It’s the number one question we get asked at AI21 Labs:
“Is AI going to replace writers in the near future? ”
It’s obvious why we are asked that question. After all, our company is developing the next generation of language models for AI writing..
The topic of AI writing has been heavily debated across the world, with readers often becoming more confused by the opinions presented by various media outlets.
Take this article published on Forbes titled: “AI Is Not Going To Replace Writers Anytime Soon—But The Future Might Be Closer Than You Think.” The title alone is enough to prompt a rollercoaster of emotions.
“AI Is Not Going To Replace Writers Anytime Soon”
Writer: Oh good, that means my job is safe!
“But The Future Might Be Closer Than You Think”
Writer: Wait, what? But you just said… How much closer is this future you are speaking of?
Many writers are, with good reason, worried about their future career prospects as the technology behind these AI writing tools become more advanced. But fear not! As insiders to the world of AI writing, we can confidently say that the age-old profession of writing will be with us for longer than the media would lead you to believe. Let us explain!
Why it’s easy to get confused
At AI21 Labs, we do not believe that AI will replace writers—at least not anytime soon! In fact, we are building our entire business model around the premise that AI tools will serve as writing assistants rather than replacements for writers. This affects many aspects of our own product, Wordtune, so you could say we have a lot of skin in this game.
However, AI will likely change the writing profession in a number of ways. Some roles will be made redundant, some will need to be adapted, and many new career opportunities will open up for writers who embrace the change. Once you understand the current capabilities of AI and the technology behind it, it’s much easier to envision where these tools are really headed.
Before we dive into the details of why AI won’t replace writers, it’s important to grasp exactly how computers have the ability to understand written text and the spoken word. For that, we’ll need to briefly explore NLP (Natural Language Processing), the branch of AI that makes this possible.
The age of expert systems
AI is not a new idea. In fact, it has been around for decades. Over the years, however, the meaning of the term “AI” has changed, and its popularity has fluctuated dramatically.
In the 80s, AI generated a lot of hype. It was just as popular as it is today, easily filling standing-room lecture halls to the brim.
But the term “AI” was used to mean something quite different back then. In those days, AI machines were expert systems that tried to mimic the human thought process. These systems were handcrafted by surveying experts and were creating “if-then” rule sets, a sort of flow chart that would form the computer’s decision-making process.
Researchers interviewed experts and tried to extrapolate decision trees of some kind, and although machine learning existed at this time, it didn’t play a significant role in this process.
The hype of this period was followed by what was often known as “the AI winter”, after these systems failed to produce practical models or results.
Transition to machine learning
From 2010 onward, an entirely new kind of AI began to emerge—one based on Big Data, statistics, deep learning, and neural networks.
Deep learning is a type of machine learning that allows machines to learn from data. In image classification tasks, for example, machines are able to analyze large quantities of images and extract hidden features or information from them. If you show the computer millions of images of cats, it will be able to “learn” the common hidden features of these images, so it can recognize whether the next picture is indeed an image of a cat.
In terms of AI, the last decade has been all about vision, and object recognition was radically transformed during this time. In the current decade, the same advancement is happening for language.
AI can be used for language through what is known as a “language model”. It’s often easier to understand this if it is seen as an information model. The computer’s neural networks have pretty much the entire span of all of the internet’s texts within their grasp to “learn” from.
I’ve put “learn” in brackets because what these networks do is closer to prediction than learning. When presented with a certain piece of text, these machines can relatively accurately predict the next word.
These neural networks boast an unfathomable number of parameters—over 170 billion with models like GPT3 by OpenAI or Jurassic X by our own AI21 Labs.
The scale of this technology seems limitless, but it does have its restrictions. Since these networks are based on predictions and statistical calculations, they lack the basic building blocks of understanding that we, as humans, take for granted.
Even the simplest sentence a five year old might say:
“I cried because Tommy hit me”, Is completely incomprehensible by these networks, because they lack the basic rules about reality that we take for granted.
The glass ceiling of AI writing
AI is not at a point where it can compete with human intelligence, since it is based on statistics and probabilistic models. So why is there so much confusion regarding the current capabilities of AI in the field of writing?
This could have to do with the fact that many AI writing companies are simply over-promising in terms of what their tools can do. There are many AI companies that promise complete article generation from a short prompt. You tell the machine you need a “top SEO tools” article, press “GO”, and then get a complete article in no time.
If that’s possible, then why not stop using writers altogether and switch to this cheaper, faster machine-led solution?
1. These are not conscious machines.
Transformer neural networks are statistical machines, as we mentioned earlier. Human input is still required to prompt the machine. Contrary to the recent Google hype that says otherwise, these aren’t conscious machines. They are statistical systems designed to predict the next word. What it produces is impressive, but it can’t think for itself. Making something meaningful still requires the human element. However, there’s no doubt that both ideation and creativity can be enhanced with these tools.
2. Writing is the art of turning ideas into words. We still need great ideas.
Even if a computer is developed with the writing level of a modern Shakespeare, we still have to direct it toward what we want it to write about. In other words, there is still an element of human-machine interaction. You need to direct the model to express what you intend to say. Essentially, you need to supervise the AI.
