Getting past the word salad: A guide to understanding AI

If you attend as many AI events and conferences as I do, you will hear a lot of tropes: that AI is not really new, that there are many types of AI, and the term “GenAI” or generative AI is the trend now.

Usually, these conversations just confuse and alienate the layman on what AI is, and with the escalating discussions on regulating AI and its harms, I think it’s important to split some hairs and stop casting a wide net on what constitutes artificial intelligence.

Generative AI vs. Discriminative AI: Understanding the difference

At the heart of AI lies the distinction between generative and discriminative models. These two categories serve different purposes and are foundational to how AI functions in various applications.

Generative AI (GenAI) models are designed to generate data, hence the name. They can generate text, images, music, and more, mimicking the style and structure of the input data they were trained on. The most common examples of generative AI today are chatbots like ChatGPT, Gemini, and Claude and image generation tools such as DALL-E and Midjourney. Now we also have generative music from Udio and sound and voices from Eleven Labs. These tools are capable of producing coherent and contextually relevant responses in natural language or creating realistic images, music, and sound from textual descriptions.

Discriminative AI, on the other hand, is focused on distinguishing between different kinds of data. It classifies input data into predefined categories. Outside of its uses in data analysis, one of the most pervasive uses of discriminative AI is in recommender systems. Applications like Waze, Netflix, YouTube, Facebook newsfeeds, and Google search results rely heavily on discriminative models. These systems analyze user behavior and preferences to recommend routes, movies, videos, news articles, or search results that are most likely to be of interest.

The mechanism behind AI: Machine learning

At the core of both generative and discriminative AI is machine learning (ML). ML is the mechanism that enables AI to learn from data and improve over time. It involves training algorithms on large datasets to recognize patterns, make decisions, and predict outcomes.

There are various types of ML, including supervised learning, unsupervised learning, and reinforcement learning, each suited to different kinds of tasks.

•    Supervised learning involves training a model on a labeled dataset, where the correct output is known. This approach is often used in classification and regression tasks. Examples include models that detect faces, identify cats and dogs in photos, or forecast sales from past data.

•    Unsupervised learning deals with unlabeled data, where the algorithm tries to find hidden patterns or intrinsic structures within the data. Clustering and association are common unsupervised learning tasks. These are models that group populations into segments, match products to customers, and identify anomalies for fraud detection.

•    Reinforcement learning involves training an agent to make a sequence of decisions by rewarding desirable actions and penalizing undesirable ones. This approach is frequently used in robotics and game playing. The infamous AlphaGo that beat Lee Sedol, the world master in Go, is an example of a reinforcement learning application. The amazing Boston Dynamics robot dogs and humanoids also use reinforcement learning to learn how to walk life-like in varying terrains and environments.

Data science: The intersection of AI and analytics

Data science is the discipline that encompasses AI, machine learning, and analytics. It involves using statistical methods, computational tools, and domain expertise to extract insights from data.

Data science is the backbone of AI, providing the techniques and processes required to build and refine AI models. Analytics is the use of data to aid in human decisions, while AI primarily concerns itself with automated decision-making, subject to human-defined parameters.

The rise in popularity of data science and the data scientist as an occupation in recent years is arguably a contributor to the mainstream popularity of AI today.

AI is mainstream

Finally, as many generative AI tools are immediately usable without writing programming code, AI has truly become mainstream, which for me implies that all occupations should be into artificial intelligence, not just data scientists.

GenAI can even be the gateway for the public to learn more about AI itself, and become the new mandatory skill in every CV, alongside word processing, spreadsheets, and the internet.

I will confess that all this information can be a lot to absorb at first read, but hopefully, by understanding these core aspects of AI, machine learning, and data science, we can move past the jargon and grasp the true potential and limitations of these technologies.

This knowledge enables us to appreciate how AI-driven systems impact our daily lives, from personalized recommendations on streaming services to advanced chatbots that facilitate customer support.

Dominic Ligot is the founder, CEO and CTO of CirroLytix, a social impact AI company. He also serves as the head of AI and Research at the IT and BPM Association of the Philippines (IBPAP), and the Philippine representative to the Expert Advisory Panel on the International Scientific Report on Advanced AI Safety.