Chapter 3: Who Watches The Machines? Ethical Auditing Explained
3. Who Watches the Machines? Ethical Auditing Explained
As artificial intelligence becomes more involved in our everyday lives — from deciding who gets hired to suggesting what we watch or read — an important question comes up: who’s making sure these systems are doing the right thing?
It’s easy to assume that AI is completely objective, making decisions without any bias or flaws. But the truth is, AI systems learn from us — from our data, our patterns, and sometimes, our past mistakes. If no one is paying attention, these systems can end up reinforcing unfairness, making unsafe choices, or simply getting things wrong.
That’s where ethical auditing comes in. It’s like a thoughtful check-up for AI. Experts take a close look at how these systems are built, what data they’re learning from, how they’re tested, and how they’re used in the real world. But it’s not just about fixing bugs or improving accuracy. Ethical auditing is about making sure AI treats people fairly, avoids harmful outcomes, and truly works for everyone — not just a select few.
As we hand over more decisions to machines, keeping a human lens on their behavior is more important than ever. Ethical audits help us do just that — making sure that as technology grows more powerful, it also stays responsible.
Table of Contents
3.1. When AI Starts Picking Favorites: A Cautionary Tale
Imagine this: a company which decides to speed up its hiring process by using AI to screen job applicants. Sounds smart, right? Faster decisions, less paperwork, and fewer human errors. But there was a catch, guess what? After a while, they noticed something odd. More men were getting shortlisted than women. And not just slightly — it was a noticeable trend.
The company’s internal ethics board investigated and uncovered a startling truth:
The AI was biased. It was favoring male candidates. But it wasn’t doing this on purpose — it was simply doing what it had been trained to do.
Turns out, the AI had been fed historical hiring data, and that data reflected a time when men were more likely to be hired. The system looked at that pattern and assumed it was the “right” one to follow. In other words, the AI had quietly inherited our old biases — and nobody noticed until it became a real problem.
So, how did they fix it?
The company took a thoughtful approach:
They retrained the AI using a more balanced dataset — one that treated male and female candidates equally.
They added fairness rules to the system so it wouldn’t fall into the same trap again.
And most importantly, they brought humans back into the loop — especially for sensitive decisions like who gets hired and who doesn’t.
What’s the bigger lesson here?
AI isn’t neutral. It learns from us — and if we’re not careful, it can repeat (and even amplify) our worst mistakes.
That’s why we need to audit these systems, question their decisions, and make sure there’s always a human sense-checking what the machine says. Because at the end of the day, technology should help us grow, not reinforce the problems we’re trying to fix.
3.2. Expert Roundtable: What the Leaders Are Saying
As artificial intelligence weaves itself more deeply into the fabric of our daily lives, global leaders from tech companies and government agencies are speaking up. Their message is clear: we need smarter, fairer, and more transparent AI and that starts with ethical auditing.
Here’s what some of the key voices are saying, and why it matters.
IBM: Making Ethics Part of the Code
IBM has long been a strong advocate for building trust in AI from the start. Francesca Rossi, the company’s Chief AI Ethics Officer, says that ethics shouldn’t be an afterthought.
For IBM, it’s about building systems that actively promote fairness, not just avoiding mistakes. Their internal ethics board plays a hands-on role in reviewing how AI is used, ensuring each project reflects the company’s values.
For IBM, ethical AI is not just about compliance — it’s about responsibility.
OpenAI: Transparency in Action
While OpenAI frequently highlights the importance of transparency, the reality is more nuanced. The organization acknowledges that the public deserves insight into how AI systems behave—especially as these tools begin to influence education, public discourse, and access to information.
OpenAI has opened some channels for dialogue around AI behavior, fairness, and safety, stressing the importance of both internal and external scrutiny. However, critics argue that the company’s actual practices around transparency sometimes fall short of its stated ideals. Still, its efforts are helping shape conversations around responsible AI development, even as the debate around openness and accountability continues to evolve.
Microsoft: Standards That Guide Every Step
Microsoft has developed a clear framework for what responsible AI should look like, with key principles like fairness, privacy, transparency, and accountability at its core. For every system they build, the company asks tough questions: Is this inclusive? Will it behave as expected in the real world? How do we explain its decisions to users?
These principles aren’t just posters on the wall — they’re baked into how Microsoft builds and tests its AI tools, with ethical auditing embedded throughout.
European Union: Regulation with Purpose
While tech companies are setting standards internally, the EU is leading the charge on regulation. The upcoming AI Act is one of the most ambitious efforts to govern AI use responsibly.
By classifying systems based on their potential risk — from minimal to high — the EU is setting up a framework where the most powerful systems face the most scrutiny. It’s a proactive step to make sure AI respects people’s rights, avoids harm, and builds public trust from the ground up.
