INDUSTRY CALL TO ACTION - BREAK THE BIAS: Diversity and Equality in Maritime Workforce Representation in Gen AI Image Creation Models

Understanding Bias in Generative Adversarial Networks (GANs)

The term "Generative Adversarial Networks" (GANs) refers to the class of AI models designed for image creation. These models are widely known for their capability to generate new data samples, particularly images, by learning from a dataset of real images. GAN's can unintentionally perpetuate biases present in the data it was trained on. These biases manifest in various ways, including gender stereotypes, physical appearance assumptions, and cultural misrepresentations.

In the case of Maritime, this has resulted in the outdated notion that all seafarers are male with beards and ignores the significant contributions of both women and the diverse appearances of modern maritime professionals. Such biases can skew public perception and hinder progress toward a more inclusive industry.

REQUEST FOR ASSISTANCE FROM THE GAN DEVELOPMENT COMMUNITY

The Maritime Sector: A Pillar of Global Trade and the Essential Role of Seafarers

The maritime sector is a cornerstone of global trade and economic stability, serving as the primary conduit for international commerce. This industry is responsible for the transportation of approximately 90% of the world's goods, including essential commodities like oil, natural gas, food, and manufactured products. The efficiency and reliability of maritime logistics are crucial for sustaining the global supply chain and facilitating international trade.

Central to the maritime sector's success is the seafarer community, whose expertise and dedication ensure the safe and efficient operation of vessels across the world's oceans. Seafarers play a vital role in navigating complex maritime routes, maintaining ship systems, and handling cargo, often under challenging conditions. Their contributions are essential not only for the economic vitality of nations but also for ensuring the smooth flow of goods that underpin daily life. Recognising the importance and identity of seafarers is paramount, as their efforts directly impact global commerce, economic growth, and the overall connectivity of the world.

Fair Representation of the Seafarer in GAN Models

We are all striving for continuous improvement and inclusivity in our respective fields.

The accurate and inclusive representation of all professions is crucial, and the maritime industry is no exception. It's time to challenge outdated stereotypes, ensuring that seafarers are represented with equality, diversity, and inclusivity in the training data for GAN's.

Roles and Responsibilities: The notion that senior maritime-related professions are a "man's world" is somewhat dated. More and more women are occupying senior roles in the maritime sector. By proactively placing emphasis on women in senior positions, captains, chief engineers, and other leadership posts in the maritime industry will help us to address the existing imbalance. Portray gender balance in the ranks of all maritime professions from junior to senior levels.

Gender Representation: The stereotype that all seafarers are male is outdated and inaccurate. Women serve as seafarers and in maritime professions across the maritime sector. Responsibilities are wide and varied with women holding senior positions as Master Mariners, Chief Engineers, Admirals and business leaders.

Physical Appearance: A stereotype is that all seafarers are supposed to have beards. Seafarers are like any other professional group, they come in all sorts of appearances and grooming styles.

Cultural and ethnic backgrounds: Portray seafarers from their various cultural and ethnic backgrounds to reflect the incredible industry diversity.

Uniforms and Clothing: Seafarers' uniform and clothing choices can be diverse depending upon their rank, position, the maritime sector, and regional fashions. It is important to represent this diversity. Use current and contemporary uniform designs that reflect current maritime standards including the use of appropriate Personal Protective Equipment (PPE) when working.

Cultural and Regional Diversity: Maritime professionals are worldwide citizens, contributing to the cultural and regional diversity that defines the industry. It is essential to represent this diversity.

Some Sample Descriptions:

1. A Diverse Maritime Crew: "An inclusive crew of seafarers, men and women from diverse ethnic backgrounds, presented to be working together aboard a modern ship. Crewmembers wore different uniforms and outfits suitable for the kind of work being undertaken and included both clean-shaven and different styles of facial hair."
2. Female Master on the Bridge: "A female master, standing at a ship's bridge, in uniform. In the image, she is issuing commands to the crew—standing officers and engineers—mixed men and women of different ages."
3. Chief Engineer on Duty: "A female chief engineer at work in the machinery room, wearing overalls and safety gear. The scene showcases her managing the team, a man and women group as diverse as the engineering world."
4. Seafarers in Different Positions: "Seafarers shown navigating, maintaining, and cooking on board the ship. People on board include men and women of different ethnicities, depicting the true picture of diversity on ships."

