Generative AI: An Easy Introduction & Its application in Shipping
As generative AI shapes the maritime industry's future, reliable data remains essential. The capability to gather, manage, and effectively utilize data will determine how we can maximize generative AI's potential.
If you've ever interacted with ChatGPT, you have already experienced generative AI firsthand. Although many of us have a loose sense of what generative AI refers to, the reality is that most people would struggle to explain why it has become such a hot topic in tech and innovation circles, and how the maritime industry can leverage it.
This piece aims to break down these complex concepts into digestible information, providing an easy yet insightful introduction to generative AI.
1. Misconceptions about Generative AI
Before exploring what generative AI is, let's clarify what it's not:
A Chatbot: Generative AI, like OpenAI's ChatGPT, does more than just chat. It creates diverse content, including text, images, and music. As a matter of fact, the typical chatbots that give standardized answers are not based on generative AI.
A Human-like Thinker: Generative AI can mimic human-like patterns in text, images, or even sound. However, it doesn't "think" or "understand" as humans do. It learns and reproduces patterns from input data, creating human-like content.
An Omniscient Search Engine: Generative AI's knowledge is limited to the dataset it was trained on, and it doesn't possess the capability to comprehend or reason beyond this scope. OpenAI's ChatGPT, for instance, has been trained on data up until 2021. On the other hand, Alphabet's Bard can access and process real-world information through Google Search, maintaining consistency with search results in its responses. This means it can provide the most recent information available online. However, in scenarios where information is scarce, Bard may also experience a "hallucination". It's crucial to emphasize that while it generates information that seems plausible and contextually fitting, it cannot guarantee the absolute accuracy of the information.
2. Simple definition of generative AI
As the name states, it is a type of AI that can be used to generate new, previously unseen data samples that are similar to the training data. This could be in the form of text, images, music, or any other form of data. For example, GPT-4 is a type of generative language model that learns about patterns in language through training data. Then, given some text, it predicts what comes next, and writes it out in the form of natural language.
Below is an overview of relevant applications:
3. Understanding the AI Family Tree
We hear all the time about AI, machine learning, deep learning, and large language models. Understanding the connection will help explain why generative AI is transformational.
Artificial Intelligence (AI) is a broad field like physics or mathematics. The goal, however, isn't to perform calculations but to carry out tasks in a "smart" way, i.e. seeing computers to perform human-like behavior.
Machine Learning (ML) is a subset of AI that trains models using input data. A key feature of ML is its ability to learn without explicit programming. For example, in predictive maintenance, ML can analyze historical and real-time data to discern patterns and anomalies, predicting when a piece of equipment may likely fail.
Deep Learning is a subset of ML. It can process more complex patterns than machine learning. The special thing that makes deep learning “deep” is the artificial neural networks, which are inspired by the human brain. Like human neurons, they are made up of many interconnected nodes or neurons that can learn to perform tasks by processing data and making predictions. DL models can be divided into two types, generative and discriminative.
It’s worth noting that neural networks can use both labeled and unlabelled data. This is called semi-supervised learning. In semi-supervised learning, a neural network is trained on a small amount of labeled data and a large amount of unlabelled data. The labeled data helps the neural network to learn the basic concepts of the task while the unlabelled data helps the neural network to generalize to new examples.
Finally, Generative AI, a subset of deep learning, uses artificial neural networks to create new content based on knowledge acquired from existing data. It processes both labeled and unlabelled data using supervised, unsupervised, and semi-supervised methods. A generative model generates new data instances based on the learned probability distribution of existing data.
4. How generative AI models works
While there are several types of generative models such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and transformer-based models like GPT-3, GPT-4, the general process for training them is somewhat similar.
The following is a step-by-step explanation of how GPT works, using the question "What are the monthly maintenance requirements for the CO2 Fire Extinguishing System?" as an example.
Step 1: Input Processing
Before your question can be processed by GPT, it needs to be “translated” into a format that the model can understand.