Why It All Matters
Ethical auditing isn’t about slowing down innovation or pointing fingers. It’s about building AI that works for everyone ; systems we can rely on, understand, and trust. Across the board, leaders agree: bias must be tested for, transparency must be built in early, accountability must be clear, and no one can tackle this alone.
AI is rewriting the rules of how decisions are made. Ethical auditing is how we make sure those decisions are fair, responsible, and human-centered.
3.3. Checklist: The 12-Point Ethical AI Audit Template
If we’re going to trust artificial intelligence with decisions that shape our lives — like approving loans, reviewing job applications, or even diagnosing illness, we have to be sure it’s not just smart, but also ethical.
Too often, AI systems are designed with performance in mind, but ethics is treated as an afterthought. And when that happens, the consequences can be serious: unfair treatment, security breaches, or decisions no one can explain. That’s where ethical auditing comes in.
This 12-point checklist breaks down the essentials of responsible AI design. Think of it as a guide, not just for data scientists or engineers, but for business leaders, policymakers, and anyone involved in shaping how AI is used.
1. Use accurate, diverse, and clean data
AI is only as good as the data it learns from. If that data is flawed or biased, the system will be too. Take Amazon’s hiring algorithm as an example: it taught itself to prefer male candidates because its training data reflected years of biased hiring practices. Clean, diverse, and representative data across gender, race, age, and geography helps prevent this kind of bias from being baked in.
2. Check regularly for bias or discrimination
Even with good intentions, bias can creep in. That’s why ongoing audits are essential. Bias isn’t static; it can evolve as data changes. Set up a routine for testing how your AI performs across different groups. Disparities in outcomes should be spotted early and addressed quickly.
3. Make decisions transparent and explainable
People have the right to understand how AI affects them. Whether it’s a credit rejection or a denied insurance claim, opaque algorithms are a trust killer. Concepts like Explainable AI (XAI) help unpack black-box models, offering frameworks to make machine logic more interpretable and open to human scrutiny.
4. Assign accountability clearly
When AI makes a mistake, who answers for it? In too many cases, no one knows. That’s a problem. Ethical systems clearly define who is responsible for design, deployment, and outcomes. This isn’t about blame; it’s about being ready to respond if things go wrong.
5. Safeguard user privacy
AI often feeds on sensitive personal data – shopping habits, medical history, location, and more. Respecting privacy means collecting only what’s necessary, storing it securely, and giving users control over how it’s used. Regulators are watching too — just look at the GDPR in Europe, which enforces strict rules around data handling and transparency.
3.4. Regulatory Deep Dive: EU vs. US AI Laws
As artificial intelligence becomes more embedded in daily life, deciding things like who gets a job interview or whether someone qualifies for a loan, governments are facing growing pressure to step in and set clear rules.
Two major players in this space, the European Union and the United States, are approaching the challenge from very different angles. Understanding how each region is handling AI regulation gives us valuable insight into what the future of trustworthy AI might look like.
The European Union: Building a Rule-book for Risk
The EU has taken a more aggressive and structured stance when it comes to AI governance. With the introduction of the EU AI Act, Europe aims to create the world’s first major legal framework for artificial intelligence. And it’s not just talk ; this is real legislation with clear classifications, restrictions, and enforcement plans.
At the heart of the EU’s approach is a risk-based model. AI systems are sorted into categories based on how much harm they could potentially cause. For example, an AI-powered spam filter would be considered low-risk and barely regulated, while a system that helps decide who qualifies for a mortgage or medical treatment would be classified as high-risk and subject to strict oversight.
For these high-risk applications, the rules are clear: developers must prove that their AI systems are transparent, well-documented, and regularly tested for bias or errors. There also has to be human oversight and, in most cases, independent third-party audits before the systems can be launched in the market. Some AI uses, like real-time facial recognition for mass surveillance or social scoring, are outright banned.
What’s striking about the EU’s approach is that it doesn’t just focus on what AI can do, it focuses on what it should do. It’s about trust, accountability, and making sure that as AI grows more powerful, it also remains fair and respectful of human rights.
The United States: Setting the Tone, but Still Taking Shape
Across the Atlantic, the U.S. has taken a more hands-off approach — at least for now. In 2022, the White House introduced the Blueprint for an AI Bill of Rights. It outlines five core principles: systems should be safe and effective, protect people from discrimination, guard their data privacy, be transparent, and provide a human alternative when needed.
It’s a strong values-based foundation, but there’s a catch: it isn’t enforceable. At this point, the blueprint is more of a suggestion than a rule-book. There are no penalties for ignoring it, and no federal laws yet require AI developers to follow these principles.
That said, the regulatory landscape is beginning to shift. Several U.S. states — including California, New York, and Colorado are starting to pass their own AI-related laws, especially around data protection and algorithmic accountability. But without a unified national framework, the U.S. is currently operating with a patchwork of state-level rules and corporate guidelines, which can be confusing for businesses and inconsistent for users.
