When I was fishing and at University, we had this concept of “sana envidia,” which translates to “healthy envy.” However, I think “wholesome envy” better defines it.
It’s basically when you wish you were as clever, charming, good-looking, or whatever as someone else, but not in a mean way—more in a complimentary manner… like saying, "Good on you, mate. You’ve got it sorted."
I have that feeling with a few colleagues/friends, mainly related to their writing skills and extensive policy-writing expertise; they seem to write good stuff (seemingly effortlessly), and I admire that; hence, when jobs that require that expertise come my way, I pass them immediately to them, because they will no doubt do a much better job than I ever will… ergo, “wholesome envy” Lars Olsen and Viv Fernandes are two of them, who, besides being great colleagues, are also friends.
A few weeks ago, I blogged about the Intellectual Property (IP) of the data feed into Machine Learning (ML) to be used in EM, which was well received… Then, while talking to Viv Fernandes, it should not have come as a surprise that he wrote a great study on this topic that you can download from here.
Below is a summary of it… and I agree with the conclusion!
Artificial intelligence (AI) models integrated into electronic monitoring (EM) systems present transformative opportunities for fisheries management, especially in the Pacific region. These models can automate tasks such as catch event detection, species identification, and activity recognition, thereby significantly enhancing the efficiency and scalability of EM programs. However, the success of AI models relies on access to high-quality, representative training datasets.
Viv”s report examines the feasibility of developing a regional annotated image library to support AI in EM systems, emphasising its benefits, challenges, and implementation requirements.
Key Benefits of a Regional Annotated Image Library
A regional annotated image library would offer numerous benefits for members of the Pacific Island Forum Fisheries Agency (FFA):
Scalability: A large, diverse dataset would enable AI models to be trained effectively, supporting broader applications and larger datasets.
Improved Accuracy: Diverse datasets from different vessels, environments, and conditions would enhance the reliability of AI models, particularly for rare events like bycatch detection.
Accelerated AI Development: Shared resources would fast-track AI model development, reducing costs and time for individual members.
Program Efficiencies: Automation would lower human resource costs and improve the speed of EM data analysis.
Capacity Building: Developing AI, machine learning (ML) expertise, and data annotation would strengthen regional capabilities.
Regional Collaboration: Pooling resources would create efficiencies in fisheries management, reducing individual member costs and fostering collective progress.
Digital Upskilling: Transitioning roles from observers to validation officers would build a digital culture and workforce in the Pacific.
Training Resources: The library could complement existing programs by providing valuable training for EM analysts and observers.
Implementation Requirements
To create a regional annotated image library, FFA members need to consider several key factors:
Clear Objectives: Define EM program goals, such as compliance monitoring, catch monitoring, or protected species tracking, to guide AI model development.
Data Annotation Standards: Ensure high-quality annotations through skilled personnel, appropriate tools, clear guidelines, and robust quality assurance processes.
Representation and Diversity: Include diverse datasets to train AI models effectively, accounting for varying conditions, species, and vessel types.
Funding: Ensure sustainable financing for initial setup and ongoing maintenance, including expenses for data storage, annotation tools, and human resources.
Interdisciplinary Collaboration: Engage fisheries and IT specialists to ensure annotations are precise and meaningful.
Data Governance: Create a suitable governance framework to oversee data across its lifecycle, ensuring security, ownership, and confidentiality.
Technological Infrastructure: Invest in IT infrastructure, including cloud-based solutions, to support data storage and processing needs.
Challenges and Issues
Despite the benefits, several challenges must be addressed:
Cost: Setting up and running the library demands substantial investment in hardware, software, and personnel.
Technological Capability: Limited local capacity in AI and ML technologies requires partnerships with external providers or regional organisations.
Data Storage: Managing large datasets requires scalable solutions, such as cloud-based storage, which may involve high costs and infrastructure upgrades.
Legislative Barriers: Outdated national legislation can obstruct data sharing and the adoption of new technologies.
Cultural Factors: Traditional mindsets and resistance to change may slow the adoption of new technologies.
Common Language: A standardised technical vocabulary is essential to enable effective communication among fisheries managers, policymakers, and technology providers.
Privacy and Ownership Considerations
FFA members have highlighted the importance of data sovereignty, intellectual property protection, and privacy. National legislation must be reviewed to ensure it supports the governance and regulation of digital tools, including AI and ML systems. Intellectual property protections, such as patents, copyrights, and trade secrets, should be considered to safeguard proprietary interests. Additionally, privacy concerns related to EM footage, such as images of crew members, need to be addressed through anonymisation techniques like blurring or redaction.
Strategic Engagement with AI
To fully realise the benefits of AI and ML technologies, FFA members must adopt a strategic approach to their engagement with AI. This strategy should:
Define the level of investment and ownership desired for AI adoption.
Align AI initiatives with national and regional fisheries management objectives.
Address regulatory components, including privacy, confidentiality, and intellectual property.
Develop a standardised language for AI and ML engagement.
Leverage existing national strategic plans and regional policies.
Explore funding opportunities and partnerships for scalable solutions.
Establish performance standards for AI and ML systems.
Identify employment and socio-economic opportunities linked to AI adoption.
Recommendations
The report offers the following recommendations to assist FFA members in creating a regional annotated image library and interacting with AI technologies:
Data Annotation: Focus on producing high-quality, reliable datasets through investment in skilled personnel, tools, and processes.
EM Program Objectives: Clearly define EM program goals and the specific elements AI models should support.
Strategic Partnerships: Explore funding and partnerships for cloud-based solutions and AI development.
Regulatory Frameworks: Review and modernise national legislation to facilitate innovation and technology adoption.
Common Language: Develop a standardised technical vocabulary to improve communication and engagement with technology providers.
Legal Protections: Seek specialised legal advice to protect intellectual property and data sovereignty.
Strategic Plan: Develop a regional strategic approach to AI engagement, considering investment, governance, and long-term objectives.
Conclusion
Establishing a regional annotated image library greatly benefits FFA members, improving EM programme efficiency, accuracy, and scalability. However, its success depends on addressing key challenges such as funding, technological capability, data governance, and legislative barriers.
A strategic approach to AI engagement will position the Pacific region to grasp short-term opportunities while preparing for long-term advancements in AI and ML technologies. By pooling resources and encouraging collaboration, FFA members can enhance regional efficiencies and reinforce their leadership in global fisheries management.