WEBVTT 00:00:00.600 --> 00:00:06.701 Welcome to lesson 4.4, AI ethics, bias, and your legal responsibilities under EU law. 00:00:07.051 --> 00:00:11.638 Using AI responsibly protects your customers, your reputation and you. 00:00:11.988 --> 00:00:15.721 You will learn why ethics matters and how bias creeps into AI. 00:00:16.071 --> 00:00:19.847 You will see how the EU AI Act sorts AI into risk levels. 00:00:20.197 --> 00:00:24.080 You will learn what Article 50 and the GDPR ask of you. 00:00:24.430 --> 00:00:29.315 And you will learn to run a simple ethics audit, which is a careful check of an AI tool. 00:00:30.815 --> 00:00:33.802 By the end of this lesson, you will be able to do four things. 00:00:34.152 --> 00:00:38.589 First, explain where bias in AI comes from, and name four types of bias. 00:00:38.939 --> 00:00:43.547 Second, place an AI tool in the correct risk level under the EU AI Act. 00:00:43.897 --> 00:00:51.172 Third, apply the duties of transparency, data protection under the GDPR and human oversight to your own AI work. 00:00:51.522 --> 00:00:55.426 And fourth, run a five-step ethics audit using a 20-point checklist. 00:00:56.926 --> 00:00:58.526 First, some key terms. 00:00:58.876 --> 00:01:04.188 Bias is when an AI treats some groups unfairly, usually because of skewed training data. 00:01:04.538 --> 00:01:09.231 The EU AI Act is the EU law that regulates AI by level of risk. 00:01:09.581 --> 00:01:15.000 A risk tier is one of the Act's four levels: unacceptable, high, limited or minimal risk. 00:01:15.350 --> 00:01:21.302 Transparency means being open that AI is involved, and labelling AI-generated content. 00:01:21.652 --> 00:01:25.321 The GDPR is the EU law that protects personal data. 00:01:25.671 --> 00:01:29.085 It covers lawful use, minimal data and people's rights. 00:01:29.435 --> 00:01:35.429 And human oversight means keeping a person responsible who can review and override AI decisions. 00:01:36.929 --> 00:01:39.148 Trust is your most valuable asset. 00:01:39.498 --> 00:01:44.085 Using AI responsibly protects your customers, your reputation and you. 00:01:44.435 --> 00:01:47.528 This matters legally and for business, not just morally. 00:01:47.878 --> 00:01:56.539 According to the OECD AI Principles from 2024, fair, transparent AI that respects privacy keeps customers. 00:01:56.889 --> 00:01:59.407 People trust those who use AI carefully. 00:01:59.757 --> 00:02:03.362 But one unfair or careless result can quickly undo that trust. 00:02:03.712 --> 00:02:06.315 And the EU AI Act sets real duties. 00:02:06.665 --> 00:02:10.078 Ignoring them risks penalties, as well as harm to people. 00:02:11.578 --> 00:02:14.309 Bias means an AI treats some groups unfairly. 00:02:14.659 --> 00:02:18.157 It usually comes from the data, not from bad intentions. 00:02:18.507 --> 00:02:20.342 First, it starts in the data. 00:02:20.692 --> 00:02:23.316 AI learns from examples made by humans. 00:02:23.666 --> 00:02:28.829 If those examples reflect unfairness from the past, the AI can learn it and repeat it. 00:02:29.179 --> 00:02:34.341 For example, imagine a hiring AI trained mostly on CVs from one group of people. 00:02:34.691 --> 00:02:39.321 It may quietly score similar candidates higher, and put others at a disadvantage. 00:02:39.671 --> 00:02:42.977 This is dangerous because AI is fast and confident. 00:02:43.327 --> 00:02:47.103 Biased decisions can be made at large scale before anyone notices. 00:02:48.603 --> 00:02:52.742 Bias comes in several forms, and recognising them helps you catch them. 00:02:53.092 --> 00:02:58.617 Data bias is when training data has too many or too few people from some groups, skewing results. 00:02:58.967 --> 00:03:04.450 Sampling bias is when the data was collected in a way that misses part of the real population. 00:03:04.800 --> 00:03:10.624 Historical bias is when the data reflects past inequality, and the AI treats it as normal. 00:03:10.974 --> 00:03:13.022 Automation bias is different. 00:03:13.372 --> 00:03:15.655 It is a human failing, not the AI's. 00:03:16.005 --> 00:03:19.759 It happens when people trust the machine too much and stop questioning it. 00:03:20.109 --> 00:03:22.477 The answer to that is human oversight. 