Artificial Intelligence (AI) is in the news and becoming part of our daily lives. While you may know that AI has something to do with the advertisements you see on your favorite website, it’s less clear how the website knew you might be interested in that advertisement. What about other applications of AI? And can you trust AI? How might you interact with something differently if you know it's AI?

Terminology

Let's start with some common terminology to discuss how AI can be used with maritime data.

Model

In Skylight and other AI contexts, you may hear about “a new model” or models that perform specific tasks. But what is a model?

A model is a mathematical framework designed to represent a task or solve a problem using available data. Advanced AI models learn patterns and relationships from the data to make predictions, classifications, or decisions.

For example, a "Fishing detection model" in Skylight can recognize fishing-like behavior based on a vessel's movements. When the model receives new data, such as streaming AIS messages, it can determine whether the speed, location, and other factors suggest fishing activity, and if so, it creates a fishing event in Skylight.

Different types of model architectures exist (e.g., Random Forest). You should be aware that not all architectures are the same and depend on various use cases. 

Artificial Intelligence (AI)

AI is a broad term for when computers perform tasks that typically require human intelligence. Some AI systems are as simple as decision trees that provide responses based on predefined rules. For example, a basic email spam filter might label an email as spam if it contains certain keywords like "send money" or "lottery winner."

Machine Learning (ML)

Machine Learning is a subset of AI that uses data and statistical methods to learn and make decisions.

ML systems use data—such as images, numbers, sounds—to provide outputs like suggestions or categorizations. Unlike traditional software that relies on explicit programming, ML models learn from data patterns.

For example, in credit card fraud detection, the ML model learns from historical transaction data. If you typically use your credit card for groceries and fuel, but suddenly buy a diamond ring while on vacation, the model might flag this as potential fraud. It learns what is normal and abnormal based on the data.

ML and AI are often used interchangeably, but ML is a specific subset of AI focused on learning from data. Data is the foundation of machine learning and the quality of models is directly related to the quality and quantity of data used to “train” a model.

Deep Learning

Deep Learning is a type of ML that mimics the way the human brain makes connections.

Imagine you grew up eating Brazilian cuisine and had never seen or heard of another country's cuisine. But then one day you traveled to Vietnam. Your brain would likely still recognize a bowl of rice noodles as a meal, despite the unfamiliar ingredients and presentation, because it connects various cues like the plate, utensils, and setting (table, restaurant, etc) among other indicators.

Similarly, deep learning models use multiple layers to understand complex patterns in data. You might hear the term Neural Networks in this context. A neural network consists of layers that perform tasks, and in deep learning, many layers work together to connect vast amounts of data, like the human brain.

For instance, a deep learning model for detecting fishing behavior would understand not just fishing-specific movements but also other vessel behaviors, making it more accurate than a simpler ML model tasked with a singular job. 

It's not always possible to understand how a model arrives at an insight, much the same as you might not always be able to explain exactly how you recognize someone.