The main AI dilemmas are bias and fairness, accountability, transparency, privacy, and automation's effects on human judgement and work. They are prescribed in IB Digital Society topic 3.6E for both SL and HL.
AI outputs reflect training data, model design and objectives chosen by people and organizations. Outcomes emerge from interactions among data, algorithms, institutions and users.
| Dilemma | Mechanism | Impacts and implications |
|---|---|---|
| Bias and fairness | Biased or incomplete training data can produce unequal outputs. Fairness definitions also reflect human choices. | Discrimination can be reproduced at scale. |
| Accountability | Responsibility may be divided among developers, data providers, deploying organizations and users. | Harmed people may not know who must explain, correct or compensate. |
| Transparency | Complex or proprietary models can operate as black boxes. | People may struggle to scrutinize decisions, although disclosure may conflict with security or intellectual property. |
| Privacy | AI can require extensive personal data. | Data can support services but also enable surveillance, misuse or breaches. |
| Human judgement and work | Automation can replace tasks or decisions. | Efficiency may rise, but deskilling, over-reliance and employment displacement may follow. |
A common misconception is that an AI system independently decides what is fair. In reality, people shape its data, objectives, thresholds and deployment context, so algorithms are not neutral.
Amazon's discontinued recruitment model is a real-world example. System: Machine learning ranked applications. Example specifics: In 2018, Amazon reported that the experimental system learned from male-dominated recruitment data and penalized indicators associated with women. Impacts and implications: It was not deployed for hiring, but exposed the risk of reproducing workplace inequality. Concepts: It demonstrates power, values and unintended consequences.
In an IB response, match the command term. For explain, link each dilemma to its cause. For evaluate, weigh benefits and limitations, compare stakeholder perspectives and support the argument with a specific example.