Including such diverse and inclusive representation, will aid in destroying stereotypes and give the industry a more accurate representation.

Join SeaBot Maritime in championing accurate and fair representation of seafarers in GAN Model learning data.

Please help by sharing this information on LinkedIn and on other social media platforms, and by including the following hashtags in your posts. Thank you.

#EqualSeas #BreakGenAIBias #GenAIEqualityMaritime #GenAIDiversityMaritime #GenAIInclusityMaritime

LEARNING OPPORTUNITY:  Are you a Maritime Professional and want to learn more about the world of Generative AI (GEN AI) and Generative Adversarial Networks (GANs)?

Read on to find out more information about GAN's and understand unintentionally perpetuate biases present in training data.        

Generative Adversarial Networks (GANs)

As outlined above, the term used to describe generative AI models that create images is known as "Generative Adversarial Networks" (GANs). These models are widely known for their capability to generate new data samples, particularly images, by learning from a dataset of real images.

In addition to GANs, other terms and model types are also commonly associated with generative AI for image creation:

1. Diffusion Models: These models generate images by iteratively denoising a random noise sample, an example being Stable Diffusion.
2. Text-to-Image Models: These models specifically generate images from textual descriptions. Examples include DALL-E 2 and DALL-E 3 by OpenAI, and Imagen by Google Research.
3. Variational Autoencoders (VAEs): While less common for high-fidelity image generation compared to GANs and diffusion models, VAEs are another type of generative model used to create images by learning the distribution of the input data and generating new data points from this distribution.

These terms and models represent the forefront of generative AI techniques used to create high-quality and realistic images. Several examples of these models are in everyday use such as DALL-E 2 and DALL-E 3 by OpenAI; Stable Diffusion by Stability AI; MidJourney by MidJourney; Imagen by Google Research.

Understanding Bias in GAN's

GAN's can unintentionally perpetuate biases present in the data it was trained on. These biases manifest in various ways, including gender stereotypes, physical appearance assumptions, and cultural misrepresentations.

In the case of Maritime, this has resulted in the outdated notion that all seafarers are male with beards and ignores the significant contributions of both women and the diverse appearances of modern maritime professionals. Such biases can skew public perception and hinder progress toward a more inclusive industry.

Types of Bias in AI Systems

Algorithmic Bias: Flawed algorithm design or implementation can lead to biased outcomes, even if the training data is neutral.

Sample Bias: Insufficient or non-representative training data results in skewed outcomes, favoring some groups over others.

Prejudice Bias: AI systems learn and reinforce societal biases and stereotypes present in the training data.

Measurement Bias: Inaccurate or unbalanced data collection methods introduce errors and biases.

Exclusion Bias: Systematic exclusion of certain data points or groups leads to a lack of representation in AI decision-making.

Selection Bias: Using too little data or unrepresentative data for training can produce skewed results.

Recall Bias: Variations in human labeling or categorization of training data cause inconsistencies and unreliable training data.

About SeaBot Maritime

SeaBot Maritime is a specialist consultancy and learning solutions provider for the maritime industry. Our organisation is dedicated to advancing the sector through innovative technology integration and comprehensive training programs. We create and deliver inventive learning experiences, utilising both digital and experiential methods, to equip maritime professionals with the knowledge and skills necessary for success in a rapidly evolving industry.

Our offerings include tailored training courses, curriculum design, and online learning development, all aimed at supporting the effective adoption of new technologies while maintaining a strong emphasis on the human component of maritime operations. SeaBot Maritime is also involved in the development of standards and best practices for Maritime Autonomous Surface Ships (MASS), playing a crucial role in shaping the future of maritime operations.

For more information, you can visit our official website.