Applying an AI Ethics Framework

An AI ethics framework turns a vague question like "is this good?" into a set of concrete considerations you can weigh a specific AI system against: is it justified, does it promote equality, does it respect human dignity, does it keep a human in control, has it weighed its longer-term consequences? These aren't boxes to tick off; each one calls for judgement about how it applies to your particular case. For an ICT student, this matters because whenever you build a product that uses AI, you're expected to weigh it against considerations like these and show where it holds up and where it doesn't, not just judge it on whether it works or sells.

Starting Points

UNESCO Recommendation on the Ethics of AI the first global standard-setting instrument on AI ethics; gives you the values and principles behind terms like "human oversight" and "proportionality" that turn up in other frameworks.
OECD AI Principles — the first intergovernmental standard on AI; shows how "trustworthy AI" gets defined at the policy level that several national laws build on.
EU Ethics Guidelines for Trustworthy AI the European Commission's framework, with concrete requirements (human agency, transparency, accountability) you can check a real system against.
NIST AI Risk Management Framework a practical framework for building trustworthiness considerations directly into an AI system's design and deployment; useful for seeing these principles applied to an actual product lifecycle rather than to policy.

Key Points

You deliberately pick which principles from a framework actually apply to the AI product you're evaluating, rather than running through all of them mechanically for example, recognizing that "added value compared to a human" is the live question for one product while "human oversight" is the live question for another.
You test claims about your own product's ethics rather than taking them at face value for instance, checking whether a claim like "our data is representative" or "a human stays in the loop" is actually true of your design, instead of asserting it because the framework asks for it.
You separate the core ethical problem in your product from secondary ones, and place it in its wider societal context for instance, recognizing that a recommendation engine's real issue is who it excludes, not how fast it runs, and considering who beyond the immediate user is affected.
You build a convincing case for your product's ethical standing by weighing it against more than one principle at once for example, showing it adds real value and considering who could still be harmed rather than resting your justification on a single criterion.