This is done through a process known as tokenization and embedding. Tokenization breaks down the sentence into smaller units (tokens), which can be as short as one character or as long as one word. These tokens are then embedded, meaning each token is represented as a high-dimensional vector that the model can process. These vectors capture semantic information about the tokens.
Step 2: Context Understanding
T in GPT stands for transformer, which is a mechanism known as the transformer architecture to understand the context of your question.
The transformer model is designed to process sequences of data like your sentence, and it excels at understanding the relationships between words (tokens) in a sentence.
It does this through a mechanism known as self-attention, which allows the model to weigh the importance of each token in the context of all the others. This allows the model to generate nuanced and contextually appropriate responses, even over long distances between words.
Step 3: Answer Generation
Once the model has processed the input and understood the context, it begins to generate a response. This starts with the model determining the most likely next word (token), given the input and what it has learned during its training. The transformer architecture of GPT makes this possible by considering the context of all previous words in the sequence when predicting each subsequent word.
The predicted word is then added to the sequence, and the process repeats, with the model making its predictions based on the extended sequence.\
Step 4: Output
After generating the complete sequence for the response, the process of unembedding takes place. The vectors representing the predicted tokens are converted back into human-readable language. The answer generated by GPT reads: "According to MSC.1/Circ.1318/Rev.1 from 25 May 2021: 4.1 At least every 30 days a general visual inspection should be made of the overall system condition for obvious..." is then presented as a response to your original question.
5. Generative AI in maritime technical management
Generative AI can be employed in the maritime industry with numerous ways to help optimize operations, reduce costs, and improve decision-making processes.
We at Kaiko Systems envision (and build) a dynamic synergy where generative AI and our platform complement each other to usher in a new era of maritime technical management. We see several crucial areas where this integration could manifest:
Risk Assessment: With historic operational, safety, and compliance data of the vessel, generative AI can pinpoint risk factors that precede incidents. Furthermore, the advanced algorithms can analyze patterns in Port State Control inspections and off-hire occurrences, in combination with the vessel health data, providing a comprehensive overview of operational risks. With the power to generate timely alerts and automatically generated mitigation strategies, generative AI can significantly enhance the safety, reliability, and efficiency of maritime operations.
Document Management: By comprehending complex relationships between diverse maritime regulations, procedures, and operational data, generative AI facilitates precise and contextually relevant search results for the shore team. Additionally, regulatory changes can be automatically highlighted to ensure that the shore team is always in line with the latest compliance requirements.
Knowledge Sharing: Generative AI also brings a new way of crew learning and training. All inception and maintenance work conducted with Kaiko Systems can be fed back to the algorithms. This not only encapsulates the experiences of seasoned crew members but also integrates this knowledge with up-to-date company procedures, industry regulations, and vessel health information. As a result, crews can obtain advice, suggestions, and answers to their queries in real-time, during the course of their inspection and maintenance tasks.
Similar to ChatGPT, Kaiko Systems' apps will allow crew members to ask questions and receive first-level support to gather comprehensive information before moving on to the TSI for further processing.
Regulatory Compliance: Given a set of new inspection regimes like SIRE 2.0 and historical inspection data, generative AI could generate a list of compliance tasks, highlighting the changes from previous versions.
Automated Communications: AI can assist in drafting emails or other communications based on the data inputs, saving time for fleet managers. This could include communicating with suppliers, reporting to regulatory bodies, or internal communications within the company.
Conclusion
In conclusion, while technologies like blockchain and virtual reality (VR) continue to draw interest in the maritime industry, generative AI stands out for its immediate potential to bring transformative value. More importantly, unlike these other technologies, generative AI can be much easier and quicker to implement, given its flexibility and broad range of applications.
The future of the maritime industry is being shaped by generative AI, and reliable data is an essential piece of this puzzle. As we move forward, the capability to gather, manage, and utilize data effectively will dictate how well we can leverage generative AI to its full potential.
The future isn't just about sailing with the currents; it's about setting the direction. By harnessing the power of generative AI and quality data, we can set the course for a new era of fleet technical management, propelling the maritime industry towards an exciting and prosperous future.