Beyond Generative AI: Active Learning and AI Teachers

By Atefeh Behdadian and Krizia Silvestri

The release of Stable Diffusion forever changed the art world, and ChatGPT shook up the internet with its ability to write songs, mimic research papers, and provide thorough and intelligent answers to commonly Googled questions. These advancements in Generative AI offer compelling evidence that we stand on the precipice of an AI revolution.

However, most of these Generative AI models are foundational: high-capacity, unsupervised learning systems that train on vast amounts of data and require millions of pounds’ worth of processing power.

Currently, only well-funded institutions with access to massive GPU resources can build these models. The majority of companies developing application-layer AI, which drives widespread technology adoption, still rely on supervised learning with large, labelled training datasets.

Despite the impressive feats of foundational models, we remain in the early days of the AI revolution, hindered by bottlenecks that limit the proliferation of application-layer AI. Beyond the well-known data labelling challenge, additional data bottlenecks exist downstream, impeding the development and deployment of later-stage AI in production environments.

These challenges explain why self-driving cars have always seemed to be "just a year away", since 2014.

While exciting proof-of-concept models perform well in research environments, they struggle to predict accurately when deployed in the real world. A significant issue is their inability to meet the higher performance thresholds required for high-stakes production environments. They fall short in terms of robustness, reliability, and maintainability.

For instance, self-driving cars often mistake reflections of bicycles for actual bicycles due to their inability to handle outliers and edge cases. Similarly, a robot barista may make a perfect cappuccino two out of five times but spill the cup the other three.

As a result, the AI production gap—the divide between “that’s neat” and “that’s useful”—has proven more substantial and formidable than initially anticipated by ML engineers.

Fortunately, ML engineers are adopting a data-centric approach to AI development, and active learning strategies have gained prominence.

Active Learning: Bridging the Gap

Active learning transforms supervised model training into an iterative process.

Initially, the model trains on a subset of labelled data from a large dataset. It then attempts predictions on the remaining unlabelled data. Instead of passively waiting for the next batch of labelled data, active learning identifies the most uncertain or informative examples and requests labels for those instances.

This iterative approach refines the model’s understanding and enhances its performance. By selectively labelling the most valuable data points, active learning reduces the reliance on massive, labelled datasets. It optimises the learning process, resulting in AI models that are more robust, reliable, and adaptable to real-world scenarios.

While generative AI holds immense promise, it’s active learning that will propel us beyond “that’s neat” to “that’s useful.” We are on the edge of an AI revolution that goes beyond benchmarks and changes everyday applications.

Active Learning and AI Teachers

Active learning, as mentioned earlier, transforms the process of training supervised models into an iterative journey. Initially, a model trains on a subset of labelled data from a large dataset. However, instead of passively waiting for the next batch of labelled data, active learning takes a more proactive approach. It identifies the most uncertain or informative examples within the remaining unlabelled data and requests labels for those instances. This iterative feedback loop refines the model’s understanding and enhances its performance.

By selectively labelling the most valuable data points, active learning reduces the dependency on massive, labelled datasets. The result? AI models that are more robust, reliable, and adaptable to real-world scenarios.

Now, let’s turn our attention to AI educators. These digital instructors exist, and they are a reality. Imagine a teacher or a trainer who knows precisely when a learner needs extra help with algebra or struggles with Shakespearean sonnets. That’s the promise of AI in Education.

AI teachers analyse data, adapt to individual learning styles, and provide real-time feedback, to learners and to human teachers.

They personalise the learning experience, making education more effective and engaging.

For example, at Otermans Institute, the team is focusing on building OIAI and its AI teachers, reshaping and creating:

1. Course Delivery and Content Creation: OIAI teachers not only deliver lessons but also create course content. They transform traditional one-directional instruction into dynamic, two-way conversations between students and their AI teachers. Universities and institutions worldwide, including Arab American University (Palestine), Bahria University (Pakistan), Ibn Haldun University (Turkey), Catholic University of Eastern Africa (Kenya), and Universidad de Las Américas (Chile), have embraced this educational revolution.
2. Human-Like Interaction: AI digital human teachers developed by Otermans Institute possess a human-like appearance and voice. They interact with learners using both voice and text inputs.

Our conclusion

While high-capacity generative AI models exist, only well-funded institutions with access to massive GPU resources can leverage them effectively. Most companies developing application-layer AI struggle with accurate real-world deployment, prompting ML engineers to adopt a data-centric approach. Active learning emerges as a solution, transforming the training of supervised models into an iterative journey. Initially, models train on a subset of labelled data, then actively identify and label the most uncertain examples, refining their understanding and performance.

This approach finds practical application in AI-powered education, as exemplified by Otermans Institute. AI educators analyse data, adapt to individual learning styles, and provide real-time feedback to both learners and human teachers. These AI teachers not only deliver lessons but also create course content, transforming traditional one-way instruction into dynamic, interactive dialogues.

With human-like appearance and voice, these digital instructors personalise the learning experience, making education more effective and engaging. This approach highlights how active learning can connect theoretical AI advancements with real-world uses.

it’s crucial to stay informed about the latest advancements in AI and Education: If you are interested to know more about OIAI and its technology, feel free to contact us. Our mission is to empower humans, not substitute them.