In her new book, "Unmasking AI: My Mission to Protect What Is Human in a World of Machines," computer scientist Joy Buolamwini unveils the alarming biases present in facial recognition software. Buolamwini, while working on a project at MIT, discovered that the software failed to recognize her dark skin and only detected her presence when she wore a white mask. This revelation led her to coin the term "coded gaze," highlighting the power imbalances and biases embedded within technological systems.

Buolamwini's research exposed the social implications of biases in facial analysis systems. She warns that these biases not only harm millions of people but also reinforce existing stereotypes. One example is a test conducted on Stable Diffusion's text-to-image generative AI system, where prompts for high-paying jobs predominantly generated images of men with lighter skin. Conversely, prompts for criminal stereotypes produced images of men with darker skin, reinforcing harmful societal perceptions.

The computer scientist emphasizes that AI systems must be critically examined, questioning their purpose and actual functionality. She believes that everyone, regardless of their technical background, has the right and responsibility to engage in the conversation about AI and its potential impacts.

Buolamwini delves into the reasons behind the misidentification and misgendering of individuals by facial recognition software. These errors stem from the flawed methods used to train these systems. Machine learning, which relies on large datasets to teach AI systems patterns, often utilizes datasets that lack diversity. Buolamwini refers to these datasets as "pale male" datasets, as they predominantly feature lighter-skinned individuals and males. Consequently, this lack of representation in training data leads to higher misidentification rates for individuals with darker skin, resulting in real-world consequences such as false arrests.

The biases in gender classification are also explored in Buolamwini's book. She discovered that older women, in particular, were more frequently misgendered by AI systems. The composition of gender classification datasets often comprises images of celebrities who conform to specific beauty norms and stereotypes. This limited and biased representation fails to accurately reflect the diverse spectrum of women, perpetuating gender biases and stereotypes.

Despite initially wanting to focus on building innovative technology, Buolamwini recognized the urgent need to address the misuses of AI. She combines her technical expertise with artistic expression, using poetry and AI audits to evoke emotion and humanize the often abstract conversation around biases in AI systems. Her work resonated unexpectedly and received widespread attention, even reaching the EU Global Tech panel, where it was shown to defense ministers to emphasize the stakes involved in AI development.

Buolamwini's book also calls upon President Biden to prioritize the protection of biometric rights and encourage the U.S. to lead in preventing AI harms. She highlights the need for a comprehensive approach that includes safeguards against algorithmic discrimination, effectiveness of AI systems, and human oversight to protect civil and human rights.

In conclusion, Buolamwini's research and advocacy shed light on the biases present in AI facial recognition systems. She urges society to be critical, questioning the adoption and implications of AI in various sectors. With her powerful message, she advocates for the protection of human dignity and the recognition of diverse perspectives in the development of AI technologies.