AI for accessible design
We recently gave my 85 year old grandmother an iPad. With more and more of our family chatter moving online and relatives spread across the globe, we thought that bringing her into the digital world would be a great way to maintain contact. Besides, evidence suggests geriatrics who are more cognitively engaged and socially embedded experience less decline as they age.
This experiment did not go well. Spend even an hour with an octogenarian and it’s clear that digital design is failing the elderly. For instance: the cool greys and minimal whites of “flat design” are dreadful if your eyesight is fading. Have issues with your neck or back? The angle at which the iPad sits on the table requiring you to crane over to see it won’t be fun.
Furthermore, the token accessibility options that do exist are hilariously bad. iOS includes a “zoom button” that you can choose to display on screen. If you press and hold it, you will zoom on the section of the screen that the button is currently in. The catch? The button itself is so small that unless both your eyesight and manual dexterity are excellent, you’ll never be able to hit it.
People who are the least able stand to gain the most from technology. And yet if you look at standout tech companies of the last few years, they’re not solving problems which help people on the margins: they are devising solutions for technologically adept 15–40 year olds. Overcoming disability has long been a spur to innovation in design; Pellegrino Turri built the first typewriter in 1808 to allow his blind lover to communicate more clearly. Which is why I’m very excited about the potential for AI-first design to make technology more accessible.
The first technology which will prove transformative is voice control.
By creating agents that communicate like humans, we can make it supremely easy for somebody to effortlessly interact with new technology. For my grandmother to benefit from Uber, she currently has to navigate an app via a touchscreen interface she doesn’t really understand. Soon, she will be able to use a format she’s utilized all of her life — a conversation — to order a taxi, enjoy a lower fare, and receive precise updates as to when it will arrive. Similarly, when she gets a Facebook message from a grandchild, she will have it read to her, rather than trying to chase down the notification, find which conversation it pertained to, and struggling to read it on a tiny portion of a screen largely dedicated to adverts.
Decoupling interface innovation from technological innovation — allowing us to have a single mode of interaction with every new technology — is one of the most thrilling promises of AI, and Natural Language Processing (NLP) in particular. And this is a double whammy: we not only make people more able to consume content, but able to produce it. I want to read more Medium articles written by 90 year olds, please.
AI is already making life easier for everybody through the power of prediction.
By understanding what content a user is likely to require “right now”, we can help reduce the volume of information that must be consumed at each step. This is cognitively useful: dealing with a huge number of options is tiring, even for a young adult. Furthermore, prediction allows us to relax our physical requirements for using devices. If you only have to choose between 3 icons, you can make them much larger than if you’re presenting an array of 20.
To their credit, Apple was early to this party: from 2008 onwards, iOS keyboards made use of predictive language models to change the “hit zone” of different keys. So if you’ve typed “hors”, the effective area around the “e” key is enlarged, whilst the one around the “w” key is decreased. Using AI to provide intelligent guesses about what a user is likely to want can thus reduce both cognitive and physical load upon users, making our technologies hugely more accessible.
This is a tremendous opportunity, but it brings with it an obligation: to think carefully about what we want to optimize. We now know that maximizing temptation to drive user engagement can result in runaway cycles of click-bait and fake news. We want to make sure that the way we help people to use their devices is aligned with their personal ambitions, and not merely somebody else’s profit margins.
A third way that AI might be utilized to promote accessibility is in teaching people to use new devices.
Pride is a huge obstacle to adoption. Faced with a complex and confusing new technology, the temptation to withdraw and reject is overwhelming. It’s embarrassing to have to ask your offspring how to use a telephone. By providing assistance which is responsive to the users’ current needs (“You seem to be having trouble with the keyboard; would it help if I made it larger?”), learns with them (“You seem tired now, should I save this email and remind you to finish it later?”), and is infinitely patient, AI might allow us to fulfill the vision promised by that annoying paperclip in Windows 98.
Technology has an undeniable ability to bring us together. But the risk is that new technologies simply create new communities just as exclusive as the old ones. Geography is no longer an obstacle, but technological literacy is. Instead of the family unit, we form digital units; I know more about what my friend in Singapore did on holiday than I know about how my grandmother is recovering from a major operation.
I don’t think this is what we want from our technology, and we now have the tools to tackle these challenges. Bring on Accessible AI-First Design.
