A Positive Guide to AI: Putting People First Part 1

This guide is here to show how we can embrace all the amazing things AI can do, while making sure we put people first, always.

AI IN BUSINESS

Konrad Pazik

6/28/202516 min read

Feeling Both Excited and Worried About AI? That’s Okay!

Let's be honest, the arrival of Artificial Intelligence (AI) feels like a huge deal. It’s a bit like the invention of the internet or the smartphone – it’s changing everything! On one hand, it’s incredibly exciting. Experts think AI could add trillions of pounds to the world economy, making businesses more productive than ever before. It’s no wonder that almost every company is planning to invest more in AI. But on the other hand, it’s also making a lot of people feel worried. And that’s completely normal. This guide is here to show how we can embrace all the amazing things AI can do, while making sure we put people first, always.

It’s important to acknowledge the worry that’s out there. A huge number of us, around 89% of workers, are concerned about what AI means for our jobs. Many of us (43%) already know someone who has lost their job because of new technology. These feelings are real and they’re growing. Nearly half of us are more worried about AI than we were a year ago, and many feel it’s all happening a bit too fast.

This "AI Anxiety" isn't just about jobs disappearing. People are also worried about work becoming less human (20%), or that AI could be used to watch us or misuse our data (17%). There's also a fear that it could get in the way of our own creativity and judgement. A big worry for many (63%) is that AI could be unfair when it comes to things like hiring or promotions. Sometimes, when AI seems too human, it can feel a bit strange and unsettling, making it hard to trust.

For anyone leading a team or a business, ignoring these worries would be a big mistake. When people are anxious, they might not feel committed to their work, which can lead to them quietly quitting or looking for other jobs. This means the very technology that was meant to make things better could end up creating a stressful and less productive workplace.

Often, there’s a big gap between how leaders and employees see AI. Leaders are excited about the opportunities and the competitive edge it can bring. But for employees, it can feel like a direct threat to their careers and what they’re good at. Leaders see a cool new tool; employees see a risk. This guide is designed to be a bridge across that gap. It’s a friendly playbook for getting everyone on the same page, calming those anxieties, and making sure that when we use AI, it truly helps everyone succeed and grow, together.

The Foundation – Building AI

More Than Just Rules: Making AI Ethics Our Superpower

When new technology like AI comes along, it’s easy to get caught up in just following the rules to avoid getting into trouble. Many companies are looking at new regulations like the EU’s AI Act and thinking, "What’s the minimum we have to do?". But treating ethics like a tick-box exercise is a missed opportunity. It keeps you on the back foot, always reacting to the next new law, and it doesn’t do much to help your employees or customers feel safe and valued.

A much better, and frankly, more powerful way forward is to build AI based on your organisation’s values. This means shifting the question from "What are we forced to do?" to "What is the right thing to do?". When you decide on a set of core values for how you’ll use AI—and you really stick to them—you’re not just managing risk. You’re building something much more valuable: trust. And in the age of AI, trust is what will set you apart and make you successful.

So, what would these values look like? We can get great ideas from global experts like UNESCO and the OECD. Here are four core values that can act as a strong foundation:

Respect for People, Fairness, and Dignity: This is all about making sure AI is used in a way that respects everyone and promotes fairness, not discrimination. This directly tackles the fear that 63% of people have about AI being biased in hiring. It’s a promise to make sure AI helps create equal opportunities for everyone.
Being Open and Easy to Understand (Transparency and Explainability): This is a big one for all the major ethical guides. It means people should know when they’re talking to an AI, and if an AI makes a decision about them, they should be able to get a simple, clear explanation why. This is the key to building trust.
Being Strong, Safe, and Secure: AI systems need to be dependable and safe. This value, which is a focus for the OECD and UNESCO, is about making sure AI works properly, doesn’t cause accidental harm, and is protected from being used in bad ways, like for spying on people.
People are in Charge (Accountability and Human Oversight): This is the most important rule of all. No matter how smart an AI gets, a human must always be responsible for what it does. This means we need to have people who can check the AI’s work, and who can step in and take control when needed.

Ultimately, the difference between just following rules and truly living your values shows up in your success. Just following rules is about avoiding fines. But building AI on strong values is about building trust, and trust is what makes customers stay with you and talented people want to work for you. When people see you’re serious about using AI ethically—for example, by providing great training on AI ethics, which 80% of employees would appreciate—they will trust you more. That trust leads to happier employees, happier customers, and a brilliant reputation. So, investing in doing AI right isn’t just an expense; it’s an investment in your future.

Opening the Mystery Box: What ‘Transparency’ and ‘Explainability’ Really Mean

When we talk about making AI less of a mystery, you’ll often hear the words "transparency" and "explainability." It’s easy to think they mean the same thing, but they’re actually different, and understanding that difference is key to building trust. They’re for different people and solve different problems.

AI Transparency is all about answering the question: How was this AI built and how does it work in general? It’s like looking at the blueprint of a house. It involves sharing the technical details about the AI, the data it was trained on, and what it’s designed to do. The main audience for this is technical people, like regulators or the company’s own tech teams, who need to check that everything is working properly and safely. Transparency is mostly about being accountable and following the rules.

