Air Canada lost a tribunal ruling in 2024 because its chatbot invented a bereavement fare refund policy that did not exist. The bot did not decide anything. It predicted an answer from training data patterns. The airline treated that prediction as policy. This is the core failure mode of AI in product design today: probabilistic systems wrapped in deterministic interfaces. The AI guesses. The interface presents certainty. The user or organization acts accordingly.
The article by Smashing Magazine makes a specific, practical case for designing with probability rather than against it. Netflix does not know you will enjoy Superstore because you watched The Office. It estimates likelihood and surfaces the title. Design decisions can follow the same logic. A 60% purchase-completion confidence requires persuasion mechanics: testimonials, comparisons, reassurance signals. A 90% confidence requires friction removal. Same screen, completely different design problem. The piece also includes a structured prompt template for evaluating designs against neurodivergent user criteria, producing a SWOT analysis and a probability score for successful use. That alone is worth the read.
The bias section is where the argument gets rigorous. AI outputs are not truth. They are the most statistically likely outcome given available training data. Prime Minister Modi cited this at the AI Summit in France: ask a model to generate an image of someone writing left-handed and it often defaults to right-handed because the training data skews that way. High confidence scores are not accuracy guarantees. A 40% signal is not automatically noise. The article argues that transparency in how outputs are generated is both good design and ethical practice, and that black-box systems erode the trust they depend on.
[READ ORIGINAL →]