A calculator is essential for solving algebraic equations. But without you, the human, how does it know which equations to solve? AI for writing is very similar. Imagine you tell the machine to write about the hat you are wearing. A very specific topic, but still there are endless directions to write about. Should it write about its manufacturing? Color? Fabrics? This is the reason human direction is so integral to the process of AI writing.
3. We will always prefer to read what real people have written.
As humans, we have an intrinsic desire to follow, read and consume content from other humans—not from machines. If machines can write like humans, human readers will find ways to authenticate human writers, because who wants to read an opinion piece by a bot?
Our preference to read from real people might make the writer an irreplaceable part of the process, regardless of technological barriers.
AI’s role in the future of writing
AI won’t replace writers, but we predict that writers who use AI will replace writers who don’t. When we cease worrying about machines replacing writers, we can take a closer look at how AI can improve writing.
AI can help us express our ideas more concisely and accurately. We have many dimensions to our thoughts, and they are extremely nuanced. We have a hard time translating these thoughts into a limited set of letters, phrases and sentences.
Once these systems can process language and suggest alternative ways to express your thoughts, there’s a good chance you’ll find a better way to say what you really mean. You will be more adequately equipped to translate your thoughts into words.
Our thoughts are so abundant, and our writing is just a reductive extraction from this abundance. Now that we have this powerful technology at our fingertips, it can help us formulate a more precise reflection of our thoughts.
Let’s take our own product, Wordtune, as an example. With Wordtune, the user can input an initial phrase and the tool will immediately offer several options for them to examine and choose from. Lots of people are already using AI just like this to help them write all kinds of content, including books!
Beside helping with better expression of thoughts, machines can also help us generate ideas. Writing is not a one-directional “idea-to-final copy” process. It’s an iterative process where new ideas come as you write and as you try to express yourself. Machines can play an instrumental role in our writing process by helping us expand our thoughts.
These writers who willingly embrace and harness the power of technology have opened themselves to the opportunity to focus on their creativity, ideas, and thoughts, as the technical aspects are dealt with by the AI.
Brushes for different styles of writing
Visual editing software like Photoshop and Figma have changed how designers work, giving them the ability to use an endless variety of brushes. We can imagine how having similar “brushes” for text could help writers articulate themselves in more versatile ways. Perhaps a piece of content originally created for an internal business report needs to be adapted to a more fun, casual and engaging style for the company’s website or social media. Writers could apply their signature style to that existing content and watch it update before their eyes.
Democratizing content creation
This technology also has the ability to democratize content creation by bringing more people into the world of professional writing. This includes people with limited writing talent, as well as those with disabilities like ADD and arthritis. This leveling of the playing field not only empowers writers but also adds more people to the workforce.
Helping non-native English speakers
It is estimated that there are 1.2 billion English language learners in the world. In a recent study, the impact of AI-based writing assistants on English language learners was measured. The study found that non-native speakers tasked with writing in English benefited significantly from the help of a specially designed, structured writing tool (AI KAKU), which was developed by the researchers, in comparison to a standard word processor.
By bridging the language gap, AI writing tools can empower millions of non-natives to participate much more freely and confidently in the business world.
The dark side of AI writing
I don’t want it to seem as if I’m sugarcoating AI technology. We cannot talk about the future of AI writing tools without also mentioning the dangers of this technology.
Using this technology, some unscrupulous companies can engage in various ill-doings, such as forging negative reviews of their competitors' products or filling the web with AI-generated spam. This is not an invention of AI, but it can be done at a greater scale than before because of it.
As these darker utilizations of the technology develop, we will need to produce counter tools to detect and block the effectiveness of black-hat writing tools.
Lower-level writers are at risk
While AI won’t replace writing as a profession, it will replace certain types of writing jobs. The writers who adapt to this will continue to thrive and find new opportunities, and those that don’t, won’t.
Throughout history, we’ve seen many jobs be replaced by technology, and we will see many more go through the same process in the future. Self-driving cars are a big threat to professional drivers, for example, and self-service checkouts and online shopping have reduced the number of shop assistants required by retail businesses.
Low-skilled writers will be the first to be replaced by AI writing tools. If you are such a writer, you should start thinking about how to differentiate your services and upscale your expertise.
I agree with Ryan Law’s tips for succeeding as a writer in the age of AI:
- Write thought leadership content that is based on interviews, original research, personal experience or data analysis.
- Shift from content generation to more strategic tasks, like curation, optimization, fact-checking and directional input.
Writers—the next generation
AI writing technology is here, and it’s here to stay. Once we get over the initial urge to cling to our previous working habits, we can start to open up to the wonderful possibilities these tools offer.
For writers, these tools can open opportunities to elevate themselves from mere transcribers of words to thought strategists and leaders.
As this technology develops, writers will be able to spend less time on the writing side of the process and more time on deep thinking, honing their creativity and crafting unique perspectives. Writers with innovative ideas will be the ones who will mostly benefit from this inevitable shift in the industry. And the world will benefit from these ideas being expressed with clarity, concision and impact.