Tom Taulli
Vintage AI Artificial Intelligence Businessman Concept Searching or Scan System Problem. AI Artificial Intelligence concept about finding searching or scan system problem. Innovation for IT and technology and business category GETTY
One of the key drivers of the AI (Artificial Intelligence) revolution is open source software. With languages like Python and platforms such as TensorFlow, anybody can create sophisticated models.
Yet this does not mean the applications will be useful. They may wind up doing more harm than good, as we’ve seen with cases involving bias.
But there is something else that often gets overlooked: The user experience. After all, despite the availability of powerful tools and access to cloud-based systems, the fact remains that it is usually data scientists that create the applications, who may not be adept at developing intuitive interfaces. But more and more, it’s non-technical people that are using the technology to achieve tangible business objectives.
In light of this, there has been the emergence of a new category of AI tools called Automated Machine Learning (AutoML). This uses simple workflows and drag-and-drop to create sophisticated models – allowing for the democratization of AI.
The use of low-code platforms can also be useful. "These are ideal to build applications with beautiful interfaces and provide a user-friendly experience," said Paulo Rosado, who is the CEO of Outsystems. "Low-code platforms are also helping to close the skills gap as there is not enough developers in the workforce to fill the growing need among organizations to build apps."
But even these systems require a background in data science and this can pose tricky issues with the development of the UI.
“Our mission when we designed Dataiku was to democratize data and AI across all people and to unite all of the various technology pieces out there,” said Florian Douetteau, who is the CEO of Dataiku. “We kept this mission in mind when we embarked on our UI. Enterprise AI is the future, and that means hundreds and thousands of people are using Dataiku every day as the core of their job, spending hours a day in the tool. So we keep the UI of Dataiku simple, clean, modern, and beautiful; no one wants to work in a space -- virtual or otherwise -- that is cluttered or that looks and feels old, especially when data science and machine learning are such cutting-edge fields. Another important consideration is ease of use, but not at the expense of robustness. That means making sure that Dataiku’s UI is simple for those on the business side -- many of whom are used to working in spreadsheets -- who don’t have extensive training in advanced data science as well as the most code-driven data scientist - but none of this as a tradeoff for deep functionality.”
Yes, it’s a tough balance to strike – but it is critical. So then what are some best practices to help out? Well, here are recommendations from Megan Mann, who is a product manager at Sift.
Focus on meaningful patterns: "When Sift approaches UI/UX we are trying to make the AI invisible to our customers. It’s more about what we hide from users than what we show because if our customers were exposed to too much data, they would quickly become overwhelmed and any meaningful patterns would disappear."
Font matters: "Given that we are tasked with explaining massive amounts of data -- Sift is now processing 35 billion events per month from 7000 fraud signals -- we need to be extra mindful about text, color, and size. These are typical design problems that become much more complicated given the sheer volume of data, signals, results, etc."
Simplify labels: "An UI/UX challenge is to translate the language commonly used by a developer into conversational language that is universally understood by our audience which, for us, is fraud analysts. For example, take an account takeover event. We might see a pattern and label that ‘failed logins in the last hour greater than 10.’ But for our fraud analyst customers, we would surface that as something like ‘urgent abnormal activity.'"
What's more, success with UI/UX is about reaching out to customers and understanding their needs. This has certainly been key for Intuit’s TurboTax, which involves advanced AI systems and algorithms.
“When we went out and asked thousands of consumers about their tax preparation, most responded with emotions of fear, uncertainty and doubt,” said Eunie Kwon, who is the Director of Design at Intuit. “Once we started to unpack their reasons for these feelings, we found opportunities to influence their experience by applying some basic psychological principles and laws of UX heuristics to simplify through mindful design. To reduce cognitive load, we balanced the fundamental elements of design through content, visual expression, animation, and recreated the informational experience to reduce fatigue, friction and confusion. To improve workflow, we dissected the complicated tax forms into adaptable and consumable interview-like experiences. We added ‘breather’ screens where we acknowledge to the customer how much they’ve completed and the accuracy of their input. We also added ‘celebration’ screens to drive confidence that informs them of their progress while educating them on the changes in tax laws along the way.”
Such approaches are simple and make a lot of sense. But when developing software, they may not get much priority.
“The main lesson learned when designing for TurboTax is balancing simplicity while ensuring 100% confidence for a customer’s tax outcome,” said Kwon. “Every year, we are faced with new mindsets that evolve the behavior of how consumers interact with products and apps. The expectations for simplicity and delight change so often that we need to look at our experience and find improvements that meet those expectations, while driving complete confidence through their tax experience.”
Tom serves on the advisory boards of tech startups and can be reached at his site.
No comments:
Post a Comment