How to design good AI applications › SIDE online

Digital products that work with the help of artificial intelligence make new demands on UX and UI – and above all on strategic design. We present processes and methods

Finnish design and technology consultancy Futurice has released an Intelligence Augmentation Toolkit that makes it easier to navigate the complex world of artificial intelligence. This includes these cards

Most of us are now aware that artificial intelligence follows us everywhere in everyday life – from the search engine to the route planner to streaming offers. The smart little helpers calculate statistical probabilities, make predictions, make recommendations and support decision-making. Unlike regular digital products, they are not “finished” when released because they continue to evolve using machine learning, learn from user behavior, and sometimes explicitly re-educate and further develop.

The smart apps represent a paradigm shift in the human relationship with technology: the applications are no longer just executing services, but also actors with a certain degree of autonomy. They don’t just obey, they “think” along. We become all the more disappointed when something doesn’t work anyway. In addition, there is the problem that with complex neural networks it is no longer possible to trace how they arrived at a certain result. So there are far more uncertainties that require flexibility in design and post-release monitoring.

Backs of Futurice's Intelligence Augmentation Toolkit card

AI tool challenges: trust, control, ethics

AI has enormous growth potential in various sectors, be it marketing, industry or medicine. According to IBM’s 2022 Global AI Adaptation Index, 35 percent of companies are already using artificial intelligence, 44 percent are working to integrate it into existing products, and 42 percent are exploring its use. Particular caution must be exercised, especially when the application is not about music or film recommendations, but about recruitment, production processes or the choice of medicine. It is therefore no wonder that the management consulting firm McKinsey in its study “State of AI in 2021” came to the conclusion that Design thinking in the development of AI tools will be one of the key differentiators for high performers in the AI ​​space.

For designers, intelligent systems represent a new design material that they may (or may not) incorporate into their work. This gives rise to new challenges. Since the further development of the products is somewhat unpredictable, designers have to think more flexibly, taking into account the complexity of the systems and considering unintended consequences. In addition, many people still benefit from and there the credibility of artificial intelligence must be convinced.

Nadia Piet, Head of Creative Technology at Dept in Amsterdam and founder of the AIxDesign initiative, has taken a deep dive into the design of AI applications and lists several areas to consider. Key points are trust and transparency: How to help users understand the results? How do you set realistic expectations? How do you respond correctly to mistakes and take responsibility for them? Another area concerns user autonomy and control: Can users provide feedback? Can they customize their experience themselves? And what about privacy and data security? Piet’s third point is value matching, i.e. the comparison of values: How do you translate user needs into parameters? How to avoid bias and ensure inclusiveness? Are unintended consequences taken into account?

AI: Mental models and expectation management

User research plays an even more important role in such projects than it already does. Because it must also be about people’s mental models. That computer programs can learn new things is still unknown to many and difficult to understand. At the same time, they have high expectations for the algorithms’ performance. It is therefore important to find out exactly what these expectations are, whether the technology can meet them and how to convey the message if it cannot.

Leave a Comment