Authentic Knowledge
2022
Researcher
About the project
It all began with the recognition that bias in AI systems posed a significant ethical challenge. My passion for economics, coupled with my growing interest in AI, prompted me to delve into this issue.
My journey in the realm of artificial intelligence (AI) and bias mitigation was a transformative experience that merged my passion for economics and public policy with a deep commitment to addressing ethical and societal challenges in AI systems. Collaborating with IIT graduates, I embarked on a mission to combat bias in AI and published a research paper, alongside developing a practical tool for AI developers.
I identified various forms of bias, including racial bias, historical bias, and socioeconomic bias, as critical areas to address. The next step was to dive deep into understanding the root causes of bias in AI systems. While human biases and socioeconomic factors were evident contributors, I was determined to explore further. My research led me to the revelation that historical data could significantly influence the biases present in current AI judgments. This comprehensive understanding served as the foundation for developing effective solutions.
My role
The culmination of our efforts resulted in the creation of "Authentic Knowledge" (AK), a powerful tool designed to empower AI developers to identify and mitigate bias in their models. This was no small feat; it required meticulous planning and implementation.
Structured Data
Image-data-driven models
Language models
To address bias in structured data, such as demographic information and historical records, I designed algorithms that could detect and mitigate racial, socioeconomic, and historical biases effectively. Techniques like re-weighting and re-sampling were employed to ensure fairness.
For image-data-driven models, my coding prowess allowed me to craft neural networks with custom loss functions designed to reduce biases associated with race, gender, or other sensitive attributes in image classification tasks.
Bias in language models was equally important to address. I developed sophisticated text-processing algorithms that could identify and mitigate bias in language models. Advanced techniques like adversarial training and fine-tuning with debiasing objectives were incorporated to reduce biases in the output generated by these models.
I devoted countless hours to perfecting the program, ensuring it met the highest standards. This included extensive testing, refining algorithms, and writing a comprehensive research presentation.
This journey to combat bias in AI systems was marked by dedication, innovation, and a deep commitment to making the world a more equitable place. It was a remarkable opportunity to collaborate with college graduates and experts while blending my passion for economics and public policy with cutting-edge technology. Throughout this journey, I acknowledged and credited those who guided and supported me, recognizing that every step was a collective effort. This experience not only reinforced my determination but also highlighted the potential of AI to drive positive change when used ethically and responsibly.