An Ode to the Keyboard

Rethinking Authorship in the AI-era

The Keyboard

Cursive Launch Webinar: You're invited. 10/18 @ 3pm EST

Images from Dall E: “standard keyboard where vines are sprouting from between the keys as if it’s being overgrown”

The qwerty keyboard has been around since 18661, invented to uniformly and efficiently share information and data; it became the device of choice for interacting with computers once punch cards went out of vogue in the 1960s. Its connection to a modern computer, envisioned first by Vannevar Bush’s seminal essay, As We May Think2, introduced an efficient human-computer interface that was lossless and immediately editable. It ensured that mistakes could be remedied in real time and coding could occur in human-readable form (or at least engineer-readable). 

Today, the written word can be conjured without a keyboard or much human “direct entry.” A prompt to any emerging generative AI chatbot sites or clicking “The Button3” (as described by Ethan Mollick in his recent post on the addition of a button to create text in Google Docs) can easily create new material in about as much time as you read these last 12 words

For many educators, the new reality has created some worry and a familiar refrain: did a human write this, or did AI write it? 


The first mainstream tool to answer that question was GPTzero, created on the Yale campus and launched on X (at that time, still Twitter) in Jan ‘23. It is an AI classifier, a tool that reviews text to detect the humanness of text through mathematics. The science behind it is fascinating and a sneak peek at the future of writing analytics. Classifiers look to the text to ask some amazingly advanced mathematical questions: what was the perplexity? What is the burstiness?4  

Since then, classifiers have proliferated, and the major incumbent plagiarism detection companies released rivals to the GPTzero. Even OpenAI launched one since shuttered for low accuracy in July ‘235. While classifiers can identify AI-generated text, they are not 100% accurate and can be gamed. Leading researchers at the University of Maryland suggest that any success will be shortlived as generative AI advances6

Unfortunately, what looked like a silver bullet for the AI era has been largely proven to be another arrow in the quiver of forensics tools (plagiarism detection, proctoring, etc.) that teachers and administrators use to investigate student conduct cases. 

Shifting the Focus to Effort and Process

Using AI-generated text as a stand-in for student submissions is only the evolution of a problem that has afflicted education for decades (centuries?): contract cheating. 

Most people are at least familiar with the term “essay mill,” a subset of contract cheating in which students buy a paper on any subject for any grade level, written by an ‘expert’ and submit it to their professor (or upload it to plagiarism detection). The value of an essay mill was 1) to save time and 2) to bypass integrity tools by ordering up a bespoke, original piece of writing. The need to contract a person is weakened when ChatGPT can do it in seconds.

In either case–to combat the use of essay mills and contract cheating or to mitigate the threat of generative AI writing for submission–it is clear that the final submission’s value in verifying academic integrity is weakened. 

A better way is to focus on the effort and process of writing, which is where authentication and authorship begin. By focusing on the writing process, we can have 100% accuracy in answering the question: did a human write this?

The physical manipulation of a keyboard is easily verifiable and opens the door to many benefits. It creates a corpus of data that describes the activity of writing on a computer (and online). Each interaction adds to a stream of data that can look like this: 

{  “clientId”: “”,  “code”: “KeyC”,  “courseId”: “ENG101”,  “event”: “keydown”,  “eventId”: “Quiz”,  “key”: “c”,  “keyCode”: 67,  “personId”: “UUID”,  “resourceId”: “Why Writing is Important”,  “unixTimestamp”: “1694012181028”}
A sample of the anonymized data captured via Cursive

We can leverage this type of data to provide accurate calculations of time on task, use of editing keys and key combinations, information about the editing process overall, writing efficiency, and much more. In addition, the typing cadence and characters keyed provide the basis for verifying the same author over time, session to session, using advanced Machine Learning and AI, which we set out to do at Cursive in ‘22. 

It’s important to note that “keylogging” has been used maliciously and is not without privacy, security, and ethical concerns. Academic Integrity tools like proctoring and plagiarism detection have similarly raised privacy, security, and ethical concerns. These considerations are at the forefront of our conversations with teachers, administrators, and students as we navigate their use in verifying writing authorship. They are the basis for design decisions and policies to provide transparency to all parties. 

The verification of genuine human effort is one goal of education. There are many ways to verify that students can think critically, can create, and have learned. Even though we are 150 years removed from the invention of the modern keyboard, it can still provide the most efficient way to do that. 

In Closing

The keyboard is an amazing device. Through it, we are barely scratching the surface of new insights and information we can glean from the writing process which can be used to inform and improve writing and editing habits for students. In my opinion, authorship trumps humanness and originality, but each plays an important role in verifying the learning process.

To beat contract cheating and fraud in academics we will need to take account of the tools at our disposal, listen to our stakeholders, and adopt new approaches. Let’s get to it.

  1. ↩︎
  2. ↩︎
  3. ↩︎
  4. ↩︎
  5. ↩︎
  6. ↩︎
Joseph Thibault Avatar

Posted by

Leave a Reply