A.I. Text Generators: Uses and Limits
Innovation is exciting, but also intimidating. As technology advances, the world gets smaller, and automation can mean jobs on the line. The creative fields of writing and editing were once thought to be safe from the rising tide of AI. That was until now.
AI text generators like ChatGPT or Google’s Bard are powerful tools which can answer questions, write essays, and summarize text. Under the hood is a predictive engine that generates text by determining which word is most likely to come next. These programs do not understand what they’re writing. But by drawing on a large and diverse pool of training material, these generators can produce coherent (if not always accurate) prose about a huge range of topics, with very minimal input from the users themselves.
With the continued advancement of AI, writers will have to come to grips with the fact that these text generators are here to stay. It is essential that those looking to use this technology understand what AI text generators are, how they work, their capabilities and limitations, and best use cases.
The genie is out of the bottle, and AI chat generators like ChatGPT are here to stay. But it’s up to us what we choose to do with them.
How Does It All Work?
Nowadays, much of AI involves machine learning, and AI text generators are no exception. Machine learning involves a computer gobbling up a smorgasbord of data and digesting it with special algorithms. Then, the computer gives predictions as output. Advancements in machine learning involve specifying the sort of algorithms at work when the computer analyzes its data.
AI text generators eat text and predict words. We’re talking about a lot of text: more than a single human could ever read. The texts in the dataset are fed into a “large language model”, and advanced natural language processing (NLP) techniques are used to analyze it and detect patterns in usage.
From there the LLM is able to generate new text by predicting the following words or phrases in a sequence, based on the patterns it has learned from the training data. With continued prompting, the model will generate new text, allowing it to produce coherent passages that are similar in style and content to the training data. Before making an LLM based tool publicly available, companies usually build in safeguards to prevent the generation of inappropriate or hateful text.
“ These models are trained on large datasets, which contain hundreds of millions to billions of words. LLMs, as they are known, rely on complex algorithms including transformer architectures that shift through large datasets and recognize patterns at the word level. ”
To an extent, the content produced with text generators is able to adhere to a consistent style or tone, which makes it useful as an aid to crafting consistent content. Text produced is also free of glaring spelling and grammatical errors, a clear plus.
AI text generators are able to generate text rapidly and on an impressive scale. The potential use, and misuse, is clear. On the one hand, their speed makes them an attractive tool to expedite the generation of content for websites, articles, product descriptions, etc. On the other hand these generators, as they currently exist, have shortcomings which make them unsuitable to fully automate the writing process.
Technology is neither the herald of a bright future or simply a sinister replacement of human labor. It’s a tool, and what matters is how you use it. Text generators are best used as an aid, not as a crutch.
What Can A.I. Text Generators Do?
ChatGPT has been grabbing most of the headlines, and will serve well as an example. The AI’s capabilities are undeniably impressive. ChatGPT can write essays, summarize text, generate code, tell a joke, and more. However, ChatGPT has some notable limitations that the user should keep in mind.
Firstly, ChatGPTs responses lack originality. The text generators are trained on vast amounts of existing text that they use to generate new text by combining and rephrasing the pre-existing text. In other words, if it sounds like it’s simply parroting an inherited opinion or factoid, that’s precisely what’s happening. Accordingly, if you are looking to create creative content which displays thought leadership, then an AI text generator isn’t what you’re looking for.
Secondly, there’s the matter of style. True, ChatGPT is able to mimic different writing styles. For example, you can tell it to simplify a passage, or to instead make it unnecessarily complex. However, the AI is limited when it comes to a more fine-grained approach to style. When it comes to subtle sorts of framing, creative turns of phrase, or word choice tailored to a particular audience, AI still struggles. If you are looking to create something with a unique voice, you are better off not relying on these programs.
Writing with context and intent are further limitations that these programs have. The text produced is merely an educated guess at what is most likely to come next in a sentence based upon previous data. This isn’t how writing typically works. Usually one writes with what has come previously as well as what one wishes to convey in the future. As a result, the generated text may be incorrect or inappropriate as it relates to the piece of writing as a whole.
Finally, you cannot rely on ChatGPT for factual accuracy. Its training data currently only goes up to 2021, and even for matters touched on in its training set, it can generate wrong answers. It’s difficult to predict what sort of question will result in an error, making it hard to rely on.
When the inconsistency was pointed out, ChatGPT apologized, and settled on April 6th, when in fact the second answer, the 20th, was the correct one.
AI text generators pull from different sources in order to create text. The issue with this is that it may be pulling from contradictory or less than reputable text sources. There is no fact checking mechanism for AI generated text, so there is no way to verify whether the information is accurate other than checking the fact yourself.
How Can This AI Be Used?
AI text generators are a writing aid, not writers themselves. They cannot be relied upon to write whole pieces, as whatever they produce generally requires heavy editing to ensure accuracy and suitable style. What’s called for here isn’t pessimism, but realism. Limitations aside, AI text generators are landmark tools, which can be helpful to those in the writing industry and can be used for a multitude of ways.
One of the handiest use cases for ChatGPT involves modifying text. It can summarize, simplify, lengthen, or shorten a passage which you provide it. For example, here’s what happens when I ask ChatGPT to simplify a passage on quantum mechanics from the Stanford Encyclopedia of Philosophy:
Brainstorming becomes easier with AI as you have a personal assistant at your disposal, able to organize thoughts, generate lists, and suggest topics for you. Then, you can take the brainstorming material and use these programs to create a structure for the piece.
While AI text generators can’t be relied upon for factual accuracy, they can nevertheless be helpful in the outline stage by suggesting topics for further research. This is especially helpful when it comes to finding examples. For instance, if you tell ChatGPT “I am writing an article on great Polish composers, but I can only think of Frederick Chopin. Can you list some others?” it immediately outputs a list of eight composers. ChatGPT is fallible, so you need to double-check that these names fit the bill. However, that’s still a pretty helpful head start.
Finally, if you’re simply stuck, you can ask an AI text generator to provide an outline or an initial paragraph. If you’ve got a case of writer’s block, it might be easier to improve a so-so suggestion than to spin a new one from whole cloth.
Even Machines Need Editors
These programs are currently quite advanced, and will only grow in sophistication as time goes on. For the foreseeable future, however, there will remain a pressing need for human oversight. AI text generators can do a lot, but can’t write as fully human; nor can they produce artificial general intelligence, the ability of an intelligent agent to understand or learn any intellectual task which a human can. Since humans are the target audience for what we write, humans are still best at evaluating how a piece will be received.
Humans are able to provide quality control, subject matter expertise, technical ability, and a greater contextual understanding to the material being discussed. There will be more work done behind the scenes in order to ensure quality content is produced at an increasingly rapid pace as the technology improves.
Nonetheless, AI text generators offer potential opportunities for businesses and individual writers to simplify workflow, freeing up time to work on more projects than they would have been able to before. Human-led AI will most certainly play a significant role in the future of content.
Parting Thoughts
Ultimately, whether AI writing is worth it depends on one’s goals. AI writing can be a useful tool for generating content quickly and efficiently, but won’t produce writing that is as high-quality or original as writing produced by a competent human. It’s great as a copywriter and thought generator, but as an editor and quality control provider there is more work that will need to be done.