of this. If you want to learn more about how you can leverage AI-powered learning in your organization, please send a note to joel@sanalabs.com