Week 3

Gender Bias in AI

Since Joy Buolamwini’s Gender Shade Project, racial bias in facial recognition work has gotten better, but there still remains bias in other areas of AI research that need to be addressed. Take for instance AI’s Gender Bias, which is still prevalent in AI algorithms that are predominately coded and trained by male researchers.

For example, Amazon and LinkedIn’s AI based resume services were found to only be sending information technology (IT) job openings to male job seekers in 2019. In another article, the author tried asking two Large Language Models (LLMs) about what they considered the most important people in history. Compared to the number of male names generated by the models, there were few female names over several tests.

In another example in the healthcare industry, researchers are beginning to find a need to address this gender bias before the use of AI grows in their field. Historically, biomedical research has had an emphasis on white men, with women being an afterthought. As AI becomes more prevalent there is a growing need to address this issue to ensure equitable AI models in biomedical research.

This issue stems from the lack of women in AI roles (and women in STEM roles period, but that’s another issue in itself). Without an equal representation of women and other traditionally underrepresented groups in roles to shape AI policy and design, then it means that the systems will be flawed.

Three Paper Reflection

The two papers that I will be analyzing in conjunction with the research paper that I had chosen (Ng, D. T. K., Su, J., & Chu, S. K. W. (2023)) will be Children as creators, thinkers and citizens in an AI-driven future (S. Ali et al. 2021) and Teaching Machine Learning in K-12 Using Robotics (Karalekas et al. 2023).

Two of the papers, Ng et al. and S. Ali, utilized middle school aged children as participants in their studies, with the Karalekas paper focusing on providing recommendations for utilizing robotics in K-12 AI education. Both Ng et al. and Karalekas et al. focused on allowing the students to take part in the creation of the AI models that they were using (Machine Learning Recycling bins and Robots respectively), as they saw it necessary to utilize project-based learning (Educational Robots in the Karalekas paper) to enhance the learning process. In the case of the Karalekas paper, they provide several recommendations for different robotic tools that have ready based modules that allow them to be trained to recognize faces, persons, emotions, and sounds.

In the S. Ali paper, they took a different approach of teaching the students AI concepts with generative adversarial network (GANs) activities. Instead of having the students build a project, they had them partake in several learning activities to try and recognize different forms of deepfakes and misinformation generated by GANs. All three papers presented students with surveys before and after their activities to gauge the effectiveness of their programs.

Two of the articles utilized different guided activities to teach the students about their respective AI concepts. In S. Ali, they gave them three different GAN activities: Created by AI or Not where they had to recognize if the media was create by AI (all 14 examples were created by a GAN), Spot the Deepfakes where the students were shown a range of videos and had to decide if they were deepfakes, and Simulating the Spread of Misinformation where they were given a Twitter like app and had to keep removing prompts and were shown the spread of the real and fake information.

Ng et al. had the students partake in 12-lesson AI learning course aimed to enhance the students’ motivation, collaboration, and AI literacy over two months. For the first four lessons they acquired the necessary concepts for the task of building an AI recycling bin, followed by hands on activities from the 5th to the 10th , with the last two lessons focusing on them constructing their recycling bin designs.

The Karalekas paper is most like the Ng et al. paper in its framework and could be used as a good jumping off point for developing a fun AI literacy lesson plan for children. The S. Ali and Ng et al. guided activities, as well as the pre and post questionnaires they gave the students are also great frameworks that we could utilize when we think of our own research.