AI Explainability (or XAI), on the other hand, answers a much more personal question: Why did this AI make this specific decision about me? It’s about getting a clear, simple reason for an outcome that affects you personally. The audience here is the everyday user—the customer who was turned down for a loan, or the employee whose application was filtered out. Explainability is all about fairness and building trust. If you’re turned down for something, you deserve to know why, so you can see if the decision was fair and what you might do differently next time.

So, how do we put these into practice?

To be transparent, a company should:

Keep Great Records: This means carefully documenting where the training data came from and how it was used. Following standards like ISO/IEC 42001:2023 can provide a good guide for this.
Create ‘Nutrition Labels’ for AI: Just like food has a label with ingredients and nutritional information, AI models can have "Model Cards." These are simple summaries that explain what the AI is for, how well it works, and any limitations it has.
Publish Reports: As we’ll talk about later, publishing regular reports about how the company uses AI is a great way to show you’re committed to being open.

To be explainable, a company can use a few clever techniques:

Use Simpler AI: For some tasks, you can use AI models that are naturally easy to understand, like a simple decision tree (which works like a flowchart).
Use Special Tools for Complex AI: For the really complex "mystery box" AIs, there are special tools that can help explain their decisions. Two popular ones are:

LIME (Local Interpretable Model-Agnostic Explanations): This tool helps to explain one single decision at a time. It basically shines a spotlight on a specific outcome and says, "For this one result, these were the most important factors."
SHAP (SHapley Additive exPlanations): This tool gives a more complete picture. For any decision, it gives each piece of information (like age, income, or credit history) a score, showing exactly how much it pushed the final decision one way or the other.

Imagine an AI that helps a bank decide on loan applications. The bank’s regulators need transparency, so they’ll want to see the technical "Model Card" to check if the system is fair and compliant. But the person whose loan was denied needs explainability. They need a simple, clear reason, like, "Your application was denied because your debt-to-income ratio was too high." Giving them the technical document would just be confusing. An organisation that understands this difference can communicate in the right way with everyone, building trust at every level.

A Proactive Defence Against Unfair AI

Let's talk about something really important: AI being unfair, or "biased." This isn’t some rare glitch that might happen by accident. If we’re not careful, it’s almost guaranteed to happen. AI learns from the data we give it, and if that data has unfairness from the real world baked into it, the AI will learn that unfairness too. That’s why we can’t just wait for it to go wrong and then react. We need a proactive plan to build fairness in from the very start.

Unfairness can creep in at a few different stages:

Biased Data: This is the most common culprit. If an AI is trained on data that doesn’t represent everyone equally, or reflects old-fashioned prejudices, the AI will become biased. A famous example is an experimental hiring tool that learned to be biased against women because it was trained on historical hiring data from a time when mostly men were hired.
Biased Algorithms: Sometimes, even with perfect data, the way the AI is designed can create unfairness. For example, an algorithm might learn to use a proxy for something like race (like postcodes) and make biased decisions, even if race itself isn’t a factor it’s looking at.
Human Bias: In many cases, AI learns from data that has been labelled by people. But people have their own unconscious biases, and these can get passed on to the AI, which then treats them as the "truth."

Because unfairness can come from different places, our plan to fight it needs to have multiple layers, happening before, during, and after the AI is trained.

Here’s a playbook:

Before Training (The Data Stage): This is our first line of defence.

Get Diverse Data: We need to make a real effort to collect data that truly represents all the people the AI will affect. We should also check the data for any imbalances from the start.
Balance the Data: If we find imbalances, we can use clever techniques to fix them. We can make sure all groups are represented equally (stratified sampling), give more weight to smaller groups (oversampling), or even create new, artificial data points to help balance things out (data augmentation).

During Training (The Algorithm Stage): We can build fairness right into the AI’s learning process.

Use Fairness-Aware Algorithms: There are special algorithms designed to aim for both accuracy and fairness at the same time. They learn to make trade-offs to avoid unfair outcomes.
Adversarial Debiasing: This is a really smart technique. It’s like having two AIs. One tries to do its job (like making a hiring decision), while the second one tries to guess a protected characteristic (like gender) from the first AI’s decision. The first AI is then trained to make it as hard as possible for the second AI to guess correctly, which forces it to make decisions that are not based on that characteristic.

After Training (The Output Stage): This is our final check.

Calibrate the Results: We can adjust the AI’s decision thresholds for different groups to make sure the outcomes are fair for everyone.
Have a Human Reviewer: For important decisions, we can have a person check the AI’s output before it affects anyone. This is a vital safety net.

To make all of this work, we need good tools and a great team. There are fantastic tools available (like IBM AI Fairness 360 or Google's What-If Tool) that can help us measure and spot unfairness. We also need to decide what "fair" means to us by using clear fairness metrics. And most importantly, we need diverse teams. A team of people with different backgrounds and life experiences is much more likely to spot potential blind spots and challenge assumptions than a team where everyone is the same.

Ultimately, we need to think of fairness not as a one-time fix, but as an ongoing commitment. An AI model can change over time as it sees new data, so we need to build a kind of organisational "immune system." This means regular check-ups (audits), keeping an eye on its vital signs (fairness metrics), and having human oversight ready to step in. This way, we can ensure our AI is, and always remains, a force for good.