Why Less Does More When You’re Editing with AI
By Daniel Heuman, CEO and founder
There’s a trick we used to design Draftsmith. This trick is what makes the AI give such extraordinarily good results. It has nothing to do with data or training or anything like that. In fact, this trick is so simple that you won’t believe it at first.
Here it is.
We put the user in control of everything that matters.
It doesn’t feel that way. Draftsmith feels magical because you simply press a button and a suggestion for your text appears. You press one more button and it’s inserted into your text. It feels like the thought has been taken out of the process. In fact, the opposite is true.
Draftsmith puts a very tight bound on what the AI is asked to do. The key bits of thinking are with the user. Here’s how we did that.
The person in the box
When you think about the task you’re asking an AI to do, it helps to picture it as a person in a box. As Tom Waddell, Chief Engineer, describes:
“With people, we’ve packaged up the different tasks and put them all into one big box with a job title (an editor). But that box is far too big to give to an AI tool. What if we break that role back down and start looking at the specific tasks and workflows? When thinking about what to get AI to do, it’s important to break the workflow down into small, manageable tasks before you apply the “person in a box” concept. You need to focus on a specific task (or set of tasks) that a human would previously have done.”
Taking the person-in-a-box approach to an extreme, we can break down the task the AI plays in an individual edit. To understand that, it’s important to understand what Draftsmith does. (Skip the next section if you’ve already used it.)
What Draftsmith does
Draftsmith is an AI writing refiner. It helps with the stages in between a first draft and the final copyedit, as you redraft and revise your text. It runs inside Microsoft Word as a seamless part of your process. You simply click a Suggester button in Draftsmith when you want an improvement to a sentence. Almost instantly, the AI suggests a replacement for the sentence. If you like the suggestion, you click Replace.
Here’s how that looks.
Figure 1: Draftsmith runs seamlessly in Microsoft Word
[Insert picture of Draftsmith in Word here]
Figure 2: Three of Draftsmith’s suggester buttons
[Insert picture of Draftsmith and put a red ring/box around the suggester buttons]
The effect is magical. Where previously you might have spent minutes thinking about an edit, the work is almost instant. So how can we claim that the key thinking is with the user?
The workflow of editing a single sentence
When you think about workflows, normally it’s for wider things like the production of an entire document. However, we can do a workflow of a single edit. And in this case, it’s illuminating. If we break down an edit with Draftsmith, here’s what happens:
The user picks a sentence that could be improved.
The user identifies what could improve a sentence.
The user presses a Draftsmith Suggester button to get a suggestion that should solve the issue that the user wants to correct.
Draftsmith comes up with a suggestion.
The user reviews the suggestion to see if it solves the problem.
If the suggestion solves the problem, the user accepts it.
If the suggestion does not solve the problem, the user modifies or rejects it.
When you break down the task, the key bits of control are with the user. The user decides what sentence to improve. The user decides how to improve it. The user decides if the suggestion is right for the intended audience.
If we apply the person-in-a-box framework to the work that Draftsmith does, the AI’s task consists of just two things:
Coming up with a suggestion in response to a specific request for how a sentence should be improved.
Typing the words of that suggestion.
The reason why the AI produces such good results is because the task for the AI is small. It’s not creating entire articles. It’s working within a much shorter space. Moreover, the key change has been chosen for it. The AI doesn’t need to figure out how the sentence should be improved. It knows what the user wants and just needs to find text that has a probability of fitting that requirement.
The challenge for users
When the AI has a small, bounded task like improving a sentence, it can produce superb results. However, it’s easy to confuse the AI. Draftsmith has an option to press one Suggester button in order to turn a sentence in the passive tense into an active one. However, if the user doesn’t understand what passive and active are, and the user asks an AI to fix a sentence that is already active, the AI won’t produce sensible results. If you imagine the person in the box in that situation, they are being given confusing instructions. So of course the results aren’t good.
A tool for professionals
Draftsmith’s results seem incredible. However, the trick is that it gives so much control to users. The AI is good. But it’s not in control. It’s the intelligence of the users that makes the tool effective. You can try it for yourself for free at https://draftsmith.ai.