When I’m writing something important, I am often overcome with desperation.
Desperation to get the ideas out of my head and into words on paper before I lose them. It is like a tsunami’s worth of water that I am forcing through a tunnel the width of a dime. My brain thinks of the ideas faster than I can get them into words. I need to finish before something interrupts me and breaks my flow.
As long as I can remember, I have felt limited by my medium. By how fast I can write things down. By how fast I can type. By my vocabulary. By the translucency of my ideas as I think of them.
The anxiety that I feel in writing comes from a fear that the words won’t be up to the task of communicating who I am. When I finally do get the bottled-up words down on paper, I worry what you’ll think of them. Which means that I worry what you’ll think of me. Guessing is nerve-wracking and lonely.
When it works, it is beautiful. When I am perfectly in flow, the words fall from my fingertips. It’s like listening to the perfect song on repeat, like completing a marathon I’m well-trained for on a beautiful day. What can feel so desperate instead feels light and effortless. Finding words that exactly capture my essence is freeing and powerful.
This has never been far from my mind at Textio. In inventing augmented writing, we set out to build a new medium for writing that effortlessly expresses you the way you are. A medium that connects you to your own ideas and to others around you. A medium that is less limiting, desperate, and lonely.
Augmented writing is a learning loop for words.
Learning loops are not a new concept. Every time you drive somewhere with Waze or Google Maps, you’re part of a learning loop. The concept is simple: You provide your car’s whereabouts, and the software combines your data with the data from everyone else on the road to get everyone to their destination faster.
It is a virtuous system that benefits everyone who participates in it. As with all learning loops, when its community gets large enough, the loop becomes so strong that people outside it are at a disadvantage. If I’m driving on my own without Waze, for instance, I can’t guess that there’s a traffic accident a mile ahead of me and I’d better get off the highway at the next exit.
How does a learning loop for words work? When you write, you’re part of a community whose words, choices, and reactions all change the way you express your ideas. The learning loop learns from how you write, and from how others write, and next time the language it creates works better for everyone.
Just as I can’t know where the traffic accident is ahead of me on the highway, I can’t know on my own how you’re going to receive my words. All I can do is guess. I’m often left with the nagging anxiety that I haven’t found the right words to express myself the way I intend.
Whether you’re driving or writing, learning loop software relies on three essential capabilities: prediction, suggestion, and creation.
At its core, prediction relies on data collection at scale. Waze works because so many people use it. When we’re all providing our whereabouts as we drive, the learning loop is able to build a complete picture of what’s happening on the road right now. You and I may be driving to different destinations, but using all that data, Waze is able to predict how long it will take both of us to drive home.
The learning loop underneath augmented writing works the same way. When we’re all writing and keeping track of who responds, the learning loop can build a complete picture of how different language is resonating with real people right now. You and I may have very different ideas, values, and selves to express, but using all that data, the learning loop is able to predict who will respond to each of us.
In both cases, the bigger the community, the better the predictions for everyone inside the loop. More data points = more confident predictions.
Static writing software doesn’t attempt to make predictions. Products like Microsoft Word and Google Docs provide a basic writing experience that lets you document your ideas. Like driving with a paper atlas, it’s better than nothing.
It’s one thing to tell me how long it will take me to drive home. It’s a much more valuable thing to make suggestions while I’m driving that help me get there faster.
When a learning loop is big enough, the community is able to put the loop to work not just to predict outcomes, but to improve them. Driving software powered by learning loops can see the traffic backup that’s ahead and suggest that I turn off the highway now.
The learning loop is adaptive to the changing conditions on the road, which the software is constantly monitoring via my broader driving community. It makes new route suggestions while I’m driving and updates the prediction for my ETA accordingly. Meanwhile, I’m sending my coordinates back to the learning loop to make things better for others on the road.
Augmented writing software works the same way. It’s one thing to predict who will respond to my writing. It’s much more powerful to tell me what language I need to change to better express my ideas and get a stronger response. As I write, the learning loop edits.
Of course, sometimes when I’m driving, I don’t take Waze’s suggestions. Maybe it’s too hard for me to cut across lanes to get off the highway at the last minute. Maybe I’d rather take a route I’m more familiar with even if it takes a little longer to arrive. The software learns from my decisions. If enough drivers in the community make the same decisions, the learning loop changes its suggestions for next time.
