I Am Cyborg Watch Me Write - Umberto Tosi

I used to think of writing as a lonely pursuit. Were it a sport it would be like golf - better yet, like fly fishing in high country. Lately, however, I've realized, not so much. My author-in-the-attic time notwithstanding, I find that I must depend on the kindness of friends and colleagues - even strangers - in the process of bringing my work to fruition, not to mention contract copy editors and designers.

I'm talking only about solo works. I'm not counting staff-produced works - magazines, journals, and so forth, or even blogs. Neither am I counting High Treason - the cold war biography I coauthored with the late, CIA-KGB double agent Vladimir Sakharov in the 1980s. Nor either of my other coauthored books that were outright collaborations.

I try my best to thank all of the above on in my acknowledgements - including those who generously provided forewords and blurbs, but the list falls short. My inamorata, Eleanor, vets all of my final drafts. Besides being a noted symbolist painter, she is a fine writer, with a keen eye for inconsistencies and typos who isn't shy about sharing her unvarnished judgement when needed. My eldest daughter, Alicia Sammons - who teaches literature as well as being a stylish writer herself - offers keen pointers and encouragement often. Others include a few of my colleagues at Chicago Quarterly Review.

If I dolly back, the bigger picture shows a network of people, past and present, the writer consults - so to speak - in composing a book or a story - what we call research. Back before the Internet, I used to have shelves of reference books at hand when I wrote. What I could not find in them, I got by phone - calling sources and colleagues. The kindly folks on public library information desks could look up and supply answers to the most arcane questions. Nowadays, I  Google everything. People, nevertheless, provide the information at the root of this - from historians, scientists, and reporters alive and dead, active and retired. Somebody, after all, writes all those Wikipedia pages that set me straight or point me in the direction of other sources. (I've written a few myself.)

The Internet puts much of the accumulated knowledge and skills of humankind only a few clicks away, if only we writers can grasp and make use of what we need. It is a huge set of electronic-power tools for we writers, good and bad, equally useful to those with lofty, benign or malicious intents. It increases the number of potential collaborators exponentially.

Now we enter a new age - that of that robot writer - artificial intelligence in the form of neural networks, that can be designed to write prose or poetry on their own. Our Authors Electric colleague, futurist and science fiction writer Edwin H Rydberg wrote eloquently on the subject in his September, 30, 2018 post, "Will Robots Write the Next Novel?"

They're gaining on us, for sure. A Japanese literary contest already shortlisted "a scarily well-written" AI-generated story in 2016. It had human help, however, according to a story about it in Fast Company. The accomplishment remained impressive, nevertheless. The AI team that designed the system assembled a conventional novel as a template. The system deconstructed it into words and phrases, and created a choice-matrix for the program to follow. It generated a new novel in a similar pattern to original, doing about 20% of the overall work, the team's report says.

More recently, OpenAI - a San Francisco-based nonprofit founded by Elon Musk and programmer Sam Altman has released an AI system that can write fiction, nonfiction and even poetry. OpenAI released it in stages this summer -- GPT-1, followed by GPT-2 - amid an initial flurry of sensational coverage about its purported ability to generate "fake news." The system's creators proceeded cautiously, with a statement announcing: "Due to our concerns about malicious applications of the technology, we are not releasing the trained model. As an experiment in responsible disclosure, we are instead releasing a much smaller model for researchers to experiment with, as well as a technical paper."

The GPT-2's full scale system gets is power by being able to draw on a vast array of Internet information to provide context to the words being generated much like a human writer does -- at electronic speed, however.  MIT's Technology Review gave it high praise, including an uncannily authentic-sounding continuation of a John F. Kennedy speech that GPT-2 handled with ease. It's easy to see how such technology could be misused - but that's the same conundrum presented by all technological advances. The problem - as Shakespeare wrote "is in our stars" - our moral evolution not on technology itself.

Will such systems replace writers altogether - or become another tool writers can use to expand and enhance their imaginations and reach? The questions echoes the themes of science fiction literature and of the broader, current arguement among futurist over AI itself - whether it will replace humans, serve humans or become integrated with humans as another step in our evolution. Will it be Terminator, or I Robot or what have you.

You can try the latest - although still not full - version of GPT-2 yourself at a
Website provided by OpenAI at Talk to TransFormer. 

It's straightforward.  You give the system a prompt by typing, or pasting, words into a window. This can be a snippet of text, a passage from an article, what have you. The system has been trained, using data drawn from the Internet as context, to “predict” the next words of the passage — meaning that it transforms your prompt into a news article, a short story, or even a poem.

I followed the link to try out the AI "TransFormer" for myself.

This being about robots, I entered the opening paragraph of my novel-in-progress: "The God Machine"... in which one of the protagonists - fittingly an android, introduces himself. Below it I've pasted in how the AI neural network continued my passage.
 -------------------------------------------------------------
[My paragraph]

I leaned on a tree trunk and breathed the pungency of morning drizzle on its bark, and rising from the soil and summer foliage. It was a most human, visceral experience for which I could thank the quivering sensibility of my inner Vin. It was my seventh sortie outside of B-Zone, but the first time I had stopped “to smell the roses,” as humans say. My power was on reserve. I needed to idle. I also needed time to consider my next move.
 --------------------------------------------------------------

[After a few moments: "TransFormer" offered up the following continuation, still in the first person:]

"I'm going to go have a smoke," I said to my crew. "I think I'll enjoy the view. And then I'll see you all on the other side."

I turned to leave.

"Hey guys," said one of the girls. "A lot of people are wondering why we stopped at our campsite, though it's more complex than that. In short, we had a lot of requests from our old comrades. They needed some sleep, they wanted to hang out, and they felt like people..."

--------------------------------------------------------------

Not bad but as you can see, off the mark.

The AI system provided contextual continuation, but missed the implication - clear to human readers, I hope - that my speaker was not human. To put it in anthromorphic terms: The AI system couldn't recognize one of its own!

It seems clear to me that in its response to my chapter-starter paragraph/prompt, that the system had failed the Winograd Schema Test. ( A Winograd schema is a sentence that’s grammatically ambiguous but not ambiguous to humans — because we have the context to interpret it. ) This is because my prompt-paragraph implies - through context - that the speaker is an android - something the "TransFormer" missed in it's response-text.

Still not bad for a prototype. It won't be long before such systems' improve.
The GPT-2, however, did cause me to re-examine my own text in reaction to its error - just as I would to a human's reaction to my prose, on or off the mark. It made me think about whether I had been too subtle in introducing my android - and his "inner-Vin" - which the story goes on to show that to be the remnants of a human character's downloaded consciousness, the human since deceased. Would that make the android human, a robot, or a cyborg, or something else?

Stay tuned.

----------------------------------------

Due to having artificial intraocular lens implants, Umberto Tosi is technically a cyborg. He is also the author of Sometimes Ridiculous, Ophelia Rising, Milagro on 34th Street and Our Own Kind. His short stories have been published in Catamaran Literary Reader and Chicago Quarterly Review where he is a contributing editor. He was contributing writer to Forbes, covering the Silicon Valley 1995-2004. Prior to that, he was an editor and staff writer for the Los Angeles Times and its Sunday magazine, West. He was also the editor of San Francisco Magazine. He has written more than 300 articles for newspapers and magazines, online and in print. He joined Authors Electric in May 2015 and has contributed to several of its anthologies, including Another Flash in the Pen and One More Flash in the Pen. He has four adult children - Alicia Sammons, Kara Towe, Cristina Sheppard and Zoë Tosi - nine grandchildren, three great grandchildren. He resides in Chicago. (He can be contacted at Umberto3000@gmail.com