Artificial intelligence (AI) is the latest in a long series of technological upheavals that have changed society forever. Artificial intelligence for writing has recently become more visible through OpenAI’s introduction of ChatGPT. It is a free software service. (GPT, by the way, stands for Generative Pre-trained Transformer.)

Anyone can use it by entering a prompt. ChatGPT then produces a piece of writing while you watch. Supposedly, it can write on any subject. It can write prose or poetry. And it can do it much faster than any human can. Not only that, it can translate from one language to another.

It has naturally raised fears that artificial intelligence can replace human writers. That seems unlikely, but it also seems unlikely that any writer can avoid it for very long. It will bring new opportunities, but it will also cause problems.

A historical overview of writing and publishing technologies

For centuries, writing meant using a pen or something to mark something such as papyrus or paper. Publication took place though scriptoriums, where trained scribes made copies for distribution. Then Gutenberg invented the printing press. Publication became much faster, more accessible, and less expensive.

Writers still had to produce their work with pen and ink until the invention of the typewriter. The introduction of the personal computer paved the way for word processing software to make typewriters obsolete. Soon enough, word processors introduced spell checkers and grammar suggestions.

The World Wide Web allowed everyone to be a publisher and post online. And here is where ethical problems with writing technology began to appear. Many writers wanted to have their work posted on multiple websites, but search engines discouraged duplication of content. So someone invented article spinning software.

It became possible to place synonyms within curly brackets. The software could then {quickly|rapidly|instantly} produce any number of {similar|comparable|related} articles by choosing different combinations. Some of them were really badly done. The internet became flooded with articles that looked different only to the search engine’s bots.

Artificial intelligence for writing continues a parade of new technologies that began centuries ago. It builds both on their advantages and drawbacks. It will soon enough become impossible to avoid using it.

My first experiences using ChatGPT

I have asked ChatGPT to generate several responses, using very short prompts. If I wanted to use it as a serious tool, I would have to devise much longer and more detailed prompts. These documents, however, give me some useful experience. In what follows, I describe responses in a logical order, not the order in which I prompted them. The links will take you to ChatGPT’s output.

I have done quite a bit of research on what artificial intelligence can do for writers by searching and reading web content, but why not ask ChatGPT about ChatGPT?

I specifically asked about its ability to cite sources. It says it can but warns that its abilities are limited by material in the knowledge base used to train it. So I followed up with a question about the knowledge base.

So I asked it to describe a garden plant, the rose mallow and then cite the sources it used for its response. It denied having access to sources. In a more general prompt about flowers, I did learn that ChatGPT can write in other languages besides English, other writing systems besides the Roman alphabet, and can translated its responses.

Artificial intelligence for writing poetry

My reading turned up the claim that ChatGPT can write poetry, so I asked it to write about Elon Musk in blank verse. It failed. Rhymed quatrains are not blank verse. So I asked it to write about Musk in the style of Shakespeare. At least it responded with iambic pentameter.

Actually, my reading said that it can write limericks and haikus, so I prompted several different ones. Many of the limericks don’t scan properly. The rhymes don’t all work. The haikus all exhibit the proper structure of five, seven, and five-syllable lines, but they lack the vivid descriptiveness and emotional overtones of even mediocre haikus written by humans.

Controversial topics

Turning to controversial topics, I asked for brief essays on border security in the styles of the New York Times and New York Post. Those two newspapers have sharp ideological differences, but ChatGPT’s responses are both so neutral that I can’t detect any sign of different viewpoints.

Then, I asked an even more controversial question:  is a male who identifies as a female really female? This prompt elicited a response from a purely progressive perspective. It barely admitted that any other viewpoint exists and certainly made no attempt to describe one. Considerations of brain chemistry or genetics are completely absent.

I have not yet begun to experiment with longer and more detailed prompts. A request for a longer article about transgenderism specifically asking for consideration of psychological and biological aspects would probably have gotten a more useful response.

Overall impressions

That said, I am not much impressed with the ChatGPT output I have seen, either in my own experiments or the ones I have seen in my research. In particular, its inability to search for information outside its training knowledge base means it’s not ready for prime time.

A version of ChatGPT that can do real research and cite sources will come along sooner rather than later. I doubt, however, if any such improvements will make much difference in consideration of the promises of artificial intelligence for writing or the problems it poses for society.

For whatever it’s worth, my SEO plugin rates the readability of most of the English prose only as OK. It flags two readability problems: too much passive voice and too many sentences of more than 20 words.

Promises of artificial intelligence for writing

My post on artificial intelligence in libraries surveys some of the general promises and problems with artificial intelligence.

For writers, artificial intelligence can speed up both finding sources and producing drafts. In less than an hour, it can produce content that would take several hours for a human to do. Not only that, it can accept human feedback and learn to become even more effective.

Plenty of people have a poor grasp of spelling and grammar. Plenty of others lack confidence in their ability to write. Artificial intelligence will produce content with no spelling or grammatical errors. Therefore, it can boost writers’ competence and confidence for producing a decent first draft.

For people with writer’s block, AI tools can suggest ideas for topics or pose questions.

Human editors ought to check the content, adding their own expertise and professional judgment to the final draft. After all, AI tools know grammar, prose structure, and other mechanics, but not the logical progression from one though to another.