Sometimes when I’m writing, I disregard a suggestion too. Maybe I like the way my own words sound. Maybe I see the suggested language and it gives me a completely different idea of what I want to say and I erase the whole sentence and start over. Here too, the software learns from my decisions. If enough writers in the community make the same decisions, the learning loop changes its suggestions for next time.
In both cases, the suggestions I get depend on my context. When Waze tells me to turn left, the suggestion is informed by what’s happening on the path to my particular destination. When augmented writing software tells me to change a phrase, the suggestion is informed by what I in particular am trying to say.
Static writing software includes elements that look like suggestions, but they are backed by hard-coded rules rather than by dynamic and contextual outcomes. Plug-in browser tools for grammar-checking, along with conventional word processors, provide spell-checking and grammar support that help you put your commas in the right place.
These tools are like the speedometer on your dashboard: one-size-fits all. They provide information that helps you follow basic rules. They do not make suggestions that substantially change your outcome.
When I begin a new driving trip, the first thing I do is tell Waze where I’m going. It uses my starting point and my destination to create a set of driving directions for me to follow.
I give the learning loop an idea of my driving directions, and it uses what it knows about current road conditions to create a good route. It’s not the only possible route, but it’s the route that the learning loop believes is optimal at the time based on where I’ve said I want to go.
The more specific I am about my intentions, the better the learning loop can do for me. If I tell Waze that I am driving to Seattle, I’ll get a set of directions that take me somewhere within city limits. If instead I specify that I am heading to “505 Madison Street, Seattle, WA,” I will get a full set of driving directions that drops me off at the front door.
When I’m writing with a learning loop, I jot my ideas down on the writing canvas, telling the software where I’m headed. Just as driving software uses its knowledge of the roads to create a good route to my intended destination, augmented writing software uses its knowledge of current language outcomes to create the words that express my ideas.
The more specifically I can indicate what I mean to say, the more effective the learning loop is at creating the words that say it. Whether I’m driving or writing, the learning loop adapts to how I respond to the path it has created. Sometimes I use part of the path but not all.
I may take different surface streets to reach the highway than the learning loop provides, simply because I am not in a hurry and I like the view better on the route I’ve chosen. When I reach the highway, I’ll join back up with the route that the software has created.
Similarly, I may edit the language that the learning loop generates, taking some but not all of the content into my writing. Maybe I like the first sentence, but want to change the second.
In both cases, the software learns from my decisions. If enough other people make similar decisions, the learning loop creates something different next time.
Autocomplete technology includes elements that appear a little like language creation, but it works more like search. Autocomplete pulls static strings from a very limited dictionary to save me typing time, for instance proposing words like morning and night after I’ve typed the word good.
Whether you’re using it inside Google Smart Compose or on your iPhone, traditional autocomplete helps you say what you were already going to say a little bit faster. It doesn’t help you know what to say in the first place.
It’s like taking a short-cut on your morning commute that you’ve used a thousand times. It’s a handy hack for that one place you go all the time, but it doesn’t help you drive anywhere else.
Learning loop software and its community members collaborate to make an outcome stronger than either could get on their own.
Sometimes when I’m driving, I follow the intended route, only making adjustments when I need to swerve to avoid something immediately in front of me. Other times, I drive where I want, and the software edits the route in front of me accordingly.
Sometimes when I’m writing, I use the language that the augmented word processor creates based on my idea, just tweaking a word here and there. At other times, I form my own words more intentionally, and the software makes small suggestions as I go.
A powerful augmented writing experience relies on its community of writers. Even when I am writing something in a room by myself, I am not alone. And just as with driving software, the bigger the community powering the learning loop, the more painful it is to be outside of it. When I write within a learning loop, I have a writing experience backed by an enormous data set that draws from and propagates the best language of every person contributing to it.
The suggestions that I get as I write aren’t about making me “sound better,” but about driving the specific response I’m looking for from the audience I’m trying to reach. The language that’s created on top of my rough notes isn’t just mindless autocomplete, but a distillation of the true essence of my idea.
When we built Textio Flow, it was with all of this in mind. A learning loop for words has to keep you in flow.
Waze takes away my anxiety about navigating to somewhere new by creating a good route for me to follow. It works with me as I drive, showing my predicted arrival time and making small route suggestions as I go.
Textio Flow takes away my anxiety about having to come up with the perfect words to express my ideas by proactively creating alternatives. It works with me as I write, predicting the likely response to my words and making new suggestions as I go.
Most of all, it is less lonely. I’m confident that my words express who I am.