Gen AI for Maritime Trade: A Beginner’s Guide

Artificial intelligence (AI) broadly refers to a machine or system that displays and mimics human-like behavior, such as learning and problem-solving. These tasks include understanding natural language, recognizing images, and making decisions. AI is the overarching field that guides scientists and technologists in teaching computers to mimic human intelligence. 

Vertical AI refers to AI applications tailored to specific industries, enhancing precision and efficiency. Windward’s platform is an example.

Machine learning is a subset of AI that enables computers to learn and make decisions from data without explicit programming. It enables intelligent systems to self-learn and constantly improve from data. The algorithms are fed with historical data, which helps them process and identify patterns automatically and make predictions about the future. They imitate the manner in which humans learn, but are able to perform tasks more accurately, saving time and money. 

There are two primary types of machine learning:

• Supervised learning: uses labeled data to train models where the expected output is known. This is like learning with a teacher. The computer is given labeled examples, such as a picture of a cat labeled “cat,” and it learns to identify similar items in the future.
• Unsupervised learning: finds patterns in unlabeled data, where the model identifies structures or relationships within the data without pre-existing labels. This is like learning without a teacher. The computer looks at significant amounts of data and tries to find patterns.

An Example of Supervised Machine Learning: Personalized Recommendations

Streaming services, such as Netflix, use AI to provide personalized recommendations. By analyzing users’ viewing history, preferences, and behaviors, AI algorithms predict what shows or movies a user is likely to enjoy next. This is achieved through supervised machine learning models that learn patterns and correlations within the data. For instance, if a user views Captain Phillips, the AI will suggest similar maritime titles, enhancing the user experience and keeping the user engaged. AI can create tailored experiences based on individual user data.

Deep learning is a subset of machine learning (either supervised or unsupervised) that uses neural networks (a logical structure of algorithms) to process and analyze complex patterns. Neural networks are designed to mimic the human brain, enabling computers to tackle more intricate tasks. They enable the processing of various data types, such as documents, images, text, and audio. Deep learning models perform best when solving extremely complex problems and using high volumes of data, quickly turning it into actionable information. 

For example, in the maritime domain, Windward’s deep learning models analyze location (GNSS) manipulation data to detect and understand complex patterns that simpler machine learning models might miss.

Deep learning can be divided into two types:

1. Discriminative AI focuses on distinguishing between different things, such as whether a picture is of a cat or a dog. It is trained on datasets of labeled data. 
2. Generative AI (Gen AI) focuses on generating new data similar to what it has seen before. When given a prompt, Gen AI uses the model to predict what the expected response might be and then generates new content. It can be trained on varied content, including text, code, images, and music.

Gen AI is a subset of deep learning (as mentioned above) and operates by learning patterns from vast datasets during its training phase. This training enables the model to generate content that aligns with the learned data patterns. The generated content can include text, code, images, and music, reflecting the diversity of the training data.

One popular application of this technology in enterprises is developing chatbots that engage in conversational interactions with business users, providing accurate answers to their questions. By utilizing private data, these systems can deliver personalized content to target audiences while ensuring data security. 

Training and Predictive Modeling

During training, Gen AI models use algorithms such as neural networks to analyze and learn from input data. The model identifies correlations, structures, and nuances within the data, forming a comprehensive understanding of the context. When prompted, the model predicts the most likely continuation or response based on this learned information, creating new content that aligns with the patterns it has seen before.

The Limitations

A significant limitation of Gen AI is its dependence on the scope of its training data. It can’t generate accurate responses or solve problems beyond the contexts it has been exposed to. If presented with a completely novel problem or question, the model may produce irrelevant or incorrect content, because it lacks the necessary background knowledge.

The Applications

Despite its limitations, Gen AI has a wide range of applications. Natural language processing (NLP) can generate coherent and contextually relevant text for applications such as chatbots, content creation, and translation services. In software development, Gen AI can assist in code generation and debugging. It can also create realistic images and videos in creative industries and generate music compositions by learning from existing pieces. 

