Design Must Lead the Way on Artificial Intelligence and Machine Learning
I recognize that AI- and ML-driven ad-delivery systems are not at the top of anyone’s mind in the midst of this unprecedented ongoing global pandemic. One effect of the COVID-19 outbreak was the cancellation of the CHI 2020 conference last month. Though I was looking forward to reconnecting with folks from CSCW and the broader HCI community to discuss the following information, I thought I might instead use this forum to share my research about this rapidly developing field. I hope this piece helps you feel a little less isolated and, on some small scale, more connected to your fellow UX designers.
Artificial intelligence and machine learning technology are all around us. They recommend relevant movies and music, help us find the right route home, organize our photos, are the technology behind voice assistants, and filter our feed on Facebook, to name but a handful of applications. These innovations have begun to permeate all areas of our lives.
To date, the areas of investment have been primarily on the highly technical side — building better algorithms. The user experience has, for the most part, taken a backseat to the mechanics. The result is that there is a huge opportunity for the design and research communities to create much better user experiences for AI and ML.
The upsides of understanding these underserved but critically important areas are potentially enormous, but to capitalize on them, we have to start understanding how these technologies are used, as opposed to just building better algorithms.
Promise and potential
The Facebook Ads and Business Platform team, for which I am a research director, is working as hard and as fast as we can to define and design the AI- and ML-powered user experience.
It’s easy to forget just how new this space is, but in truth, we have barely begun. The most thrilling part for those of us who design for businesses is that advertising, by several measures, happens to be the biggest and most mature commercial application of machine learning by several orders of magnitude.
We’ve implemented ML and AI on several of our advertising products, and we’ll continue to run full force toward this space because we know that doing so will pay off for the businesses who use our platform.
Beyond the algorithms
Certainly, technical work is essential to making particular types of strides in creating AI- and ML-powered experiences. But understanding how businesses interact with AI and ML both at a macro level and at a product level is incredibly important.
What we’ve learned so far is that we can and should apply solid and tested user experience and human-computer interaction principles to the AI- and ML-driven delivery systems that choose ads for people, even as we design new ways for them to make use of and interact with these systems.
Simply making ML and AI technically better will not create more meaningful or relevant experiences for our users. It will not solve the real issues they face. To do that, we must understand their everyday business practices and collaborate in these areas. Luckily, this is where user experience professionals thrive.
Prioritizing design and creating best practices
As the need for well-designed AI- and ML-driven user experiences crystallizes, we’ve begun to transform our thinking from designing for outcomes to prioritizing design at the outset.
Cross-functional collaboration among user research, content strategy, engineering and others helps Facebook build products that truly resonate with people, and AI and ML are no different.
As members of our team and several academic colleagues explained and workshopped at the most recent Computer-Supported Cooperative Work and Social Computing Conference (CSCW), both industry and academia are in urgent need of leaders who understand and design for the work that occurs with AI and ML, best practices for design, and the unique methodological approaches to accomplish both.
Thus, as AI and ML move from novelty to standard, we’re naturally learning along the way. Because the field is still very much in formation, we have a massive and exciting opportunity to actually create the standards, define the best practices and develop the protocols for both our industry and our advertisers.
For example, we’ve already learned two seemingly simple but incredibly valuable best practices:
- Highlight when the system is still learning, including any reduced capabilities. This makes it clear to people when the system is still collecting data and how it affects them.
- Be transparent about confidence in estimates. This helps people understand the level of uncertainty behind estimates.
High-wire act for AI and ML in business contexts
Saleema Amershi, an advocate for responsible AI and a well-regarded researcher from Microsoft, recently articulated several potential dangers of consumer applications of AI and ML:
“As automated inferences are typically formed under uncertainty, often producing false positives and false negatives, AI-infused systems may demonstrate unpredictable behaviors that can be disruptive, confusing, offensive and even dangerous.… Attempts at personalization may be delightful when aligned with users’ preferences, but automated filtering and routing can be the source of costly information hiding and actions at odds with user goals and expectations.”
The consequences of misapplying AI and ML in a business setting go far beyond providing an unusual movie or receiving a head-scratching voice assistant response.
At minimum, technology that dynamically updates itself and learns on the fly doesn’t mix well with historical expectations of interface consistency and behavior predictability. And, as Amershi notes, “AI components are inherently inconsistent due to poorly understood, probabilistic behaviors based on nuances of tasks and settings, and because they change via learning over time.”
The potential impact of misapplication in business is high stakes, especially in the areas of manufacturing, health care and financial services, where unintended consequences can disrupt lives.
The road ahead
We don’t know everything about UX for AI- and ML-powered experiences — not even close. But we do know that this technology dramatically shapes the experiences that come out the other side. We know that we want to help the businesses that use Facebook’s ad platform achieve their goals, and that this won’t be accomplished by building better algorithms alone.
We are also aware that in addition to calling on standard techniques, we will need to define new metrics and usability practices. Further, we’ll need to create ways to implement and measure the resulting data.
Finally, we know that we don’t know all the ways ML and AI can lead to unforeseen experiences. So it’s of utmost importance that we make sense of these challenges and opportunities as a collaborative design team, from beginning to end and back around again.
What a tremendous opportunity we have to stake out this ground and design meaningful user experiences for the people who use our systems. What an incredible chance to lead the industry in solving these hard problems.
Thank you to my thought partners in this space: David Randall, Joe Gonzales, Chinmay Karande, Yanling Wang, Serene Lewis, Janice Jung, Chris Langston, Anne Yamanaka and all of my friends from our CSCW Workshop.
