Book Review - Unmasking Al: My Mission to Protect What Is Human in a World of Machines

By Dr. Joy Buolamwini, Ph.D. (MIT), published by Random House

You often hear the plaintive refrain from the fragile who troll social media comment sections that "not everything is about race." Never more so than at the slightest hint that white supremacy is under attack. And it’s true - not everything is about race. When it comes to the decision-making tools used by the government, law enforcement, businesses, insurance brokers, banks, and landlords to make decisions affecting our daily lives, though, we’d better be sure very sure it isn't about race. But we aren't sure, and that’s Unmasking AI’s message.

In Unmasking AI, Dr. Joy Buolamwini describes how a simple master’s project she built using AI to track facial movements in a mirror failed to detect her face unless she wore a white mask. She recounts how, through this, she discovered tech giants had failed to address bias and other ethical issues of AI in facial recognition software before taking their tools to market.

Part memoir, part exposé she explains what ultimately led her to create The Algorithmic Justice League and fight for the ethical use of facial recognition tools. She found that software sold to law enforcement and security companies was routinely misidentifying dark-skinned men and women. She also uncovered the underlying privacy issues surrounding AI face recognition and its training data. She illustrated this by explaining how companies like Facebook used millions of our photos without consent to train their AI.

A Canadian-born Ghanaian woman raised by an academic and an artist, first in Ghana and later in the United States, her perspective is unique. Despite her Fulbright and Rhoads Scholarships and MIT MSc & Ph.D., you sense the ‘otherness’ she feels in the worlds she inhabits. When she considered dropping out of her Ph.D. program, she rhetorically asked if her three academic credentials weren’t enough. She mentions famous tech dropouts like Bill Gates and Mark Zuckerberg but recognizes that women aren’t granted the same latitude where credibility in tech is concerned.

Unmasking AI was interesting because there’s no undertone of a social justice warrior searching for a fight in Dr. Buolamwini. Hers is the story of an incredibly accomplished woman in STEM with unlimited options who found her calling through an accidental discovery. She's a woman who reveals she wasn’t even sure at first if she wanted to make it her mission to right the wrong she’d discovered. So yes, she is a socially minded warrior but not a frivolous or undisciplined one, who has been recognized by both Congress and President Biden for her work.

In today’s misguided social and cultural environment that divides us by race, gender, and a multitude of other thin wedges, we must recognize our own bias to see bias in AI. I was not fully alive to the alternate realities Dr. Buolamwini discusses until my daughter was born over a quarter century ago. I knew then that she would experience the world differently than I did as a man. While there is comfort in egocentricity, I let go of my ego’s safe shore to learn about the world she would face, and it was eye-opening.

Unmasking AI is equally eye-opening.

Dr. Buolamwini explores the coded bias built into AI systems and how they reflect human biases. What struck me profoundly was that AI systems don’t just mirror biases, but they reinforce and amplify them. Although not mentioned in her book, she now describes this effect not as a mirror but as a kaleidoscope because of how AI negatively refracts bias in subtly unexpected ways.

Does the state have any business monitoring its citizens without cause, she asks? It doesn’t, but omnipresent surveillance is coming. It is too tantalizing a technology not to come. Ironically, Dr. Buolamwini’s work to audit and improve facial recognition algorithms will ultimately extend their proficiency – a fact she acknowledges.

Finding systems were fine-tuned on large data sets of mostly white males. She found they were excellent at identifying white adult men but had difficulty detecting darker faces and gendering women; when given a binary choice, they failed 35% of the time.

So what? Consider how a facial recognition system that could only identify white men 65% of the time would be received. But systems that can only identify women of colour 65% of the time? They are available today and have the real-world impact you'd expect.

I trained teams bound for Afghanistan to make lethal strike/no-strike decisions through the lens of an armed drone's camera. AI could easily replace several steps I taught, including positive target identification, collateral damage assessment, and the lawful authority to strike assessment based on the rules of engagement.

I can only imagine how happy commanders would be to have AI’s assurance before ordering a strike. But is 65% enough? I don’t think so. The future use of remote weapons systems guided by flawed AI facial or visual recognition is a concern we all should have.

My own encounter with coded bias was revealing. While searching for images of ‘good-looking couples’ on a popular graphic arts website, it took 45 minutes to realize I’d been exclusively served white people. What startled me was what that said about my own unconscious racial bias.

It wasn’t until that moment that I began to see the tip of the iceberg that is algorithmic bias. People of colour have complained about it for several years. I’d heard their arguments but hadn’t understood them until that moment. Unmasking AI reveals this iceberg in a way the Titanic’s captain couldn't ignore.

Her book contributes significantly to the broader conversation about technology and society. It shatters the notion of AI as a monolithic, unbiased entity. She reminds us that AI is only as good as the data it's trained on.

?We need continued vigilance to monitor the inherent weaknesses in AI systems. Things are changing, but with general resistance to the idea of unconscious bias in the chattering class, let alone the bias AI produces, we can’t assume this will solve itself.

This is where the third-party validation and government regulation Dr. Buolamwini advocates for is crucial. We need an objective understanding of an AI system’s strengths and limitations before deploying them. We need to insist that our technologies are not only equitable and beneficial but that they fundamentally protect the people they are designed to help.

?Dr. Buolamwini is a charming writer and poet who didn’t bog the book down in technical jargon or concepts. Her ideas were well explained and easily accessible to those with a limited understanding of the issues. The timeline shifted more than I liked, but that was only a minor distraction. I read it on a beach in Punta Cana, so to say she held my interest is, I think, self-evident.

Ironically, my readings were immediately reinforced by a practical lesson - the facial recognition scan I was subject to on leaving the Dominican Republic. I wondered, where did my face go? Who will use it? How will they use it? Is it for sale? Will I ever know? as I presented my face to the camera. These are the questions Dr. Buolamwini teaches us to ask.

Unmasking AI is a call to action. It demands that we not, with blind faith, give our unreserved trust to AI and the companies developing it. It demands that we highly regulate facial recognition and the data supporting it.

As a researcher and engineer, Dr. Buolamwini does not suggest there is anything wrong with progress. She is not a doomer but a realist grounded in fact. As AI advances, she encourages us to let go of our safe shore to insist that the systems increasingly governing our lives do no harm.

Author's Note. I considered calling Dr. Buolamwini "Joy" throughout this review, but it didn't seem right. Her doctorate came at a great personal cost, and she deserves recognition for the sacrifices she made to achieve it. That, and I found myself repeatedly shouting, "No, don't do that!" as I read her thoughts on dropping out, and I was relieved when she didn't (despite knowing full well she didn't); I wanted to honour her in this small way to show the joy [sic] I felt when she was named a Doctor.

Unmasking AI can be found here - https://www.amazon.com/Unmasking-AI-Mission-Protect-Machines/dp/0593241835