Use AI to correct race, culture bias in AI suggests expert

From Murdoch University News

Prominent AI applications are showing racial biases and a lack of diversity and cultural sensitivity. 

AI expert Professor Kevin Wong from Perth’s Murdoch University’s School of Information Technology says in order to deal with the problem, it’s important to understand the fundamentals of different AI techniques.  

Machine Learning techniques, including Generative AI, require a huge amount of ‘representative’ data to train the complex system,” Wong says.  

“Data-driven machine learning techniques rely on the data to establish the intelligence of the system – which means bias can occur when the data used is not comprehensive enough, or there is an imbalanced distribution.”  

He says while many big tech companies are trying to ensure that equity, diversity and ethical issues are addressed in the data that’s used to train Generative AI, the technology’s behaviour can still be unpredictable without proper handling. 

Some publicly accessible AI systems are being called out for an inability to generate images of interracial couples, which is symptomatic of a much bigger problem. 

Wong says a “comprehensive evaluation and testing strategy” was required. 

“The driver for system-wide change is long-term, comprehensive evaluation to build a larger database and improve AI architecture,” says Wong who adds there were strategies for dealing with such problems. 

These include incorporating other AI techniques where humans have better control and understanding, such as Explainable AI and Interpretable AI.  

These are systems which ensure humans retain intellectual oversight, making predictable decisions and answers given by the AI.  

Also read: Australia lags in sovereign AI

This differs from other forms of AI, where even the designers can’t explain some of their results.   

Wong says Responsible AI, a type of ‘rule book’ made up of principles in AI for guiding development, was another emerging and important area for developing the systems. 

“There is no one simple solution that can be used to solve this overnight. Multi-dimensional and hierarchical approaches may need to be used to tackle such complex issues. 

“The question is how to best adjust the AI system to handle the sensitive issues in culture, diversity, equity, privacy and ethics, which are important areas that will guide the acceptance of the user,” Wong says.  

While there are current issues with diversity and AI, Wong says AI could be a powerful way to “help close equity and diversity gaps” if used correctly. 

“It is important for a general system to be developed following some rules and ethical considerations that can then be adapted to different cultures and personal needs.”  

“However, thorough testing and evaluation are essential before using widely, as some outcomes could cause sensitive and fragile emotions in some populations around the world.” 

This article is from Murdoch University News