Better understanding how generative AI and predictive AI are transforming professional learning

Artificial intelligence is currently at the center of attention in the field of professional training. But what exactly do we mean when we talk about Generative AI or Predictive AI? What makes these two forms of artificial intelligence different, how do they work, and what do they bring to the table? 

In the white paper Learning and sharing knowledge differently thanks to Artificial Intelligence, Frédéric Oru, CEO of AI for Better and AI expert, explains both concepts in detail and provides concrete examples of how they can be applied to learning. Here’s an excerpt from his interview.

When we talk about AI in training, what do we mean?

We’re mainly talking about two things: Generative AI and Predictive AI. 

Generative AI is today’s star, largely thanks to ChatGPT. It’s a branch of AI that creates content based on user input. Feed it a sentence, an image, or even a piece of music, and it generates something new from that. It’s incredibly versatile: it can create images, compose music, write text, summarize articles, explain tasks, or even edit photos. 

Most importantly, it doesn’t stick to one format. It can combine different types of content, for instance, creating complete training materials: a learning synopsis, an e-learning module, a slide deck for an in-person workshop, or a virtual class.

How does it learn to do all that?

This is where it gets interesting. Generative models learn from existing data. They analyze that data to understand patterns and structures, then use that knowledge to generate new, similar data. But here’s the catch: even though they can produce entirely new content, they don’t actually understand the meaning of what they’re creating. 

Essentially, they learn by imitation. They take existing creations, translate them into mathematical language, and train themselves to reproduce or generate something similar based on a directive, what we call a “prompt.”

So they’re good at creating and imitating, but without really understanding?

Exactly. It’s a form of creativity, but very different from human creativity. This opens up exciting opportunities, but also reminds us of the gap between imitation and true understanding of creation.

And what about Predictive AI?

Predictive AI (also called discriminative AI) is the most widely used, yet often the least understood. Based on machine learning models, it focuses on predicting outcomes from input data, hence the name, because it “discriminates to predict.” 

It’s capable of differentiating, sorting, and selecting (in a positive sense). Predictive AI is built on a few key elements: data, model, training, evaluation, and inference, the process that leads to making predictions.

What does that look like in practice?

Let’s take a simple example. Imagine developing a corporate training program. Predictive AI could analyze employee performance, learning styles, and feedback from past sessions. Based on that, a machine learning model could predict which type of training would be most effective for each employee or group, aligned with the company’s goals. 

It may sound straightforward, but done well, it can yield highly impactful results.

Can you tell us more about the different machine learning models?

There are three main types of machine learning: supervised, unsupervised, and reinforcement learning. 

• Supervised learning can be used to predict employee performance based on historical data. 
• Unsupervised learning can identify trends or clusters among employees, helping tailor training programs. 
• Reinforcement learning can optimize training strategies in real time.

But how do we manage the risk of bias in these models?

That’s a crucial point. Biases like stereotyping or confirmation bias can creep into AI models and distort predictions. In professional training, that could mean inappropriate recommendations for certain groups of employees. 

The solution is to work with diverse datasets and regularly test models to detect and correct biases.

Can you share a concrete example of Predictive AI applied to professional training?

Of course. Imagine an AI system designed to analyze the skills and knowledge of a group of learners. Using supervised learning, the system could evaluate not just where each learner excels, but also pinpoint areas where they need improvement or additional learning. 

It could take into account multiple data sources, past test results, trainer feedback, even learner interactions with specific learning materials. Based on that analysis, the AI could then generate personalized recommendations: targeted training modules, learning resources, or even teaching approaches best suited to their individual needs. 

This has two major benefits: 

• Relevance and efficiency, every learner receives training tailored to their actual gaps and strengths, maximizing both time and resources. 
• Engagement and motivation, learners see that their path is personalized, which boosts motivation and confidence as they progress.