Man wearing a bright orange sweater types on computer while working in office. As AI use increases, many are looking for ways to humanize its writing outputs which can sound clunky or unnatural.
Getty
The backlash against AI-generated content is growing. Many readers are tired of reading the same formulaic AI-generated text. Repetitive phrases like “it’s not X, it’s Y” have been widely criticized as being indicators of AI slop, while many content creators turn to AI humanizers to address these concerns.
With AI humanizers, users can enter AI-generated text and remove the robotic formatting that’s characteristic of many large language models (LLMs) outputs. Changing the format helps to make AI-generated text flow more naturally and produces a more readable end product than the content produced by LLMs. But while AI humanizers are used to make writing sound more human, there are still challenges when using AI, particularly around accuracy and authenticity.
An AI humanizer is a type of software that rewrites AI-generated text to sound more human. AI humanizers aren’t just AI detection tools. The products are designed to remove robotic phrasing from input text to improve readability while emulating the user’s writing style and voice.
Humanizers work by using AI and machine learning to process input text for patterns with natural language processing (NLP), breaking up repetitive patterns in speech that are characteristic of AI-generated writing. It also adjusts tone to make the text appear more natural. Examples of AI humanizers include Humanize AI Pro and Undetectable AI.
More and more people are using AI humanizers because they want to be able to create content rapidly with the help of generative AI, but without producing an output that reads like AI-generated text. Undetectable AI, one of the biggest AI humanizers on the market, has over 20 million users.
AI humanizers first began to appear after the launch of ChatGPT in 2022, as the adoption of generative AI tools increased. Over time, these tools have increased in capabilities as advances in natural language processing have been made.
Consumer-grade AI tools like ChatGPT and Gemini are trained on massive text data sets, which help them to generate verbose outputs. However, this also leads to them producing highly-formulaic outputs. Humanizers are optimized to edit and paraphrase the outputs of these large language models (LLMs) to make them flow more naturally and avoid AI detection. These tools are used by a range of groups, including content creators, marketers and even students.
While AI humanizers can help disguise AI use, it’s not foolproof. On the one hand, these tools can make AI-generated content harder to spot, but they can also be detected by plagiarism detection tools like Turnitin by processing their formatting.
AI humanizers are limited in the sense that they can only imitate the human voice to a point, and they also provide no assurances against detection by AI detection tools. They can also include unnatural phrasing of their own, for instance, changing a word like “use” to “utilize,” or overuse transition words like “furthermore.”
For this reason, if you want to humanize your writing, you are likely better off writing work from scratch or producing a draft with an LLM and then rewriting it in your own words.
The biggest pros offered by AI humanizers are that they provide quick workflows with output text that’s more readable than the content produced by an LLM. These benefits have led many writers and marketers to add such tools to their workflows to create content faster.
However, AI humanizers come with some significant drawbacks. Most notably, they fall short of offering an authentic human voice, which is a problem given that many readers object to AI-generated content. For example, researchers from the University of Florida released a study that found people automatically downgrade stories they believe were written by generative AI.
This suggests that any potential tells of AI writing, such as repetitive formatting or even the em dash, could result in pushback from the target audience. That being said, readers can respond more positively to AI-generated content if they think it was human-authored.
Bynder, surveyed 2,000 UK and U.S. consumers and found that 56% chose an AI-generated article over one written by a professional copywriter, though 52% felt less engaged with it once they found out it was AI-generated. Humanizers can help make AI use harder to spot, but it can also lead writing to become overly polished and unnatural.
Humanizers are built on LLMs which they use to rewrite existing text. LLMs are trained on vast datasets of text which are then used to predict the next word in a sequence to generate outputs that are contextually relevant to the user’s prompt.
However, although LLMs are verbose and adept at formatting human language, they ultimately have no understanding of what they are saying. As Yann LeCun, executive chairman of AMI Labs and Chief AI Scientist at Meta said in a post on LinkedIn, “LLMs mostly memorize knowledge and retrieve answers which is one reason they need to be so big, with so many parameters. They don't understand reality at a deep level.”
This lack of understanding in LLMs, means that you can’t count on AI to produce accurate outputs. For instance, a humanizer might incorrectly paraphrase a reference to factual information because it doesn’t think like a human being does. This means it could share errors. Hence, it’s important to fact check AI for accuracy.
If you want to humanize your work, the best way is to write it yourself. The University of Oxford’s Centre for Teaching and Learning recommends setting a writing improvement goal, writing (either writing a paragraph a day or for 30 minutes per day) and keeping a diary of your progress.
It’s worth noting that while an AI humanizer might be more efficient than writing manually, the use of AI-generated content can be problematic, particularly in industries like journalism and marketing, as LLMs can introduce hallucinations and bias. If you’re creating content in those industries, consider disclosing the use of AI to the reader.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。