Copywriters will be among the first to benefit from AI. They need to produce a steady stream of emails, articles, white papers, advertising copy, and more. Learning to use artificial intelligence for writing can only make them better organized and more productive.

For more scholarly projects, some artificial intelligence tools can summarize articles. Scholars can then use those summaries to determine which articles are worth reading in depth. Other tools can compare a draft of an article with already-published literature to avoid the risk of accidental plagiarism.

Social problems from artificial intelligence for writing

ChatGPT can quickly generate natural human language on a stunning variety of topics. Unfortunately, we can’t depend on the answers ChatGPT generates to be correct, as OpenAI freely acknowledges.. This problem can probably be fixed. Surely OpenAI doesn’t intend to keep the tool tied to its training materials indefinitely.

Other problems can neither be fixed nor avoided.

Decline in human writing skills

Teachers are already devising ways to use AI as a tool and discourage cheating with it. Nevertheless, we as a society have become addicted to convenience. People will be tempted to rely on AI as a crutch instead of a tool.

It’s nothing new. Calculators appear to interfere with the ability to learn basic math skills. GPS systems appear to interfere with the ability to read maps and plan routes.

Future students will probably do far less real writing than they do now. Less experience in writing means less opportunity to develop good writing skills.

Students can already pay someone else to write term papers. Tools such as ChatGPT will enable them to cheat more efficiently and less expensively.

The ability to write and the ability to think are closely related. If people rely on a robot to do their writing, they’re relying on it to do their thinking.

Not only that, those people will read less and less, and therefore be less likely to be well informed. It’s happening even without AI. Florida Polytechnic University opened in 2014. Its administration decided that its library would have no printed books.

One freshman enthusiastically observed, “When you get a print book you spend time reading to find what you need, but with digital, you can go straight in. That this much information is available at our fingertips is something other libraries can’t provide.” How can anyone possibly understand what parts they need without reading first? But technology makes it possible to bypass reading, thinking, and learning.

Job losses

Artificial intelligence is likely to take jobs from humans. To be sure, it will likely create new ones at the same time, but these new jobs will require skills and aptitudes that the displaced workers will not have.

This problem is as old as the Industrial Revolution. Sophisticated machinery impoverished traditional craftspeople, who could not compete against it. It’s not that the machinery produced superior products. In fact, the traditional craftspeople could probably produce superior products. They just couldn’t produce products as quickly or in as much quantity as the machines, and if they tried, their quality suffered.

Back in the days of mainframes, I remember seeing a cartoon of someone showing off the company’s new computer. “This baby can replace ten secretaries.” But there were eleven people on ladders maintaining the computer. Surely each one made more than any of the secretaries.

Count on it. Artificial intelligence tools will help some people become more affluent at the expense of pushing others into poverty. Why should this revolutionary technology turn out differently from the previous ones?

Scams and misinformation

Artificial intelligence makes not only legitimate work easier, but it’s also a boon to scammers. Non-English-speaking scammers now give themselves away with their poor spelling and grammar. If they use artificial intelligence for writing, the software will take care of competence in English. Potential victims will lose an important clue.

Similarly, some people deliberately work to spread propaganda and misinformation. The speed and volume of artificial intelligence content writing can make it impossible to detect scammers and deliberate misinformation.

Environmental impact

Artificial intelligence consumes a lot of energy. Before OpenAI could introduce ChatGPT, it had to train it. Training required storing vast amounts of information, which required keeping lots of computers cool.

That is, before anyone outside of OpenAI had a chance to use it, ChatGPT had a huge carbon footprint. It’s similar to that of other digital energy hogs such as the cloud. The industry is actively seeking more energy-efficient and environmentally friendly equipment and procedures, but it hardly counts as green yet.

Uncontrolled bias

ChatGPT was trained on a wide variety of human writing, worldwide and spanning a wide span of time. But as vast as the knowledge base is, it still had to be selected by a team of humans.

How diverse is this team? I don’t mean just diversity of such factors as race, ethnicity, or gender. How many team members are past middle age? How well does the team represent all the various ideologies of the present day?

Our media (news, entertainment, social platforms, etc.) is dominated by the left side of the American political spectrum. How well is conservatism represented on the team that trained ChatGPT? For that matter, how well represented are any of the various non-western ideologies?

If any kind of group think controlled the choice of training material and other aspects of ChatGPT’s design, it will display a hard-wired bias that is even more serious than the factual errors it demonstrates. And the team will not be able to assess its own biases without soliciting outside critiques.

Unfortunately, that’s not how businesses operate. OpenAI considers a lot of its models and procedures proprietary. The future of ChatGPT therefore rests on OpenAI’s own visions, goals, and ethical standards, none of which outsiders can discover.

OpenAI’s introduction to ChatGPT promises to curb bad use of artificial intelligence:

While we’ve made efforts to make the model refuse inappropriate requests, it will sometimes respond to harmful instructions or exhibit biased behavior. We’re using the Moderation API to warn or block certain types of unsafe content, but we expect it to have some false negatives and positives for now. We’re eager to collect user feedback to aid our ongoing work to improve this system.

That’s supposed to be reassuring, but who will protect society from any hardwired bias in the Moderation API?

Sources:

6 big problems with OpenAI’s ChatGPT / Garling Wu, Make Use Of. March 21, 2023
ChatGPT, artificial intelligence, and the future of writing / Glenn Geher, Psychology Today. January 6, 2023
Introducing ChatGPT / OpenAI