00:03:22.827 --> 00:03:27.115 For every AI output you rely on, ask: who might this treat unfairly? 00:03:28.615 --> 00:03:33.671 The EU AI Act sorts AI by risk: the higher the risk, the stricter the rules. 00:03:34.021 --> 00:03:36.517 Place your project in a tier before you build. 00:03:36.867 --> 00:03:39.129 The top tier is unacceptable risk. 00:03:39.479 --> 00:03:45.025 These are banned, such as government social scoring, or AI that manipulates vulnerable people. 00:03:45.375 --> 00:03:49.173 These bans have applied since 2 February 2025. 00:03:49.523 --> 00:03:54.237 High risk means strict duties, like risk checks, human oversight and records. 00:03:54.587 --> 00:04:00.582 Examples are AI used in hiring, credit scoring, grading or essential public services. 00:04:00.932 --> 00:04:04.025 Limited risk means you must tell users it is AI. 00:04:04.375 --> 00:04:08.685 Examples are customer-service chatbots and AI-generated content. 00:04:09.035 --> 00:04:11.360 Minimal risk has no special rules. 00:04:11.710 --> 00:04:14.334 Think of spam filters or AI in games. 00:04:14.684 --> 00:04:17.969 Your chatbot from lesson 4.1 is limited risk. 00:04:19.469 --> 00:04:24.440 Most tools you build in this course, such as a customer-service chatbot, are limited risk. 00:04:24.790 --> 00:04:27.350 Your main legal duty is transparency. 00:04:27.700 --> 00:04:31.753 The GDPR also applies whenever personal data is involved. 00:04:32.103 --> 00:04:34.834 So say clearly that an AI is responding. 00:04:35.184 --> 00:04:38.768 Label AI-generated content where the law requires it. 00:04:39.118 --> 00:04:42.126 Check outputs for bias, and protect personal data. 00:04:42.476 --> 00:04:43.671 But be careful. 00:04:44.021 --> 00:04:48.565 If your AI helps decide on jobs, credit or grades, it is high risk. 00:04:48.915 --> 00:04:52.605 A CV-screening bot for a client, for example, would be high risk. 00:04:52.955 --> 00:04:55.238 Get expert advice before building one. 00:04:56.738 --> 00:05:00.194 Now, transparency under Article 50 of the AI Act. 00:05:00.544 --> 00:05:03.957 This duty applies from 2 August 2026. 00:05:04.307 --> 00:05:06.419 Simply, be honest about AI. 00:05:06.769 --> 00:05:08.433 First, say it is AI. 00:05:08.783 --> 00:05:14.714 Chatbots and AI assistants must make clear that users are talking to a machine, unless it is obvious. 00:05:15.064 --> 00:05:17.453 Second, label AI-made content. 00:05:17.803 --> 00:05:23.691 Deepfakes, which are fake but realistic images, audio or video, must be disclosed as AI-made. 00:05:24.041 --> 00:05:26.964 So must AI text published to inform the public. 00:05:27.314 --> 00:05:29.789 And AI tools must mark their output. 00:05:30.139 --> 00:05:31.717 Third, why it matters. 00:05:32.067 --> 00:05:35.630 Transparency lets people judge AI output for themselves. 00:05:35.980 --> 00:05:38.476 It is the foundation of trustworthy AI. 00:05:39.976 --> 00:05:41.469 You have four duties. 00:05:41.819 --> 00:05:43.483 First, be transparent. 00:05:43.833 --> 00:05:48.292 Tell people clearly when they deal with AI or AI-generated content. 00:05:48.642 --> 00:05:50.306 Second, check for bias. 00:05:50.656 --> 00:05:54.965 Test that your AI treats different groups fairly, not just fairly on average. 00:05:55.315 --> 00:05:56.915 Third, protect data. 00:05:57.265 --> 00:06:02.663 Collect only what you need, and store personal data securely, as the GDPR requires. 00:06:03.013 --> 00:06:04.826 Fourth, keep a human in the loop. 00:06:05.176 --> 00:06:10.573 A person reviews important AI decisions and can override them before they take effect. 00:06:10.923 --> 00:06:15.169 For high-risk AI, bias checks and human oversight are legal duties. 00:06:15.519 --> 00:06:17.801 For everything else, they are good practice. 00:06:19.301 --> 00:06:23.376 If your AI touches personal data, the GDPR applies. 00:06:23.726 --> 00:06:25.539 First, have a lawful reason. 00:06:25.889 --> 00:06:31.628 You need a valid basis to use someone's data, usually clear consent or a genuine business need. 00:06:31.978 --> 00:06:35.754 Second, collect only what you need, called data minimisation. 00:06:36.104 --> 00:06:38.152 Data you do not hold cannot leak. 00:06:38.502 --> 00:06:40.486 Third, respect people's rights. 