Designing With Uncertainty: How AI Supercharges Probabilistic Thinking

Designing With Uncertainty: How AI Supercharges Probabilistic Thinking In 2024, an Air Canada customer asked a chatbot about bereavement fares. The bot confidently gave him a refund policy that didn’t exist. The airline refused to honor it. A tribunal ruled in the customer’s favor. The bot hadn’t decided anything; it had predicted an answer based on patterns in its training data. The company treated that prediction as policy. This is the risk at the heart of designing with AI today: probabilistic systems wrapped in deterministic interfaces. The AI offers a guess, the interface presents it as truth, and the user, or the organization, acts on it. Humans are wired for deterministic thinking. We prefer to believe that past actions determine future outcomes. Flip a coin 999 times and get heads every time, the deterministic mind assumes the coin is rigged. The probabilistic mind accepts that the 1000th flip could still go either way. That second mindset is harder to hold onto, but it is exactly what designers need right now. Products operate in complex, nonlinear environments, and AI is accelerating that complexity. When designers and product teams treat AI outputs as the answer rather than one of many possible answers, they build fragile experiences, and in some cases, like medical diagnostics or financial forecasting, genuinely dangerous ones. This article is a practical guide to designing probabilistically with AI as a partner. It is about using AI to sharpen your thinking rather than outsource it, accounting for model bias, human sentiment, and perceived risk along the way. Probabilistic Thinking + AI Most questions we ask AI do not produce binary answers. They produce probabilities based on patterns in data. If you ask, “Do aliens exist?” the answer will be somewhere between plausible and uncertain. Scientists consider life elsewhere in the universe likely, but without any concrete evidence, we cannot confirm it. The answer doesn’t resolve the question; it frames it as a probability. Designers should read AI outputs the same way. They are signals, not conclusions, possible outcomes that have to be interpreted within the context of product goals, user behavior, and business constraints. Many digital products already work this way. Netflix doesn’t know you’ll enjoy Superstore because you watched The Office; it estimates the probability and surfaces the title accordingly. The interface is responding to a prediction. Design decisions can follow the same logic. AI models can combine behavioral analytics with research insights to estimate the likelihood of certain outcomes, and those probabilities can act as a yardstick for design strategy. Consider a scenario where analytics suggest a 60% versus 90% confidence that users will complete a purchase. At 60%, the design has to do more persuasive work, testimonials, explanations, comparisons, and reassurance signals may help the user move toward a decision. At 90%, the user is already motivated, and the design should start removing friction so the action can happen quickly. Same screen, very different design problem. AI can also simulate outcomes using historical data and behavioral models before you commit to a direction. The value of those simulations depends heavily on how prompts are structured, the context they define, the hypothesis being tested, user motivation, and the edge cases you want stressed. I can think of one such practical use: evaluating early designs through structured prompts, especially when you don’t have direct access to the user group you’re designing for. The prompt below is a starting point for evaluating a design from the perspective of neurodivergent users as well. Treat it as a template, adapt the user group, criteria, and output format to your product, and use it as a conversation starter with your team rather than a verdict. Evaluate the [design file or weblink] for usability, accessibility, and content relevance from the perspective of neurodivergent users such as those with autism spectrum disorder, ADHD, learning disabilities, etc. Please consider the following criteria: - Is the layout and navigation intuitive for neurodivergent users? - Is the language and content appropriate and engaging for neurodivergent users? - Are there any barriers (technical, cognitive, or sensory) that this group might face when using the site? - How well does the site meet the specific needs or goals of neurodivergent users? Provide a SWOT analysis, probability score for successful use by neurodivergent users, and any recommendations for improvement. Note: This is an oversimplification of the idea. Please be mindful of the intricate details of your product and make any appropriate changes. That said, simulations do not replace experimentation. Because models are trained on historical data, they reflect past behavior more strongly than they predict future change. Imagine designing a voice interface for elderly users who struggle with touchscreens. A model trained on mobile interaction data might predict low engagement, not because the idea lacks value, but because the dataset reflects different user behavior. Simulations should always surface assumptions, not prevent innovation. Be Cautious of Skewed Probabilistic Thinking Using AI AI systems are built on historical data, more specifically, on the datasets they are trained on. That foundation shapes the outputs we receive. During the AI Summit in France, India’s Prime Minister Narendra Modi shared an example that illustrates this well. If you ask an AI model to generate an image of a person writing with the left hand, the output may still show a person writing with their right hand. The reason is statistical: most people are right-handed, and the training data reflects that. This may have improved over time, but the point remains relevant. I still occasionally see this behavior when generating images with similar models. What you receive is not truth. It is the most statistically likely outcome given the data available. Always ask whether past data meaningfully predicts future behavior. If additional context can improve the prediction, include it. Without context, the output is just one of many possible answers dressed up as the only one. Confidence scores deserve the same scrutiny. Overtrusting a high-confidence output leads to the Air Canada situation. Dismissing a low-confidence one can cause teams to miss a real signal buried in noisy data. A prediction with 90% confidence is not necessarily correct, and a 40% signal is not necessarily useless. Designers must still weigh the possibilities, consider the case in front of them, and bring judgment to what the AI recommends. Transparency is how you make that possible. As AI systems increasingly shape decisions, people need visibility into how outputs are generated, the sources, the reasoning, and the summaries behind a recommendation. Black-box systems breed distrust. Systems that reveal their reasoning let users evaluate outputs for themselves. That transparency is good design and ethical practice. It respects the trust people place in these tools. Thinking in probabilities often means resisting the temptation of quick answers. AI can accelerate research and surface patterns faster than ever before, but those outputs are starting points, not final decisions. Practice Probabilistic Design with AI Design shapes how a product is ultimately experienced — the decisions designers make determine whether the experience feels adequate, intuitive, or exceptional. And design is inherently full of assumptions and bets. Even the most rigorous research can yield multiple valid solutions to the same problem, each carrying a different probability of success. Thinking probabilistically means recognizing that design decisions rarely produce binary outcomes. They lead to a range of possible results, and the role of the designer is to navigate those possibilities and identify the path most likely to create value. This mindset also builds adaptability: user needs evolve, strategies change, and sometimes ideas fail. Teams that lean on data signals, experimentation, and learning loops move faster toward the most effective solution. Before the practical principles, one fundamental idea: Design decisions should be optimized for likelihood, not certainty. “ Design for Likelihood, Not Certainty Every design decision is a bet, not a guarantee. Even when decisions are informed by research and data, they are still based on smaller samples and assumptions about how users will behave at scale. A well-researched idea can still fail in the real world. The Air Canada chatbot from the introduction is a design lesson as much as a legal one. The bot was doing what language models do, predicting plausible text. The interface, however, communicated that prediction with complete confidence, no caveats, no “here’s what our policy usually says,” no obvious path to a human. The user read confidence as commitment, and legally, so did the tribunal. This is what happens when probabilistic systems are wrapped in deterministic interfaces. The interface transforms likelihood into certainty, and that is where the risk emerges. Designing for likelihood means letting the interface continue to have uncertainty, visible fallbacks to human support, and clear labeling when content is AI-produced, preventing unforeseen issues. Designers should avoid binary thinking — a great idea does not mean guaranteed success, and a familiar idea is not guaranteed to fail. Examine variations, confidence levels, and edge cases instead. AI can certainly help here, acting as a portfolio-thinking engine that surfaces different interpretations, highlights risks, and generates structured recommendations. The goal is not to optimize for certainty, but for value: it should always be value-driven. Think of the moment in Avengers: Infinity War when Doctor Strange tells Tony Stark that out of millions of possible futures, there is only one w