The Ghostwriter typewriter brings generative AI to the printed page | Engadget

Running from 1992 through 1995, Ghostwriter, the beloved PBS children’s television show, followed a diverse group of friends as they solved mysteries around their Brooklyn neighborhood with the help of their haunted typewriter, a cursed item possessed by the trapped soul of a murdered runaway Civil War slave. The Ghostwriter typewriter developed by interaction designer, artist and Lumen.world CTO, Arvind Sanjeev, on the other hand, comes with none of the paranormal hang-ups of its coincidental namesake. Instead of a spirit bound to this hellish plane of existence, forced to help tweens solve low-stakes conundrums, the deus in Sanjeev’s machina is animated by OpenAI’s GPT-3.

He first devised this artistic endeavor in 2021 as a, “poetic intervention that allows us to take a moment to breathe and reflect on this new creative relationship we are forming with machines.” Built over the course of weekends and evenings, Ghostwriter interacts with its user through the written word, allowing the two to converse and co-create freely through the physical medium of paper.

“I wanted Ghostwriter to evoke warm feelings and make people comfortable playing with it,” Sanjeev told Engadget via email. “I chose the mental model of the typewriter for this reason. It is an artifact from our past, a world where technology was more physical and mindful of people’s lives.”

“People trust typewriters and feel comfortable with them because they know their sole purpose is just to create stories on paper,” he added. “This is contrary to today’s technology, black boxes that try to propagate unethical business models based on the attention economy.”

Ghostwriter began as a vintage electronic Brother AX-325 typewriter (chosen on account of its encodable keypad matrix). Sanjeev selected the GPT-3 model in part due to his familiarity with it through his adjunct faculty position at CIID and in part to its impressive “capability to generate creative content,” he noted. “The easily accessible API convinced me to integrate this into Ghostwriter.”

Sanjeev stripped out much of the machine’s existing mechanical guts and replaced them with an Arduino controller and Raspberry Pi. The arduino reads what the human user has typed on Ghostwriter’s keyboard, then feeds that input to OpenAI’s GPT-3 API through the onboard Raspberry. The AI does its generative magic, spits out a response and Ghostwriter dutifully prints it back onto the page the person’s perusal.

“The Ghostwriter’s tactile slow-typed responses made people meditatively read each word one after the other, bringing out all the quirks and nuances of the AI through its finer details,” Sanjeev said.
“Fast digital interactions that live on a word editor tend to hide things like this unintentionally.”

Teaching the system to tap the correct keys in response proved one of the project’s greatest challenges. Sanjeev had to first decode the existing electronic keyboard’s matrix — the device that converts a key’s physical press into its corresponding digital signal. “I pressed each key, read its triggered signal-scan lines, mapped it to the corresponding key, and finally made a driver that ran on an Arduino,” he wrote. Users can even influence the AI’s answers using two physical knobs that adjust Ghostwriter’s “creativity” and “response length” parameters.

Ghostwriter will remain unique for a while longer, unfortunately, though Sanjeev is working to opensource the project so that makers around the world might build their own. “I hope to carve out some time to clean up the code and package everything together soon,” he said.

“Generative AI is definitely not a fad,” Sanjeev declared, though neither is it a silver bullet for content creation. “It is evidence that we have crossed the tipping point for AI creativity that pioneers of AI thought was impossible,” he continued.

These tools help shape our ideas and can even inspire new ones, but at the end of the day are still merely tools for our creativity, not replacements. “AI is a glorified brush that a painter can use to tell their stories,” Sanjeev said. “Humanity and life will always be the center of any successful work, regardless of whether it is realized through AI.”

GPT’s ultimate applications will depend on the medium in which it is employed — as an active, hands-on instrument for digital content creation but more as a “library of ideas for inspiration” for makers in the physical space. “The key to unlocking the potential of chatGPT in maker spaces lies in creating meaningful physical interfaces for it,” he said. “The role of an artist or creative using AI becomes that of a bonafide curator who selects the best works from the AI, filters it, and passes it to the next phase of the design process.”

He expects a similar synergy from knowledge workers as well. Automated text generation systems have been the focus of intense media and industry scrutiny in recent months amidst ChatGPT’s rocketing popularity. The technology has shown itself adept at everything from writing linux code and haiku poetry to Wharton Business School entrance exams and CNet financial explainers. Knowledge workers — lawyers, business analysts and journalists, amidst myriad others — are rightly concerned that such automated systems might be used to replace them, as BuzzFeed recently did to its newsroom.

However, Sanjeev believes that AI will instead have a less conspicuous role to play, instead trickling down from its generalist creative uses specializing into specific knowledge fields as it goes. “Just like how cloud computing has become pervasive and powers most of the applications today, AI will also become ubiquitous and recede into the backgrounds of our lives once the hype cycle fades away,” Sanjeev argued.

The AI revolution should lessen the rigors of such jobs and automate much of the drudgier aspects of the work. “The ability to synthesize vast amounts of niche data catered specifically to domains like software engineering, law, and business is being used to train hyper-specialized AIs for these respective fields,” Sanjeev noted.

OpenAI itself offers custom training packages for its systems so that customers might more easily spin up their own personalized AI doctors and robolawyers. Who ultimately bears responsibility when something goes wrong — whether it’s an AI doctor pushing quack diagnoses or an AI lawyer getting itself disbarred — remains a significant question with few easy answers.