00:06:40.836 --> 00:06:44.655 People can ask what data you hold, correct it, or have it deleted. 00:06:45.005 --> 00:06:50.914 Fourth, if AI alone decides something important, like a loan, people can ask for a human to review it. 00:06:51.264 --> 00:06:54.165 That is Article 22 of the GDPR. 00:06:54.515 --> 00:07:01.406 So if a bank's AI refuses a loan, the customer can ask for an explanation and a human review, and can challenge it. 00:07:02.906 --> 00:07:06.639 An ethics audit is a simple, repeatable check in five steps. 00:07:06.989 --> 00:07:12.749 Step one, set the scope: pick one tool, such as your chatbot, and define what it does and who it affects. 00:07:13.099 --> 00:07:14.870 Step two, check the data. 00:07:15.220 --> 00:07:17.567 Is any group missing or shown unfairly? 00:07:17.917 --> 00:07:23.165 Step three, test for bias: try inputs from different groups and look for unfair differences. 00:07:23.515 --> 00:07:27.461 Step four, confirm it says it is AI and protects personal data. 00:07:27.811 --> 00:07:30.051 Step five, log the gaps and fix them. 00:07:30.401 --> 00:07:34.305 Write down what is wrong, plan changes, and check again after fixing. 00:07:34.655 --> 00:07:41.397 In exercise 4.4 you will use a 20-point checklist on your chatbot from exercise 4.1. 00:07:42.897 --> 00:07:48.806 Responsible and irresponsible AI use are separated mostly by honesty, fairness and care. 00:07:49.156 --> 00:07:52.825 Responsible use tells people clearly when AI is involved. 00:07:53.175 --> 00:07:55.949 It tests outputs for bias and fixes problems. 00:07:56.299 --> 00:07:58.560 It collects and protects data carefully. 00:07:58.910 --> 00:08:01.363 And it keeps a human accountable for decisions. 00:08:01.713 --> 00:08:05.020 Irresponsible use hides that AI made the decision. 00:08:05.370 --> 00:08:07.717 It ignores who might be treated unfairly. 00:08:08.067 --> 00:08:10.221 It hoards personal data carelessly. 00:08:10.571 --> 00:08:13.345 And it lets the machine decide with no one checking. 00:08:13.695 --> 00:08:16.319 Aim to be responsible in everything you build. 00:08:17.819 --> 00:08:20.699 Please pause the video and answer these four questions. 00:08:21.049 --> 00:08:24.590 In your own words, where does bias in AI usually come from? 00:08:24.940 --> 00:08:29.079 What are the four risk tiers of the EU AI Act, from highest to lowest? 00:08:29.429 --> 00:08:32.949 What must you disclose under Article 50, and from what date? 00:08:33.299 --> 00:08:36.776 And can you name the five steps of a simple AI ethics audit? 00:08:37.126 --> 00:08:38.854 When you are ready, press play. 00:08:40.354 --> 00:08:42.381 Here are some trusted resources. 00:08:42.731 --> 00:08:46.592 The European Commission has a page on the AI Act and how it works. 00:08:46.942 --> 00:08:52.041 There is also an AI Act explorer, where you can read the full text article by article. 00:08:52.391 --> 00:08:55.313 And you can read the OECD AI Principles. 00:08:55.663 --> 00:08:58.117 You will find all the links on the lesson page. 00:08:59.617 --> 00:09:01.387 Here are the key takeaways. 00:09:01.737 --> 00:09:03.999 Ethics protects people and protects you. 00:09:04.349 --> 00:09:06.887 Trust, fairness and the law all depend on it. 00:09:07.237 --> 00:09:12.059 Bias usually comes from the training data, so test outputs across different groups. 00:09:12.409 --> 00:09:17.529 The EU AI Act ranks AI by risk, and most tools you build are limited risk. 00:09:17.879 --> 00:09:25.089 From 2 August 2026, under Article 50, you must disclose AI and label AI-made content. 00:09:25.439 --> 00:09:31.413 And remember your four duties: be transparent, check for bias, protect data, and keep a human in the loop. 00:09:32.913 --> 00:09:35.409 Now let's put it into practice on the platform. 00:09:35.759 --> 00:09:39.001 First, try the activity called which risk tier, which duty? 00:09:39.351 --> 00:09:43.597 Then do exercise 4.4, where you run an AI ethics audit. 00:09:43.947 --> 00:09:45.589 It takes about one hour. 00:09:45.939 --> 00:09:47.945 Finally, take the lesson quiz. 00:09:48.295 --> 00:09:51.751 It has five questions, and you need 70 per cent to pass. 00:09:52.101 --> 00:09:54.682 You can retake it as many times as you like. 00:09:55.032 --> 00:09:57.144 Well done for finishing the last lesson!