As Co-founder and CTO of Seaber, I’ve been deeply involved in every aspect of our software development since day one. Watching AI evolve and reshape the industry over the years has been nothing short of incredible. Now, I’m thrilled to share how Seaber’s AI strategy is revolutionizing the bulk cargo shipping industry. At Seaber, our mission is clear: to tackle the complex scheduling challenges in bulk cargo shipping with cutting-edge software solutions. AI is at the heart of this mission, driving us to create smarter, more efficient maritime operations.
At Seaber, we take data privacy seriously. By default, all customer data remains private and is never used to train our AI models without explicit permission. Our core AI capabilities are built using a combination of advanced optimization techniques and broadly available training data. For customers who wish to further tailor the model’s performance to their specific operations, we offer an opt-in approach—your data is only used if you choose to share it.
AI for predictive models
We are developing advanced predictive models that accurately forecast costs and durations for all aspects of cargo vessel schedules. This empowers our clients to make data-driven decisions, accurately forecasting and reducing uncertainty, and thus optimizing their operations. Our models can learn how schedules change as they approach the present and are actualized, enabling confident data-driven decisions.
The types of AI we have in our pipeline
- Generative AI for schedule improvements
-> Leveraging generative AI, we can create optimized shipping schedules that consider a multitude of variables, such as weather conditions, port congestion, vessel characteristics, port limitations, and cargo availability. This results in increased efficiency and cost savings.
- AI Agents performing complex tasks
-> Our AI agents are designed to handle complex tasks autonomously. They can adapt to changing conditions in real-time, ensuring that scheduling adjustments are suggested swiftly and effectively without constant human intervention.
- Reinforcement learning with human feedback (RLHF):
-> We utilize RLHF to fine-tune our AI systems. By incorporating feedback from human expert planners, our AI learns to align its objectives with their organizational goals, leading to better decision-making and improved performance over time.
The foundation models we’re building
We’re developing foundation models for several reasons. For those unfamiliar with the term, a foundation model, also known as a large X model (LxM), is a machine learning or deep learning model trained on vast datasets, enabling its application across a wide range of use cases. The foundation models will be trained on the synthetic data that will be produced in the pre-training step.
To train our foundation models, we generate a large number of problem instances of varying sizes and apply different optimization strategies based on the problem's difficulty. For problems of a manageable size, our foundation model is trained on analytically optimal solutions, ensuring the highest level of efficiency and effectiveness.
Another approach is to use heuristic optimization algorithms. When addressing large-scale problems, we utilize heuristics, meta-heuristics, and the most advanced optimization algorithms. These methods allow us to find near-optimal solutions within a reasonable time frame, which is crucial in the fast-paced shipping industry.
Heuristics are like smart shortcuts that help us find good solutions quickly when fully exploring every possible option would take too much time or effort.
Meta-heuristics provide practical, efficient methods for finding satisfactory solutions for complex scheduling and optimization problems, especially when a fully optimal solution is computationally infeasible. Heuristics rely on experience-based rules and approximations to yield good enough results quickly, enabling effective decision-making in dynamic and large-scale situations.
The optimization algorithms enable us to scale both pre-training and training without requiring precious human feedback. This can improve the quality of suggestions even for the most challenging scheduling problems.
When desired, our models can continuously learn from human input, refining the models' understanding of what our client organization values most, whether it's cost efficiency, speed, or reliability.
By integrating these advanced techniques, we're not just solving scheduling problems—we're enhancing overall operational efficiency, setting new standards in the industry.
Data privacy
Our foundation models are trained on purely synthetic data and independent information sources, ensuring that the models do not contain any sensitive customer data. If a customer desires, we can fine-tune the foundation models for their specific needs based on their objectives, historical data, and their use of our system.
We will also enable opting in to allow the use of their data in a shared generic model. If many customers are willing to share some of their data for training purposes, then they can all benefit from increased accuracy due to a more diverse dataset.
By default, all customer data is private, and nothing is used for training purposes without explicit consent.
Our long-term goals with AI
Looking ahead, we're investing in AI that can predict market demand with high accuracy. This will enable our clients to anticipate trends and adjust their strategies proactively, giving them a competitive edge.
We're also developing tools to perform break-even analysis of competitors. Understanding competitors' cost structures will help our clients make more informed decisions about pricing and market positioning.
In the long run, we predict that not only will users manage teams of AI agents to do computationally difficult and tedious tasks for them, but the agents will also request help from their human colleagues for tasks that require the experience, wisdom, and imagination of humans. Whenever agents have difficulty or even in principle could not make the proper decisions and judgements, it is always best to rely on the expertise already gained by maritime professionals.
AI for our clients’ needs
In summary, Seaber's AI strategy is a multifaceted approach aimed at solving today's challenges while preparing for tomorrow's opportunities. By combining predictive models, generative AI, and advanced optimization techniques, we're not only improving scheduling but also driving significant operational efficiencies for our clients.