Critics worry this leaves too many gaps — especially in high-stakes areas like hiring, healthcare, or law enforcement, where algorithmic bias or opaque decision-making can cause real harm.
Looking Ahead: Different Roads, Shared Destination
At this stage, the European Union is clearly ahead in terms of turning ideas into action. The AI Act is not just a set of principles — it’s enforceable law with real consequences for noncompliance. It’s likely to become a global benchmark, much like GDPR did for data privacy.
The U.S., meanwhile, is still laying the groundwork. Its approach emphasizes innovation and flexibility but has yet to fully address the risks that come with unchecked AI development.
Still, both regions are moving and the gap may narrow as pressure grows from citizens, advocacy groups, and even companies themselves. In the end, the goal is the same: to create AI systems we can trust, understand, and benefit from without fear of harm or discrimination.
The challenge now is to build that trust not just with powerful technology, but with thoughtful regulation that keeps people at the center.
3.5. Case Study: AI in Healthcare — From Bias to Fairness
Imagine you’re at a hospital, and your doctor is using AI to help decide who should receive care first.
It sounds efficient—a smart system prioritizing patients based on urgency.
But what happens when the algorithm doesn’t treat everyone equally?
That’s exactly what happened with an algorithm developed by Optum, a subsidiary of UnitedHealth Group, which was widely used by major hospital systems across the United States. A 2019 study revealed that the algorithm significantly underestimated the health needs of Black patients compared to white patients with similar medical conditions. The issue stemmed from using healthcare costs as a stand-in for health needs—a decision that inadvertently reinforced systemic inequalities.
They had introduced an AI tool to identify patients for high-risk care programs, hoping to catch serious conditions early and improve outcomes. But over time, a troubling pattern emerged. The algorithm was prioritizing white patients more often than Black patients — even when those Black patients had more severe medical needs.
So, what went wrong?
The AI had been trained to predict who would benefit most from extra care by looking at how much money had been spent on each patient’s healthcare in the past. It seemed logical — more spending typically means more serious illness, right?
Historically, less money has been spent on Black patients, not because they were healthier, but because of long-standing disparities in access to care, trust in the system, and treatment decisions. The algorithm couldn’t see these social factors — it simply learned from the data it was given. As a result, it underestimated the needs of patients who had historically been underserved.
Once this issue came to light, the developers made a crucial change. They rebuilt the model using direct health data like blood pressure, lab results, and existing medical conditions — rather than relying on financial records as a stand-in for health. This shift led to a much more accurate and equitable system that better reflected who actually needed care, regardless of race or socioeconomic status.
This case offers a powerful lesson. AI systems, no matter how well-intentioned, can mirror the flaws of the society they learn from. In healthcare, those flaws can be life-threatening. That’s why it’s essential to use diverse data, continually test for bias, and remain vigilant at every stage of development.
Ethical AI in healthcare isn’t just about getting the technology right — it’s about making sure it works for everyone.
Final Thoughts: Building AI You Can Trust
AI can be transformative, or it can go terribly wrong. Whether we’re designing recommendation engines or diagnosing diseases, we need AI to be transparent, ethical, and accountable.
The good news? We already have the tools, the frameworks, and the knowledge. Now we need to use them and hold ourselves to higher standards. Because the future of AI isn’t just about power. It’s about trust.
4. Verifiable AI — From Blockchain Anchors to Zero-Knowledge Proofs
We trust our GPS to get us home, our smart assistants to answer our questions, and our AI writing tools to make sense of language, but can we trust how these systems actually work?
As AI models grow more powerful and opaque, a new frontier of transparency is emerging: verifiable AI.
The idea is simple but powerful to prove that an AI system is doing what it claims to do, without needing blind faith. Whether it’s a large language model generating legal advice or an image recognition system screening for medical issues, verifiability ensures that what’s happening under the hood can be checked, traced, and held to account.
What Is a Verifiable Model?
In basic terms, a verifiable AI model is one that can prove its actions, inputs, and outcomes — either to a human or to another system. This means:
You can trace where its data came from
You can confirm which model version was used
You can validate that it wasn’t tampered with after training
And in some cases, you can audit a decision without revealing sensitive inputs — thanks to cryptographic tools
These capabilities are vital in high-stakes areas like healthcare, finance, law, and governance — where decisions need to be transparent, defensible, and accountable.
Contributor:
Nishkam Batta
Editor-in-Chief – HonestAI Magazine
AI consultant – GrayCyan AI Solutions
Nish specializes in helping mid-size American and Canadian companies assess AI gaps and build AI strategies to help accelerate AI adoption. He also helps developing custom AI solutions and models at GrayCyan. Nish runs a program for founders to validate their App ideas and go from concept to buzz-worthy launches with traction, reach, and ROI.