[AI Ethics Seminar 2025 EP.2] Beyond Bias, A Journey Toward Fair AI - LG AI Research Blog

As AI technology becomes deeply integrated into daily life, responsible development and use are becoming as important as technological innovation. LG AI Research explores ethical challenges through an AI Ethics Seminar that brings together diverse members across the entire AI lifecycle, including researchers, business developers, and UI/UX designers.

The core topic of this seminar is AI Bias. AI bias refers to a phenomenon in which biases present in human society are reflected in training data or algorithms, causing outcomes to be skewed in a particular direction. This goes beyond a simple ethical issue and gives rise to real social problems that disadvantage specific groups.

Real-world examples of AI bias include the following:

• Bias in facial recognition accuracy: A phenomenon where error rates are much higher for certain races (e.g., women of color) than for white men
• Reflection of occupational bias: Image generation models reproducing social stereotypes by depicting high-income occupations with lighter skin tones and low-income occupations with darker skin tones
• Discriminatory language use: Generating discriminatory expressions containing stereotypes about gender or race

Because there is a strong belief that AI is inherently neutral, the risks are even greater when such bias occurs. Therefore, it is essential to clearly recognize that AI outputs can be distorted just like human judgment, and to accurately assess their impact.