Just over thirty years ago, the first Terminator movie hit the theaters. It was a game changing moving starting starring Arnold Schwarzenegger as the Terminator, a cyborg assassin sent back in time from 2029 to 1984 to kill Sarah Connor, whose son will one day become a savior against machines in a post-apocalyptic future. An artificial intelligence network will become self-aware and initiate a nuclear holocaust.
The storyline seemed far-fetched back in 1984 but today seems surprisingly accurate. With only 12 years left until the singularity event portrayed in the movie, we are far down the path to smarter machines. Elon Musk has been famously warning of an impending disaster. Machines already with access to all the world’s information are starting to learn, change and adjust themselves to the world. While I hope we will not see a future at war with the machines we, as a society, need to start engaging with this new reality.
Recently, the concepts arounds machine learning and artificial intelligence have been elevated to forefront of tech world. Every service we use is getting smarter and smarter. Each time you search for a concert or dogfood, machine learning algorithms are deployed to improve the quality of your results. They are becoming smarter with each question. In addition to text search, the quality of voice and visual search tools are improving exponentially.
I recently took part in a hackathon where teams were tasked with creating a new service with the help of Machine learning. With no experience in machine learning it seemed a daunting task. Our team made up of three developers, a data scientist and myself representing design and product had never used machine learning tools. Time boxing work can be a magical thing and within 72 hours our team took a concept from design to a working prototype.
Our solution was built to solve a challenge of the retail world. Shoppers like search by color and style but brands generally organize products individually by category. Using visual recognition, we built an app that allow retail customers to capture images of head-to-toe looks from social media or their phone. The app then used visual recognition to find styles and colors from our brand that would be a good match. Using images and tags from target products we were able to train our app producing matching results with surprising accuracy. I was floored when hours before our deadline we ran through the demo journey end to end and it worked!
With new tools such as IBM Cloud and AWS Tensorflow, developers can quickly implement advanced machine learning solutions without the complex math used by data scientists. These tools have changed the game for developing intelligence in user experiences. Organizations can quickly deploy these algorithms without needing to understand the underling intelligence layer.
The experience was a bit shocking to me. As someone who with a career building systems intended to improve lives, it is alarming to imagine how easily we can be surrounded by smart machines without a deeper understanding of how they make decisions and how we should manage them.
Designing for the Machine
As user experience designers, how do we arm ourselves for a future with smart machines? As far back in 1959, Arthur Samuel described machine learning as: “the field of study that gives computers the ability to learn without being explicitly programmed.” Adoption of the technology has slowly grown over the year but has recently exploded into every industry. While ML and AI is being applied across every platform, design for human interaction with these systems has a lot of room to catch up.
Call to Action
With such low barrier to entry to use, designers need to take an active role in shaping the experience powered by the machine learning rather than letting the tools define the experience.
The two main categories of machine learning include supervised and unsupervised. Supervised learning requires examples of correct solutions to train the algorithm. For example, when you sort your Gmail and tag some into the spam folder, machine learning algorithms are improved by adding to the data set.
Unsupervised learning, by contrast, uses algorithms to categorized data without knowledge of expected results. Sets of data can be grouped. This can be helpful in analyzing data and identifying cohorts needing further study. For example, groups of consumers can be categorized into cohorts based on their past behavior without intentionally knowing what attributes are important.
Design Best Practices
Challenges arise when you need to get users to take action to improve an algorithm. In the example of an email spam filter, the user, not knowing about the benefits of sorting their mail, deletes the email and loses the chance to have less spam in the future. As a user, the quickest route might seem most expedient but small changes in your interaction might improve your life in the long run. Designers need to consider how to craft messaging and flows to correctly position learning features so that they appear to be in service of the user, not the machine.
The Google Nest has been a leader in the smart home segment. With learning algorithms, Nest promises to reduce costs and improve our comfort. It is a great example of an elegant, streamlined onboarding process. One thing the designers at Nest have done well is explain how interacting with the thermostat will make it smarter. This little incentive to the home owner to make more thoughtful decisions will create better results down the line. More recently, the Nest app has implemented a geo-fencing feature that allows it to set Home/Eco mode based on your location. On iOS users must grant location access to each app which many hesitate to allow. Nest could improve their messaging to explain the benefits of geo fencing and how it will make the Nest smarter and save you money.
As experience designers, we need to develop best practices to ensure we apply machine learning ethically. It is important to communicate with the end user how their interactions will used. With a goal of transparency, we must communicate immediate benefits to the user, longer term benefits to the community and the impacts of data storage over time.