TJ's Insights

As Artificial Intelligence (AI) use in daily life has only become more prevalent since generative AI tools (e.g. ChatGPT) rose to fame, it is more crucial than ever to stay aware of and address the bias present in AI systems.

Because AI models are trained based on prejudiced information (the training data includes subjective information from human opinions and values), there is a significant bias in many models. AI systems can sometimes have discrimination towards LGBTQ+ individuals, people of certain races and genders, people with disabilities, and other typically marginalized groups. 

If the issue of notable bias isn’t addressed, systems can turn out to be inaccurate, harmful, and/or make decisions based on false prejudiced beliefs, possibly creating significant negative impacts, depending on what the system is being used for. Common examples in recent months include cases where the use of AI-powered facial recognition technology by law enforcement has led to cases of mistaken identity, or those where students have been falsely accused of generating work using AI.

It is important that the general population is aware of the bias in AI models, especially those who use AI-generation and AI-powered tools. By not being aware, there is a risk of people carrying these false beliefs with them. Further research must be done, to understand more of what can potentially cause bias, and how it can be addressed without severely declining a model’s efficiency. While bias is inevitable and can not be entirely removed in AI systems, as it would impact model performance to the point where it is inefficient, work can be done to greatly reduce the amount of bias and prejudice in AI. By addressing biases, AI models can be further improved so as to not reinforce stereotypes.