There has been increasing media and public attention to Artificial Intelligence(AI) based products. AI advocates and proponents espouse the opportunities that AI presents to solve major problems in healthcare and medicine, education, finance, human resource management, marketing, predictive maintenance, and other industries. Despite its promises and advances, there are still major challenges to AI. If you follow technology news, it is ripe with many stories of AI failures including wrong recommendations, sexism, discrimination, and questions regarding ethics and privacy.
While the engineers work to improve algorithms to provide better products and governments and regulatory bodies work on the ethical and privacy ethical issues, we as designers can do our part to design better AI-based products as the technology becomes more ubiquitous. We face unique challenges whether we are designing completely new AI-based products or introducing AI into some aspects of an existing product portfolio and we must prepare for these challenges.
As a starting point, major industry players such as Google, IBM, Amazon, Deloitte, McKinsey, and Microsoft provide guidelines on AI design. These design guidelines synthesize several design disciplines, including voice user interface design, interaction design, visual design, motion design, audio design, and UX writing. They provide useful guidelines and frameworks for the development of trusted AI systems. Whilst these guidelines are very valuable, the ideas are a little dispersed and there is a need for a unified body of work to unify the various learnings that these industry leaders have individually developed. In this series of articles, I will address some of the important considerations when designing AI-based products. In this first article, we briefly look at what I consider the top design considerations for designers namely: Explainability, Transparency, Graceful failure mechanisms, Feedback loops, and Customization and issues about Privacy, Ethics, and Biases.
Making sense of what algorithms output and conveying this in a format that end-users of Ai driven applications can use. Designers of AI products need to familiarize themselves with techniques and approaches known as Explainable Artificial Intelligence(XAI) that enable users to understand the products, build trust, and effectively interact with AI-based products.
When designing for AI, it is important that the capabilities of the system are transparently conveyed to the user. In some critical systems such as health, driverless vehicles, space exploration, and medical laboratory analysis, lack of transparency can when discovered will destroy confidence in your product and might have legal consequences as well. It is also important to address issues like trust, fairness, and discrimination. On the other hand, increasing transparency increases the attack surfaces for malicious actors leading to the so-called “transparency paradox”. Designers are thus faced with finding a balance in this regard.
It is important to set expectations for users to know what errors they might encounter when using AI-based products, designers must do a proper mapping of possible error sources in the product and then provide avenues for users to recover when they encounter errors. For instance, when a bad recommendation is made in recommender systems, avenues must be provided for recovery. Deciding when and how to gracefully fail and recover is also a design challenge that needs to be delicately balanced.
Unlike traditional Software products, AI-based products need user feedback to continuously learn and improve the underlying models. If you design for products where AI is used for very personalized experiences then you should consider allowing users to control and customize their experiences to be effective. Designers have the challenge of finding non-invasive, privacy-preserving methods of data collection especially in recommender based systems like e-health, e-learning, e-commerce systems allowing for allowing user feedback to improve the models and thus improve personalized health, learning and shopping experiences.
While most of the issues related to Privacy, Ethics, and Biases can only be addressed at the level of the underlying technologies themselves, better algorithms and models, training data better, designers can do their part as well. This is perhaps the biggest hurdle of AI-based product design as major products have been suspended recently due to these concerns. Designers should be transparent as much as possible about how data is collected, processed, and stored and be in general more thoughtful about biases and ethical use technology.