Workforce Learning in the AI Era: Skill Mastery vs. AI Supervision

In October 2024, I had the privilege of being invited as a guest speaker at the 2024 ATD APAC Conference in Taiwan. The theme of this year’s conference was Empowering Future Talent. During my session, titled A New L&D Paradigm Shift with AI Beyond Technologies, I was deeply inspired by the thought-provoking questions from the audience. Their curiosity and insights motivated me to write this article.

At conferences, speakers often take the spotlight, but it’s the audience’s questions that ignite new ideas and open doors to innovative thinking. One question, in particular, resonated with me: “As AI continues to improve at an incredible pace, what should we prioritize in workforce development, and how can we effectively measure their performance accordingly?”

This is a broad question, and its answer depends on specific areas of profession. At the risk of oversimplifying, I would argue that organizations should help their employees advance their skills to evaluate, synthesize, and create solutions for problems where they are domain experts.

Let’s unpack this argument with a quick quiz:

What is the answer to this math equation?

5 - 2 + 2 × 3

I hope you got the correct answer (it is nine!). But what does this have to do with the question at the start of this article?

Although it is a fairly simple math question, arriving at the correct answer without a calculator required not only completing an end-to-end computation process but also remembering the rule that multiplication takes precedence over subtraction and addition. In other words, we knew the right way to approach this question based on our background knowledge. Furthermore, we were able to think critically to know it is incorrect if another person (or AI) says that the answer is 15.

Bloom’s Taxonomy & Autonomous AI Systems

Based on Bloom’s taxonomy, there are different levels of cognitive skills. These cognitive skills are structured into six hierarchical levels in the following:

When paralleling Bloom’s taxonomy with recent breathtaking developments in large language models (LLMs), we see that generative artificial intelligence (Generative AI) and its applications are now capable of remembering and understanding information humans provide (e.g., summarizing and interpreting knowledge from academic textbooks or research articles). It can also analyze information in bulk and produce desired outputs prescribed by humans (e.g., performing data analysis and creating data visualizations or image-to-text translations).

Furthermore, exciting progress in agentic action and Retrieval-Augmented Generation (RAG) suggests that AI will soon mature its ability to apply knowledge and take actions on our behalf to achieve tasks based on success outcomes we define. This might leave us wondering: in that case, what contributions can human experts make when tasks are highly automated by AI?

The Changing Nature of Workforce Learning

As our economy has shifted from labor-intensive work to knowledge creation and innovation, albeit not universally, the nature of workforce learning is also changing. In the past, the drill-and-kill method or apprenticeship by modeling was effective in sustaining a workforce that followed an optimal recipe to perform tasks. The cognitive skills these training approaches focused on were remember, understand, and apply.

However, we are now at a time where the boundaries of knowledge and skills are constantly being pushed and expanded—not just by human experts but by generative AI as well. This revolution naturally prompts us to focus on higher-level cognitive skills where human expertise continues to outperform AI.

Fields such as data analysis, software development, content creation, and even healthcare have seen AI systems outperform humans in certain tasks that involve mostly basic cognitive skills. While we should continue to research the biases and errors that AI makes from one realm to another, it might be a matter of time before AI can remember facts and rules better, understand the logic and subtle patterns of problems quicker, analyze the pros and cons of solutions more thoroughly, and apply actions faster than us.

Practically, when AI can write news articles, build websites, and analyze data faster than us, what do organizations want their employees to excel at?

Remember the math quiz: 5 - 2 + 2 × 3?

The Shift Toward AI Supervision

As autonomous AI agents execute actions based on the goals we set, the role of the human expert is increasingly shifting toward supervision. This role requires a different set of skills—skills that may not involve completing specific tasks but rather guiding AI systems to deliver the right outcomes. Task definition, problem framing, and providing corrective input when AI veers off course are becoming as valuable as the tasks themselves.

For example, in industries like marketing or journalism, AI-generated content needs human oversight to ensure that it aligns with brand messaging or journalistic standards. In data science, AI can crunch vast amounts of data, but human experts are needed to ask the right questions and ensure that output analytics positively impact the business. In software development, different versions of code may be proposed to build a banking app, but a senior software engineer’s intuition to compare pros and cons in each code version and identify bugs or safety risks ahead of time will remain important.

This shift calls for a new focus in workforce training—problem-solving, critical/contextual thinking, and ethical judgment—where people act as supervisors, curating and refining the outputs of AI systems. This new training focus is crucial for developing a workforce’s high-level cognitive skills based on Bloom’s taxonomy: analyzing, evaluating, and creating solutions for problems across different realms.

Where Skill Mastery Still Matters

While AI supervision is an essential skill in today’s workforce, skill mastery remains critical as a guardrail for evaluating and implementing solutions created by autonomous AI agents.

Fields like medicine, law, and engineering require a level of human judgment, empathy, and critical thinking that AI cannot easily replicate. For example, while an AI system might analyze medical images or suggest potential treatments, the final decision often involves ethical considerations, patient interaction, and a nuanced understanding of human biology that only a seasoned medical professional can provide. Similarly, in law, AI might assist in legal research or contract drafting, but interpreting the law and addressing ethical considerations based on years of practice is essential for meeting a client’s specific needs.

Conclusion: We Will All Become Solution Engineers or Designers

In the future, we can imagine everyone becoming a solution engineer or designer, not just those in IT or creative industries. With domain-specific expertise, our focus will be on asking the right business questions, orchestrating the information and answers generated by AI agents, and creating innovative solutions to solve problems that existing answers cannot.

The future of workforce learning is not about constantly looking for distinct skill sets that AI cannot master. As AI systems evolve—and at a rapid pace—the key will be to embrace the duality of skill mastery and AI supervision. The critical question to ask in order to find this balance is not “Which skills should human experts focus on so they can keep their jobs?” but rather, “Which levels of cognitive skills should they prioritize in a specific domain?” Balancing knowledge mastery and AI supervision will define success in the AI-driven world of tomorrow.

A Heartfelt Thanks to the 2024 ATD APAC Conference Organizers

It was an absolute pleasure to attend the 2024 ATD APAC Conference in Taiwan. I am grateful for the opportunity to learn about the L&D challenges facing the APAC region and to collaborate with fellow speakers and attendees in brainstorming innovative solutions and pushing boundaries together. I eagerly look forward to next year’s conference!