Large language models (LLMs) are advanced deep learning models trained on vast amounts of text data to understand and generate human language responses when queried. They are a subset of Gen AI specialized in natural language processing and generation. They can handle various language tasks with minimal data fine-tuning and continuously improve with more data and parameters. LLMs excel at analyzing documents, summarizing unstructured text, and converting it into structured table formats.

Training LLMs 

There are three training methods for adopting LLMs to a new domain-specific model:

1. Prompt engineering: giving the model a set of clear examples and instructions on how to respond to certain inputs. This method is quick and often very effective when you just need to guide the model’s responses, without changing its core knowledge.
2. Fine-tuning (retraining): retraining the model with new, specific data from a specific domain to improve its performance in that area. This method is more resource-intensive and costly, but can be very powerful for creating highly specialized models.
3. Retrieval augmented generation (RAG): enhancing the model’s responses by integrating additional data, which can be private or real-time external information, that the model retrieves as needed. This is useful when you need the model to provide information that changes frequently, or isn’t included in its original training data.

Gen AI can empower decision-making with comprehensive risk assessments and insight summaries: 

• Expert integration: future-proof risk management with a virtual maritime expert that provides AI-based actionable recommendations to assess sanctions risk exposure accurately

• Enhanced productivity: accelerate decision-making processes and standardized communications with automated risk assessments highlighting key insights that truly matter for your business’ bottomline

• Optimized resources: leverage an integrated expert to complete screenings faster and without prior maritime knowledge, reducing hiring costs, onboarding time, and reliance on limited staff expertise

Navigating the Waters With Maritime AI™

Maritime AI™ is Windward’s vertical AI foundation. It uses deep learning AI models to understand maritime industry factors. AI predictive algorithms use behavioral patterns and historical data, and analyze AIS transmissions, weather forecasts, sanctions and blacklists, and much more. The multiple sources and maritime-centric algorithm help deliver accurate, real-time data on events occurring at sea, to mitigate risks, ensure regulatory compliance, provide actionable insights, and improve operational efficiency.

Windward has expanded our Maritime AI™ portfolio to introduce Windward Gen AI Agent – MAI Expert™. The industry’s FIRST Gen AI agent is a virtual maritime risk subject matter expert that leverages Windward’s proprietary AI models and human expertise, using innovative Gen AI engines. It is generally available now. 

Designed for precision and efficiency, MAI Expert™ empowers your decision-making with comprehensive risk assessments and insights summaries. It seamlessly integrates a reliable maritime and risk expert into your daily workflows and automates repetitive tasks to offer a strategic edge, and reliability.

MAI Expert™’s Area of Expertise

MAI Expert™ has been trained, and will continuously be trained, as a maritime and risk expert with the following areas of expertise:

A Gen AI agent can only be as good as its foundational data and instructions on what to extract. We have paired our proprietary, AI-driven knowledge-base with a unique, sanctions-focused prompt created by maritime risk experts. 

The prompt instructs MAI Expert™ to consider the following risk and maritime aspects:

• General maritime knowledge: ownership, vessels, port, state control, P&I insurance, ownership structures, and more 
• Sanctions knowledge: deceptive shipping practices, OFAC-specific regulations and advisories, UK and OFSI advisories, and EU advisories
• Windward-unique product knowledge: Organization Defined Risk (ODR), Windward activities, and AI models. ODR enables users to tailor behavioral indicators to meet their unique business and risk needs. Indicators can be applied to both behavioral and static screening data sets

It’s a natural evolution for Windward, to help pioneer the application of Generative AI to the maritime and ocean freight spheres. Some companies were excited to start talking about Gen AI following the Chat GPT explosion – we’ve been developing AI models since 2015. 

By leveraging our advanced models and technology, we aim to enhance productivity and streamline digitalization and automation in risk management. With our knowledge and tools, we are ready to support our customers on their Gen AI journey, ensuring a solid foundation for success.