For the past several years, machine learning (ML) and artificial intelligence (AI) experts have worked on algorithms that can write articles and other types of content in ways previously thought to be exclusively writable by human beings. As these technologies get more advanced, we’re gradually approaching an era where machines could write content better the humans.
But just how close are we to having machine writers that are better than humans? And what would that future actually look like?
Key Challenges for AI Writing
Let’s start with . Machine learning and AI are incredibly good at a handful of different applications, but there are critical challenges that prevent them from being utilized in a broader range of contexts.
With regard to writing, there are several distinct challenges faced by AI developers:
Natural language processing. One of the biggest hurdles to overcome with AI writing is natural language processing (NLP). We’ve seen an explosion in the capabilities of NLP in recent years, in a variety of applications, with the most obvious being the rise of digital assistants. Thanks to natural language recognition, assistants like Siri and Alexa have gotten incredibly good at learning to “understand” basic human speech, even in a conversational style. Some assistants are even as good as a human speech identifier. However, learning to recognize human commands and trying to write human language are two entirely different tasks, and it’s remarkably difficult to train an algorithm to write in a comprehensible way. It’s one of the ways traditional link building for SEO is getting obliterated by natural language processing.
Information processing. Before an AI writing algorithm can write an article on a given topic, it must somehow “understand” that topic. There’s no conscious learning or understanding going on, but the algorithm must seek, digest, analyze, and reprocess information from other sources in order to write material that is accurate. In some cases, this is trivial; algorithms can easily look up current stock prices and report on them. In others, it’s almost impossible; algorithms can’t judge the events of a grainy video.
Creative ideas. Many forms of human writing depend on creativity; writers must come up with unique ideas, or unique angles, and present them in some compelling way to an audience. The spark of human creativity that gives rise to novel ideas has yet to be replicated in any kind of machine form. It’s uncertain how a programmer could go about trying to replicate this feature of human thinking.
Voice and subjective appeal. We find books, articles, and other forms of content appealing for a variety of reasons, including the voice of the writer and the overall subjective appeal of the content. This is important for everything from novel writing to . While algorithms might be easily trained to gather and reword certain types of information, they may struggle to figure out exactly how to frame the information in a subjectively appealing way.
Opinions and experiences. Some articles depend on the presentation of an opinion—especially in contexts like political reporting or advice articles. Others are presented with a unique perspective because of human experience; the writer has undergone some personal experience that gives them information that can’t be gleaned in another way. Algorithms would have difficulty inventing new human experiences and opinions that seem realistic; they can only mimic what’s already been presented.
Where AI Excels
With all these drawbacks, where does AI excel?
AI is exceptionally good for tasks that are predictable, repeatable, and objectively measurable. In these situations, an AI algorithm can repeat a given task or solve a simple problem incredibly quickly, and at scales impossible for human beings to achieve. In the context of writing, this means that predictable, formulaic writing is trivially easy for AI algorithms, and articles of this level can be written many thousands of times faster with an algorithm than with a human writer.
Another key advantage of AI is , which could push it into new territory when it comes to writing; over time, algorithms can be trained to evaluate the quality of their output based on many different factors. Given enough time and resources, algorithms could learn to write much more complex, arguably creative works.
The Current State of AI Writing
Already, AI algorithms are working hard to produce content. Chances are, you’ve read many articles that were written by algorithms, and you don’t even realize it. For several years, major news outlets have relied on algorithms working in the background to create articles designed for simple reporting; for example, algorithms are frequently used to report updates on sports matches or stock market fluctuations.
Some algorithms have attempted to go even further, pushing the limits of what machines can do. For example, some programmers have created algorithms designed to replicate the form and writing style of famous authors and poets, creating brand-new works that are hard to distinguish from original human writings. Of course, these advanced algorithms are still in an experimental phase, and have yet to see widespread use or even consistent results.
Online Articles and Search Engine Optimization (SEO)
One of the most promising areas for development in AI-based writing is in the world of online marketing and search engine optimization (SEO). Content marketing has become one of the most popular marketing strategies in the world, due to its ability to simultaneously. While the best content often requires human creativity, opinions, and emotions—things AI can’t currently mimic—the bulk of published online content could feasibly be written by machines in the near future.
Of course, even low-level content marketing fodder is still more complicated than reporting on sports or stocks; it’s going to take time to develop programs capable of rivaling human writers in terms of comprehensibility and style.
Books and Creative Works
It’s going to be even more impressive when AI algorithms make a breakthrough in the world of creative writing, tackling short stories, novels, and poetry. Algorithms have gotten exceptionally skilled at , and some specialty algorithms have gotten good at recognizing speech patterns, including local dialects and styles. When these algorithms are sufficiently advanced, they may be able to accurately and consistently replicate the styles of famous authors, possibly even hybridizing different styles together, much in the same manner that Deepfakes can convincingly fake the presence of a specific human being in a recorded video.
Still, it may be some time before algorithms are able to generate truly unique ideas; when the first consistent creative writing algorithms emerge (and they will almost certainly emerge soon), they’ll likely focus on conjuring ideas generated from a pool of previously successful works. In other words, they’ll be better at mimicking existing good ideas than they are at coming up with new ones.
Human Emotions and Subjective Perspectives
One of the crucial obstacles standing in the way of AI writing algorithms is finding a way to gain the power of human emotions, opinions, and subjective experiences. There are some reasons to believe this is achievable; for example, are able to detect changes in human emotions based on the vocabulary and tone of an interaction. Other chatbots are developed for therapeutic purposes, and can effectively imitate certain human emotions. It may only be a matter of time before these concepts are better integrated into writing algorithms.
The Ongoing Evolution of AI
It’s hard to tell where the future of AI-based writing is headed. Currently, algorithms are quite capable of writing articles that are indistinguishable from human writing efforts, given subject matter that’s sufficiently simple and easy to understand. In the near future, algorithms may be harnessed to make advancements in online marketing, journalism, and even creative writing. However, for the foreseeable future, AI-based writing algorithms will still have shortcomings, based on their inability to replicate human emotions, their limited capacity in handling complex topics, and their lack of “natural” creativity.