Will Artificial Intelligence Replace Writers?

Automation and artificial intelligence are changing our world at an alarming pace. In Phoenix, AZ , Waymo One, a company founded from Google’s self-driving car project, allows people to hail driverless taxis. In medicine, AI algorithms can comb through mountains of data and identify patterns in medical outcomes that would take years to discover by conventional methods, or assess medical images and detect the presence of disease better than humans with decades of experience. These technical advances are amazing, and quite frankly a little bit scary. With machines outperforming humans on so many fronts, we as humans are left facing the question of our own obsolescence.

One hope remains. The arts. Sure you can get a machine to do math because math is ultimately a pattern of rules. But art by its definition is an expression of the human experience. Isn’t it?

AI and Art

Here’s something that freaks me out. I just went to https://hotpot.ai/art-maker and from a relatively simple string of input text generated this picture of the main character from my novel First Command, Cassi.

Cassi Requin–the main character in my novel First Command, as rendered by from a simple string of descriptive text by the artificial intelligence tool at https://hotpot.ai/art-maker on the left, and the commissioned cover art from https://miblart.com/ on the right.

I mean, holy crap! That could be an actual person on the left!

I certainly couldn’t draw that, and it would take years for me to develop the skills to generate something that complex with a computer graphics platform like Blender. Generally speaking AI art like this is based on a set of algorithms called generative adversarial networks of GANs. The system learns a specific aesthetic by analyzing thousands of input images and then generating new images consistent with that aesthetic. Images are generated randomly and judged in an adversarial way in relation to the input request until an acceptable output is attained.

If you really look at these GANs, they do a great job with what they’re meant to do, but they can struggle with things like faces. The thing is, even if you don’t get it right the first time follow-up routines can take something that looks face-ish and rework it to appear a lot more realistic.

I thought art was one of the last impenetrable bastions of the human race against our AI overlords.

The Turing Test

The reason the AI struggles with a face is that it doesn’t know it’s generating a face. It’s generating a pattern, based on patterns it has identified. These patterns are hugely complex, but they don’t account in any human experience of interaction with a face… tears from the first time you’ve broken up with a significant other are different from the tears when your dog died… your mother’s smile from when she tucked you in as a child is different from the smiles on the faces of your flag football team mates when you’ve one won the 2007 University Division III championship.

In 1950 Alan Turing proposed a test he referred to as the imitation game. It has since become more popularly known as the Turing test. The basic idea is this: an interrogator can interact only in a textual basis, asking questions of and receiving answers from two candidates A and B. The interrogator cannot see or otherwise inspect A and B. One of the candidates is a machine and the other is a real person. It is possible for the machine to do well at the game and make itself imperceivable from the human?

For early programs, sure, it was pretty easy to spot when a computer was just applying simple logic rules and drawing from random tables when necessary to generate responses. But as time goes on, the amount of data that computers can process in an effective and timely manner has grown exponentially. From the interrogator’s point of view, it’s not like there’s a single question that can be asked either. Are you a human? Either one can lie.

How would you interrogate the participants?

What if you fell in love with a participant and felt that love reciprocated? Is that real love, or just a machine emulating love? How would you know the difference?

AI and Novels

I’ve mentioned before in my blog that in his book Sapiens, author and historian Yuval Noah Harari has argued that one of the defining characteristics of the human race is our ability to tell stories. In fact this trait may just be responsible for the tremendous success of our species. Through stories we can comprehend the value of money or the fear wrath of a god. Over history stories have enabled collective, cooperative behavior on massive scales.

Is it possible for an artificial intelligence to write something as complex as a novel?

Certainly. Many popular tools available right now are based on an algorithm called the generative pre-trained transformer 3, or GPT-3. This is a deep learning algorithm that’s used as a natural language processor. Basically, given a block of text, it makes a prediction at what comes next and it often does well enough so as to be indistinguishable from the prose of a human. But according to this article, it’s not without its flaws. Sure it can string words together that make a certain logical sense, but in one of the examples it talks about the wind of a passing train knocking a character off her feet… not exactly logical for anyone who has ever been passed by a train… but of course we humans are not immune from plausibility holes in our fiction either.

I’m not sure whether it has the capacity to string together a coherent and engaging plot for the entire length of a novel, but I don’t doubt that is possible.

In the foreseeable future, what is likely to happen is that AI won’t act so much as a replacement for writers, but increasingly it will serve as an augmentation. Writers will play an editorial role, generating ideas where needed and then smoothing over AI-generated text with the lens of experience.

On a recent forum I frequent, the question came up, if AIs can produce decent books (and probably by the thousands) why would people pay humans to do the work as a much slower pace?

To this, I have several answers. I suppose time will tell if they hold up.

Craftsmanship

Robots can manufacture things like furniture, clothing, jewelry, etc. much faster and with higher precision than humans, and they have been able to for years. But there is still a market for hand-crafted items. In fact, in many cases customers are willing to pay extra for it. Often that’s because hand-crafted items are of higher quality because he goal of the work was to produce a quality item, and not to optimize profits off of a Minimum Viable Product.

And as noted above, there are aspects of the process that will still require supervision and navigation through the complexity of story production. For example, we’ve had calculators and spreadsheets for years, yet today we still have accountants.

Originality

Artificial intelligence, is largely based on emulation. It looks at what exists, and with a little poke, replicated the patterns within its experience. So if/when we get the point where AI is producing stories, my guess is that, many of these will be iterations on existing themes. This could be a boon, in for example sub-genres of romance novels where the audience demands rather formulaic stories. The theme repeats, but there are unique elements to each story than could be generated either through human input or random generation.

But the truly original ideas–those based on emerging trends, predictions about the future, challenges to social and cultural norms, metaphors for the deep problems encountered within the human experience–will be quite challenging to generate in any reliable manner.

Economics

This isn’t so much of an argument as to why AI books won’t happen, but more of a why they shouldn’t happen. If you consider a series of “bots” churning out stories, they would presumably be controlled by some kind of corporation much bigger than a typical independent author. The money taken in from these machines can essentially fall into a black hole, increasing the wealth of the corporation that owns them, but when you support an independent author, that money tends to go back into that author’s community.

Why We Read (And Write)

People read for all sorts of reasons… to relax, to escape reality, to learn… but I think one big reason is that there is a social element to it. We read to connect with other humans, and to enhance our own experience. Through books we experience what it was like to be an explorer with Shackelton or to walk on the moon, or to fall in love in Regency England. We contemplate what it might be like to be left on Mars, or what our lives might like if we could travel across the galaxy in the blink of an eye.

I would also add the writing is also a part of this connection. People don’t always write to share their work, but when they do, particularly today, it’s not really for the purpose of making money. People write out of a need for creative expression and that need won’t disappear even if there are machines that can do it better than us. People write to share a part of themselves with others.

I’ll close out this post with one of my favorite quotes from Carl Sagan.

“What an astonishing thing a book is. It’s a flat object made from a tree with flexible parts on which are imprinted lots of funny dark squiggles. But one glance at it and you’re inside the mind of another person, maybe somebody dead for thousands of years. Across the millennia, an author is speaking clearly and silently inside your head, directly to you. Writing is perhaps the greatest of human inventions, binding together people who never knew each other, citizens of distant epochs. Books break the shackles of time. A book is proof that humans are capable of working magic.”

Carl Sagan – Cosmos, Part 11 The Persistence of Memory (1980)